The research, conducted by Jaya Rajkumar Ramchandani and Suddhasheel Ghosh of Jawaharlal Nehru Engineering College at MGM University in Chhatrapati Sambhaji Nagar, India, formulates flyover design as a three-objective minimization problem. The objectives are flyover length, construction cost expressed in Indian rupees at 2023 base-year prices, and construction-related carbon emissions measured in tonnes of carbon dioxide equivalent. Rather than treating environmental impact as an afterthought to be assessed once a design is complete, the framework builds emissions directly into the optimization loop, drawing emission factors from the IPCC 2019 greenhouse gas inventories, Indian Standards for steel, and Maharashtra state road development project reports.

The case study is grounded in a real infrastructure context: the announced 25-kilometre Shendra–Waluj corridor in Maharashtra, which includes an 8.5-kilometre double-decker metro section. Megaprojects of this scale involve substantial public investment and long-term environmental consequences, which makes the choice of optimization algorithm a strategic decision rather than a technical footnote. The researchers modeled a prestressed concrete box-girder flyover with piers and abutments, loaded according to Indian Roads Congress Class A and 70R live load classes in seismic Zone III, with traffic assumptions of 45,000 vehicles per day growing at six percent annually.

At the heart of the framework are 29 design variables, ranging from clear span lengths of 20 to 45 metres and roadway widths of 7.5 to 15 metres, to deck slab thicknesses, girder depths, bearing types, skew angles, and load distribution factors. The bounds for each variable were established from Indian Road Congress codes, AASHTO specifications, and analysis of existing flyovers in Aurangabad. A variable-importance ranking revealed that clear span length, roadway width, pier height, and slab thickness exert the greatest influence on cost and environmental outcomes, while reinforcement diameter and foundation depth play more moderate roles.

Evaluating thousands of candidate designs with full finite element analysis would be computationally prohibitive, so the team built Radial Basis Function surrogate models for each objective. These models interpolate exactly at training points and approximate smoothly between them, allowing the optimizers to explore the design space at a fraction of the cost. Validation using 500 designs generated by Latin Hypercube Sampling, with ten-fold cross-validation, yielded mean absolute errors of 2.1 percent for length, 3.4 percent for cost, and 4.0 percent for carbon emissions, with coefficients of determination above 0.95. The surrogates achieved efficiency ratios of 95 to 98 percent relative to full finite element evaluation, meaning the entire optimization could run on a standard engineering office computer in under an hour.

Nine algorithms were compared under identical conditions: NSGA-II, NSGA-III, a genetic algorithm, differential evolution, CMA-ES, a classical evolution strategy, a stochastic ranking evolution strategy, Nelder-Mead, and pattern search. Each ran ten times with different random seeds, a population of 100 where applicable, up to 500 generations, and a budget of 50,000 function evaluations. Performance was judged on the hypervolume indicator, which measures the volume of objective space dominated by a Pareto front; the inverted generational distance, which measures closeness to a reference front; spacing, which measures how evenly solutions are distributed; computational time; and the fraction of solutions satisfying IRC and AASHTO constraints.

The results were decisive for the evolutionary multi-objective algorithms. NSGA-III achieved the highest hypervolume at 0.847, narrowly ahead of NSGA-II at 0.839, and both substantially outperformed every single-objective method adapted through weighted-sum scalarization. NSGA-III also produced the most evenly spread front, with a spacing metric of 0.0127 versus 0.0156 for NSGA-II, and generated 134 non-dominated solutions against NSGA-II’s 127. Statistical tests, including Wilcoxon signed-rank and Friedman tests with p-values below 0.05 and Cliff’s delta effect sizes, confirmed that the superiority of the two algorithms was not a product of random variation. Notably, the stochastic ranking evolution strategy was withdrawn from the final comparison due to constraint-handling inconsistencies.

The trade-off structure of the Pareto front itself tells a compelling story. Designs in the 800 to 1000 metre range can be built for 45 to 55 crore rupees with moderate emissions of 35 to 42 tonnes of CO2 equivalent, but they require more piers and denser construction. Designs longer than 1400 metres cost less per kilometre, between 32 and 42 crore rupees, yet emissions climb to 48 to 58 tonnes because of greater land take and material consumption. The balanced region between 1100 and 1400 metres forms the knee of the front, where all three objectives reach reasonable values. For the Aurangabad case, a compromise solution of roughly 1250 metres, 42 crore rupees, and about 40 tonnes of CO2 equivalent was selected using knee-point detection combined with expert weighting. Compared with standard IRC-compliant baseline designs, optimized solutions promise cost reductions of 8 to 12 percent, material savings of 10 to 15 percent, construction schedules shortened by 5 to 7 months, and carbon emission reductions of 18 to 22 percent.

Sensitivity analysis added an important practical dimension. When parameter bounds were varied by plus or minus 20 percent, the multi-objective evolutionary algorithms kept hypervolume variations under 5 percent, while Nelder-Mead’s final solutions shifted by up to 18 percent when span length bounds changed. This robustness matters because real projects rarely have perfectly certain boundary conditions. The study also offers scenario-based guidance: NSGA-II or NSGA-III for comprehensive trade-off exploration, CMA-ES with weighted sums when objective priorities are clear, Nelder-Mead or pattern search for rapid refinement of existing designs, and differential evolution for large infrastructure programs where many structures must be optimized and simplicity aids deployment. For the full 25-kilometre corridor, the authors recommend a hybrid strategy: explore the design space with NSGA-II, present the Pareto front to stakeholders, then refine the preferred design with CMA-ES using stakeholder-derived weights.

The authors acknowledge limitations, including the focus on bound-constrained optimization rather than the full complexity of geometric compatibility and safety-factor constraints, and they point toward future work on uncertainty quantification, hybrid algorithms, integration with multi-criteria decision-making methods such as TOPSIS and AHP, and transfer learning from previous flyover designs across cities. They also note practical barriers to adoption, including stakeholder resistance to non-traditional design parameters, procurement limitations, and regulatory approvals. All data and code are deposited in a DOI-bearing Zenodo repository to support reproducibility. As cities worldwide grapple with the twin demands of mobility and climate responsibility, the message of this study is clear: no single algorithm dominates every scenario, but with the right tool matched to the right project context, flyover design can become measurably cheaper, faster to build, and significantly greener.

Subject of Research: Multi-objective optimization algorithms for sustainable flyover design

Article Title: Comparative analysis of multi-objective optimization algorithms for sustainable flyover design framework

Article References: Ramchandani, J. R., & Ghosh, S. (2026). Comparative analysis of multi-objective optimization algorithms for sustainable flyover design framework. Frontiers in Built Environment, 12, Article 1915058. https://doi.org/10.3389/fbuil.2026.1915058

Image Credits: AI Generated

DOI: 10.3389/fbuil.2026.1915058

Keywords: multi-objective optimization, NSGA-II, NSGA-III, flyover design, radial basis function surrogate models, carbon emissions, Pareto front, sustainable infrastructure, structural engineering, differential evolution, CMA-ES, Aurangabad

The concept behind the study is elegantly practical. Crude oil arriving at a desalting unit is typically cold and viscous, and its water-oil emulsion is stubbornly stable, which makes salt removal difficult. Heating the crude before desalting reduces viscosity, destabilizes the emulsion, and improves separation. Meanwhile, the furnace exhaust leaves at approximately 973 kelvin, about 700 degrees Celsius, brimming with recoverable sensible heat. Rather than routing that heat through an intermediate conversion step, the researchers proposed transferring it directly to the crude stream through a shell-and-tube heat exchanger, avoiding the losses and complexity associated with boilers, organic Rankine cycles, or thermoelectric devices. Direct reuse, they argue, is the most attractive route when a compatible heat demand exists within the same plant.

Deciding which fluid goes where inside the exchanger was itself a consequential engineering choice. Following the guidance of industry specifications API 660 and IOGP S-614, the team assigned the crude oil to the tube side and the exhaust gas to the shell side. The rationale is rooted in mechanics and maintenance: the higher-pressure, fouling-prone crude benefits from the tube side, where mechanical cleaning and hydroblasting are straightforward, while the high-volume, pressure-drop-sensitive gas stream performs better on the shell side, where baffles can enhance heat transfer without excessive hydraulic penalty. This allocation reduces required shell thickness, controls fabrication cost, and keeps the dirtiest fluid in the easiest-to-clean passage.

What sets this work apart from earlier heat-exchanger studies is the sheer breadth of the design exploration. The researchers constructed a discrete design space of 192 configurations by combining four tube diameters ranging from 19.05 to 38 millimeters, four tube lengths from 1 to 3 meters, three segmental baffle types, and four tube arrangements including triangular and square patterns. Each configuration was simulated using three-dimensional, steady-state computational fluid dynamics with the RNG k-epsilon turbulence model, chosen for its established reliability in flows with curvature, recirculation, and strong strain rates. The simulations ran on the Simorgh supercomputer at Amirkabir University of Technology, with a final mesh of roughly 14 million cells verified through a Grid Convergence Index analysis showing discretization uncertainties below 0.6 percent.

The team deliberately avoided the classic trap of optimizing for a single metric. Instead, they evaluated four normalized indicators for every design: shell-side heat-transfer performance, total pressure drop, capital cost, and entropy generation, the last serving as a second-law measure of thermodynamic irreversibility. Entropy generation in a heat exchanger arises from two competing mechanisms, heat transfer across finite temperature differences and viscous friction, and minimizing it forces a design to respect both thermal and hydraulic realities simultaneously. Cost was estimated through established power-law correlations linking heat-transfer area and pumping power to equipment expense, providing a comparative economic yardstick across the 192 candidates.

The parametric results revealed a landscape of stubborn trade-offs. Increasing tube length reliably improved heat transfer by enlarging the exchange area and extending fluid residence time, but it simultaneously raised pressure drop and capital cost, and its effect on entropy generation was sometimes non-monotonic as thermal and frictional irreversibilities shifted in relative importance. Larger tube diameters eased hydraulic constraints but did not guarantee better overall performance; at the largest diameter tested, cost penalties mounted sharply. Baffle type emerged as a key lever on entropy generation, while tube arrangement played a comparatively minor role. Sensitivity analysis confirmed that tube diameter dominated the cost indicator, tube length governed thermal performance, and baffle configuration most strongly shaped thermodynamic losses.

To navigate this crowded trade-off space, the researchers applied Pareto analysis, identifying 141 non-dominated designs out of the 192 candidates, a large set that reflects how richly the objectives conflict in a discrete design space. They then ranked the Pareto solutions by their Euclidean distance to an ideal point representing perfect performance on all four normalized criteria, a transparent method that avoids the subjective weighting that can bias multi-objective decisions. The winning configuration, designated D2L4P4B1, pairs a 25.4-millimeter tube diameter with a 3-meter length, a 45-degree rotated square tube arrangement, and single-segmental baffles. It scored 0.903 on normalized heat transfer, 0.954 on entropy-generation performance, and a respectable 0.76 on pressure drop, with the shortest distance to the ideal point at 0.4433.

The CFD visualizations of the optimal design bring the physics to life. Temperature contours show a searing hot zone near the exhaust-gas inlet where the largest temperature gradients, and therefore the most intense heat transfer, occur, followed by a gradual cooling of the gas as it flows across the tube bundle. On the tube side, the crude oil warms steadily along its path, confirming effective absorption of the recovered heat, a rise that directly translates into lower viscosity and easier emulsion breaking downstream. Pressure contours reveal a smooth, gradual decline from inlet to outlet with no abnormal jumps, indicating that the optimized geometry keeps hydraulic losses within acceptable bounds for the viscous crude service.

The validation of the numerical framework lends credibility to the findings. Comparisons against published CFD data from Li and He showed maximum deviations of just 2.34 percent for the shell-side heat-transfer coefficient and 9.8 percent for pressure drop, good agreement for a system as geometrically complex as a baffled shell-and-tube exchanger. The authors are candid about the limits of their approach: the study relies on steady-state simulations with discrete design variables, excludes fouling evolution and transient operation, and validates against numerical rather than experimental data. They also note that a plant-specific design would need to impose the desalter’s required inlet-temperature range as an explicit constraint.

Even with those caveats, the study offers a compelling template for refineries seeking to squeeze value from their stacks. It demonstrates that the best heat-recovery design is never the one that maximizes heat transfer alone, nor the cheapest, nor the thermodynamically cleanest, but the one that balances all three against hydraulic reality. By coupling a rigorous CFD campaign with Pareto-based multi-objective selection and entropy-generation analysis, the researchers have shown that furnace exhaust at Marun can be converted from an environmental liability into a process asset, preheating crude oil, easing the burden on the desalting unit, and cutting the fuel that would otherwise be burned to do the same job. For energy-intensive industries worldwide, the message is that the geometry of a heat exchanger, down to the millimeter of a tube diameter and the angle of a baffle, can decide whether waste heat is truly recovered or merely relocated.

Subject of Research: Multi-objective CFD optimization of a shell-and-tube heat exchanger for recovering refinery furnace exhaust heat to preheat crude oil before desalting

Article Title: Multi-objective design and optimization of a shell-and-tube heat exchanger for recovering Marun furnace exhaust heat and preheating crude oil entering a desalting unit

Article References: Ghodrati, M., Deymi‑Dashtebayaz, M., Khoshnevis, A. B., & Yazdani, M. (2026). Multi-objective design and optimization of a shell-and-tube heat exchanger for recovering Marun furnace exhaust heat and preheating crude oil entering a desalting unit. Results in Engineering, 32, Article 113398. https://doi.org/10.1016/j.rineng.2026.113398

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113398

Keywords: waste heat recovery, shell-and-tube heat exchanger, CFD simulation, multi-objective optimization, entropy generation, Pareto front, crude oil preheating, refinery furnace, furnace exhaust gas, pressure drop, baffle design, energy efficiency

The study, authored by Farnam Dehghani and Mohammad Agha Shafiyi of the Department of Electrical Engineering at Shahid Beheshti University in Tehran, presents a comprehensive framework for integrating renewable energy sources into coupled electricity and natural gas systems. Rather than treating policy design as a purely political exercise, the researchers built a nonlinear programming optimization model, solved through stochastic programming, that captures the uncertainties of consumer demand patterns and the inherent variability of renewable generation. The result is a systematic toolkit that policymakers can use to design incentive policies that encourage private sector investment while respecting governmental budget constraints.

At the heart of the framework is a tension that every energy minister recognizes: the policymaker must make a trade-off between reducing public costs and increasing returns for private investors. Feed-in tariffs, grants, and tax instruments all shift money between the state and private developers, and each combination produces a different outcome for both sides. To navigate this trade-off, the authors employ a sensitivity analysis, a Pareto front diagram, and multi-criteria decision-making based on the technique for order preference by similarity to ideal solution, known as TOPSIS. Together, these tools allow the policymaker to identify the incentive scheme that best balances the competing objectives rather than simply optimizing one at the expense of the other.

What makes the study particularly timely is its treatment of two emerging technologies that most energy system models have historically ignored: bitcoin miners and power-to-gas conversion. Bitcoin mining has become notorious for its enormous electricity appetite, but the researchers show that, when deployed as a flexible load within a hybrid renewable energy system, miners can act as a financial shock absorber. Because mining operations can be switched on and off almost instantaneously, they can soak up surplus wind and solar generation that would otherwise be curtailed, converting otherwise wasted electrons into revenue. This revenue stream improves the economics of renewable projects, which in turn allows governments to offer smaller subsidies while still delivering attractive returns to private investors.

The concept of economic complementarity plays a central role in this logic. Wind and solar resources often have complementary generation profiles: the wind blows more strongly at night and in winter, while the sun shines during the day and in summer. By co-locating hybrid wind and solar capacity and sizing the mix according to these complementary patterns, the system produces power more consistently, reducing the need for backup generation, storage, and grid reinforcement. The framework explicitly models hybrid renewable capacity, curtailment, and the availability factors of both photovoltaic and wind stations, allowing the optimizer to find the blend that minimizes total system cost under uncertainty.

Power-to-gas technology extends this flexibility into the gas network. When renewable generation exceeds demand, electrolyzers can convert the surplus electricity into hydrogen, which can then be transformed into synthetic natural gas through methanation and injected into existing gas pipelines or stored in gas tanks. Conversely, gas-to-power units can convert stored gas back into electricity when renewable output falls short. This bidirectional coupling between the electricity and gas networks effectively turns the gas system into a vast seasonal storage reservoir, smoothing out the mismatches between renewable supply and consumer demand that plague high-penetration renewable grids. The model tracks gas flows, state of charge in storage tanks, and the efficiencies of electrolysis and methanation, capturing the full techno-economic picture of the coupled system.

The technical architecture of the optimization model is ambitious. It represents the electricity network as a set of buses connected by transmission lines, complete with power flow constraints, reactance parameters, and the option to reinforce weak lines at a defined cost per megawatt. Conventional fossil generators compete with renewable installations and their associated clean technologies, with operational cost factors assigned to each technology. Investment costs are specified for photovoltaic arrays, wind stations, power-to-gas and gas-to-power facilities, gas tanks, transmission reinforcement, and bitcoin miners. The stochastic layer of the model assigns probabilities to different scenarios of demand and renewable availability, ensuring that the chosen investment and policy mix remains robust across a range of possible futures rather than being tuned to a single idealized forecast.

On the policy side, the framework incorporates several incentive instruments as explicit decision variables. Grants expressed in dollars per megawatt of installed capacity, feed-in tariffs expressed in dollars per kilowatt-hour, and tax percentages all enter the objective function, alongside the operational costs, investment costs, and the profits flowing to private investors. The penetration factor of renewable energy sources can be constrained to reflect political targets, while the maximum potential capacity of wind and solar stations at each network node reflects physical resource limits. By varying these policy levers and observing the resulting cost and return outcomes, the sensitivity analysis reveals which instruments deliver the most private investment per dollar of public expenditure.

The Pareto front diagram then makes the trade-off explicit. Each point on the front represents a policy combination that cannot be improved in one objective without worsening the other: pushing further toward lower government costs necessarily reduces private returns, and vice versa. TOPSIS provides a principled way to select a single point on this front by ranking the candidate policies according to their similarity to an ideal solution, one that simultaneously minimizes governmental cost and maximizes investor profit. This structured approach replaces the ad hoc bargaining that often characterizes energy policy negotiations with a transparent, reproducible decision process.

The headline numerical result, that well-chosen policies can save governments up to 100 million dollars per year while improving conditions for private investment, underscores the scale of the stakes. Energy transitions are measured in trillions of dollars of investment over the coming decades, and even percentage-point improvements in the efficiency of incentive design translate into enormous sums. The study also carries a provocative implication for the ongoing debate about cryptocurrency: rather than simply banning or tolerating bitcoin mining, policymakers may be able to harness its flexibility as a grid service, provided the miners are coupled to renewable generation and dispatched in a way that absorbs curtailed energy rather than competing with other demand. Combined with power-to-gas storage and economically complementary hybrid renewable portfolios, the framework suggests that the tools for a cheaper, faster energy transition may already exist, waiting only for smarter policy design to deploy them. The authors report no external funding for the research and declare no conflicts of interest, and the underlying data are provided in the supplementary material accompanying the article.

Subject of Research: Techno-economic and policy optimization of renewable energy integration with clean technologies and green bitcoin mining

Article Title: Techno-economic and policy investigation of renewable energy integration into energy systems considering clean technologies and green bitcoin mining

Article References: Dehghani, F., & Shafiyi, M. A. (2026). Techno-economic and policy investigation of renewable energy integration into energy systems considering clean technologies and green bitcoin mining. Clean Technologies and Environmental Policy, 28(10), Article 269. https://doi.org/10.1007/s10098-026-03625-2

Image Credits: AI Generated

DOI: 10.1007/s10098-026-03625-2

Keywords: renewable energy, bitcoin mining, power-to-gas, energy policy, incentive schemes, stochastic programming, hybrid renewable systems, Pareto front, TOPSIS, energy transition, clean technologies, energy economics

The new method, called MOPSO-NMF, is the work of Mehdi Rahimi, Keyhan Khamforoosh, and Vafa Maihami, and it fuses two well-established but previously separate strands of computational intelligence. The first is multi-objective particle swarm optimization, a swarm-based search technique inspired by the way flocks of birds collectively explore a landscape. The second is non-negative matrix factorization, or NMF, a linear algebra tool that decomposes large matrices of non-negative numbers into lower-rank components. NMF has a long history in machine learning, famously used to learn the parts of objects in images and topics in document collections, and it has been applied to community detection before. What makes the new work distinctive is the specific role it assigns to NMF: rather than using it as the community detector itself, the researchers use it to construct the starting population for the evolutionary search.

To understand why that matters, it helps to look at how standard evolutionary community detection algorithms operate. These methods maintain a population of candidate solutions, each one a proposed partition of the network’s nodes into communities. The population evolves over generations as better solutions are selected, recombined, and mutated, guided by one or more quality functions such as modularity. The trouble, the authors argue, lies at the very beginning of this process. Most evolutionary algorithms initialize their populations randomly, which means the first generation of candidate partitions is essentially noise. On a noisy, explicit network topology, such random starts frequently lead to premature convergence, where the search locks onto a mediocre solution early and never escapes, and to poorly defined community boundaries where ambiguous nodes are assigned arbitrarily.

MOPSO-NMF sidesteps this problem by never searching the raw topological space at all. Instead, the algorithm first applies non-negative matrix factorization to the network’s adjacency representation, extracting a low-rank latent representation of the network. In this compressed latent space, each node is described by its membership strengths across a set of latent factors, and the underlying block structure of the network, obscured by noise in the original topology, becomes far more visible. The researchers then use this latent representation to generate structure-aware initial particles for the particle swarm. In other words, the swarm does not begin its journey from random guesses; it begins from educated hypotheses about where the communities are, hypotheses derived directly from the statistical structure of the network itself.

The multi-objective side of the framework matters just as much. Community detection is inherently a trade-off problem: one quality function may reward densely connected groups while another rewards well-separated ones, and optimizing a single measure can distort the final partition. Multi-objective evolutionary algorithms handle this by searching for a whole family of compromise solutions, the Pareto-optimal set, from which a decision maker can later select. By running the particle swarm in this multi-objective mode, MOPSO-NMF explores the trade-off surface between competing objectives rather than committing to a single, possibly misleading, definition of what a good community looks like. The swarm’s particles move through the latent space, guided by personal and collective experience, gradually assembling a front of high-quality candidate partitions.

One further innovation refines the details that matter most in practice: the boundary nodes. In almost every real network, the vast majority of nodes are easy to classify because they sit deep inside a community, surrounded by their own kind. The difficult cases are the nodes at community borders, which maintain ties to several groups at once and are the usual source of errors in any partition. MOPSO-NMF incorporates a boundary-based local search operator that specifically targets these ambiguous nodes. Working in the latent space, the operator re-examines the assignments of border nodes and adjusts them when a better placement is available, polishing the partitions produced by the swarm without disturbing the parts of the solution that are already well settled.

The experimental evaluation is where the approach shows its teeth. The researchers tested MOPSO-NMF on a battery of synthetic and real-world benchmark networks, comparing it against fifteen state-of-the-art community detection algorithms. On the Karate club network, the famous social network of a university karate team that split into two factions, and on the Dolphin network of bottlenose dolphin associations in Doubtful Sound, New Zealand, the algorithm achieved a normalized mutual information score of exactly 1, meaning it recovered the known community structure perfectly. On the Football network of American college football games, it scored 0.9547, and on the Political Books network of political titles linked by co-purchase patterns, it reached 0.8430. Precision, recall, and F-measure scores told the same story: perfect 1.00 across all three measures on Karate and Dolphin, 0.95, 0.97, and 0.96 on Football, and 0.94, 0.95, and 0.95 on Political Books.

Those numbers are more than a leaderboard. Normalized mutual information is a stringent measure that quantifies how much information the detected partition shares with the ground-truth partition, and a score of 1 indicates exact agreement. The fact that a single framework can hit that ceiling on multiple networks while remaining competitive on harder, noisier benchmarks suggests that the three ingredients, NMF-based initialization, multi-objective optimization, and boundary refinement, are genuinely complementary rather than redundant. The initialization gives the swarm a head start informed by the network’s latent structure, the multi-objective search keeps the exploration balanced across competing quality criteria, and the local search cleans up exactly the nodes where automated methods typically stumble.

The broader significance lies in a lesson that resonates across computational intelligence: where a search starts often matters as much as how it searches. Population initialization has long been recognized as a critical factor in the performance of metaheuristics, and a growing body of literature shows that informed, deterministic, or hybrid initialization strategies can dramatically accelerate convergence and improve solution quality. By importing a matrix factorization technique from the machine learning toolbox into the initialization stage of a swarm algorithm, the Iranian team illustrates a productive convergence between the two fields. NMF provides the statistical intuition; the evolutionary framework provides the global search power and the flexibility to juggle multiple objectives; and the boundary operator supplies the fine-grained local polish that global methods lack.

There are, of course, practical questions that remain open. The authors note that the code used to implement the algorithm is available from the corresponding author upon reasonable request, and the benchmark datasets are publicly available, which should make independent replication straightforward. Scaling structure-aware initialization to networks with millions of nodes, handling overlapping communities where nodes genuinely belong to several groups at once, and adapting the latent representation to dynamic networks whose structure changes over time are all natural next steps. But the core result stands on its own: when the search for hidden communities begins from a representation that already understands the network, rather than from blind random guessing, the communities come into focus with remarkable clarity. For a problem that underpins everything from recommendation systems to epidemiology, that is a starting point worth paying attention to.

Subject of Research: Community detection in complex networks using a multi-objective particle swarm optimization algorithm initialized with non-negative matrix factorization

Article Title: A multi-objective evolutionary algorithm based on non-negative matrix factorization for community detection in complex networks

Article References: Rahimi, M., Khamforoosh, K., & Maihami, V. (2026). A multi-objective evolutionary algorithm based on non-negative matrix factorization for community detection in complex networks. Neural Computing and Applications, 38(19), Article 804. https://doi.org/10.1007/s00521-026-12599-7

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12599-7

Keywords: community detection, complex networks, non-negative matrix factorization, multi-objective optimization, particle swarm optimization, evolutionary algorithms, Pareto front, modularity, benchmark networks, machine learning, graph clustering, swarm intelligence

The algorithm, called GSCMO, was developed by Junhua Liu, Lei Fan, Wei Zhang, Yuelin Gao, Mengnan Tian, Meng Wang, and Xueqing Yan, with affiliations spanning Xi’an Polytechnic University, Northeastern University, and North Minzu University. Their work, published in the journal Cluster Computing, introduces a dual-population framework in which two separate pools of candidate solutions evolve in parallel, each guided by a different kind of information extracted from the population itself. The key insight is that the process of generating new solutions, often treated as a mechanical afterthought in algorithm design, can dramatically shape whether the search succeeds or stalls.

To understand why this matters, it helps to picture how evolutionary algorithms work. Inspired by biological evolution, these methods maintain a population of candidate solutions, evaluate how well each one performs, and then repeatedly select the fittest individuals, mix their traits, and introduce random mutations to produce offspring. Over many generations, the population converges toward a set of trade-off solutions known as the Pareto front, which represents the best possible compromises among competing objectives. When constraints enter the picture, the search space becomes littered with infeasible regions where solutions violate one or more limits, and the algorithm must somehow navigate toward the narrow corridors where feasibility and optimality intersect.

Most existing constrained multi-objective evolutionary algorithms concentrate their cleverness on selection, deciding which solutions survive to the next generation based on objective values and constraint violations. The GSCMO team argues that this focus leaves a critical lever untouched. If the offspring are generated poorly, no amount of sophisticated selection can rescue the search. A population that clusters prematurely in one corner of the decision space, the high-dimensional region where the actual decision variables live, may never discover the feasible pockets that hold the best answers. Conversely, a population that wanders too widely may never refine promising solutions to the precision that real engineering demands.

GSCMO addresses this by splitting its labor between a main population and an auxiliary population, each equipped with its own information-assisted generation strategy. The main population’s strategy draws on entropy information, a statistical measure borrowed from information theory that reflects how the population is distributed across the decision space. When entropy indicates that the population is becoming too concentrated, the algorithm dynamically adjusts its mutation strength to push individuals apart, enhancing diversity and maintaining broad exploration. This adaptive tuning gives the algorithm a better shot at locating potential feasible regions that a rigid, fixed-strength mutation scheme would simply walk past.

The auxiliary population takes the opposite approach, exploiting neighborhood information that captures the proximity relationships between individual solutions. Using Manhattan distance, the grid-like distance measure that sums absolute differences along each dimension, the algorithm identifies the closest neighbors of each individual. It then selects parents from within these tight local neighborhoods to breed high-quality offspring, sharpening the algorithm’s ability to conduct fine-grained local search. In effect, the two populations form a division of labor: one casts a wide net across the decision space while the other drills down into the most promising areas, and the interplay between them balances exploration against exploitation, the classic tension at the heart of all search algorithms.

The technical elegance of this design lies in how the information is gathered. Neither entropy nor neighborhood structure requires extra evaluations of the objective functions, which are often expensive in real applications such as aerodynamic shape optimization or optimal power flow analysis. Instead, both measures are computed from the population’s existing coordinates and performance records, making the assistance essentially free. This matters because evaluation budgets in industrial optimization are frequently so tight that even a handful of additional function calls per generation can double the computational cost of a run.

The team put GSCMO through extensive experiments against a battery of state-of-the-art constrained multi-objective evolutionary algorithms, testing on benchmark suites that include the well-known CEC 2009 test instances and the difficulty-adjustable constrained multiobjective test problem toolkit developed by Fan and colleagues. Performance was assessed using standard quality indicators and validated statistically with the Wilcoxon rank-sum test, the field’s conventional tool for determining whether one algorithm’s advantage over another is genuine or merely noise. The reported results show that GSCMO achieves high competitiveness across the comparison set, performing particularly well on problems where the feasible regions are hard to reach and where maintaining decision-space diversity determines whether those regions are found at all.

The significance of the work extends beyond a single algorithm’s benchmark scores. Constrained multi-objective optimization underpins applications from collaborative path planning of multi-unmanned surface vehicles to constrained aerodynamic design, and the field has seen a wave of innovation in recent years, including two-archive strategies, push-and-pull search frameworks, coevolutionary schemes, and evolutionary multitasking approaches that borrow knowledge from auxiliary problems. By shifting attention to the solution generation process itself and showing that population statistics like entropy can steer mutation adaptively, GSCMO adds a new dimension to this design space that other researchers can build upon.

The research, supported by the National Natural Science Foundation of China and the Scientific and Technological Program of Xi’an, arrives at a moment when the demands on optimization algorithms are growing faster than classical methods can handle. As systems become more interconnected and design spaces more tangled with constraints, the difference between a mediocre and an excellent search algorithm translates directly into cheaper designs, safer operations, and better use of computational resources. The GSCMO team’s message is deceptively simple: look at how your solutions are born, not just how they are judged, and the search will thank you. For a discipline that has spent decades perfecting the art of selection, that shift in perspective may prove to be one of the most consequential ideas in constrained evolutionary computation.

Subject of Research: A dual-population constrained multi-objective evolutionary algorithm using entropy and neighborhood information for solution generation

Article Title: Information-assisted solution generation based dual population constrained multi-objective evolutionary algorithm

Article References: Liu, J., Fan, L., Zhang, W., Gao, Y., Tian, M., Wang, M., & Yan, X. (2026). Information-assisted solution generation based dual population constrained multi-objective evolutionary algorithm. Cluster Computing, 29(12), Article 712. https://doi.org/10.1007/s10586-026-06492-0

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06492-0

Keywords: constrained multi-objective optimization, evolutionary algorithm, entropy information, neighborhood information, dual population, decision space, Pareto front, mutation strength, Manhattan distance, local search, solution generation, Cluster Computing

The core insight behind the work is that classification errors across classes can conflict. A network that excels at distinguishing shirts from coats in the Fashion MNIST dataset might struggle with sneakers versus sandals, and the conventional practice of averaging errors into a single scalar loss hides these tensions entirely. By contrast, the CMaO-SL formulation defines one objective per class, so a ten-class problem becomes a ten-objective optimization problem. Because no single architecture can simultaneously minimize every class-specific error, the search targets a Pareto set: a collection of non-dominated solutions in which improving performance on one class necessarily comes at some cost to another. This trade-off information, normally invisible to practitioners, becomes an explicit output of the training process.

Solving such a problem is computationally daunting. Not only must the search evaluate many candidate architectures, but it must do so repeatedly to draw statistically meaningful conclusions about a stochastic process. The authors tackled this challenge with a two-pronged strategy combining transfer learning with evolutionary neural architecture search. Every network in their framework is split into two parts: a deep, pre-trained feature extractor kept completely frozen, and a shallower classification sub-network that is allowed to evolve. The feature extractor is derived from a ResNet-18 network pre-trained on ImageNet, with its final layers removed and its pooling adjusted to output 2,304 features. Because the backbone is never retrained or fine-tuned, the expensive part of the computation is paid only once, and the search can concentrate on the smaller, cheaper classification head.

The evolutionary search itself adapts MOEA/D, a well-established decomposition-based multiobjective evolutionary algorithm, to the architecture search setting. The user first defines a master architecture for the classification sub-network, consisting of fully connected layers with batch normalization and ReLU activations. Each individual in the evolving population is a binary mask vector that dictates which connections in the master architecture are pruned; where a mask entry is zero, the corresponding connection is removed and its weight fixed at zero. This pruning-as-search strategy generates a rich family of alternative architectures from a single blueprint. Each candidate is then trained with gradient descent, evaluated per class on a validation set, and assigned a performance vector whose components are one minus the F1 score for each class.

The evaluation procedure embeds an early-stopping mechanism to guard against overfitting. During each candidate’s five training epochs, the algorithm tracks a scalar validation error and retains the best-performing weight vector, ensuring that later epochs cannot degrade the network that gets assessed. The authors used two training loss functions in their experiments, cross-entropy and a weighted average of per-class binary cross-entropy aligned with each sub-problem’s reference direction, and two corresponding validation error functions. This produced five distinct search configurations, three evolutionary and two non-evolutionary, the latter serving as controls in which independent scalar searches replaced the coupled evolutionary process.

The experimental study spanned six benchmark datasets drawn from very different domains: Fashion MNIST, K-MNIST, CIFAR10, and three medical image collections from the MedMNIST suite, namely BloodMNIST, DermaMNIST and OCTMNIST. Each configuration was run fifteen times, allowing a rigorous statistical treatment that is rare in the Pareto-based architecture search literature. After confirming with a Shapiro-Wilks test that the results were not normally distributed, the authors compared configurations using medians and pairwise Mann-Whitney-Wilcoxon tests at a significance level of 0.05, alongside the IGD+ performance indicator, which measures how well an obtained front approximates a reference Pareto front.

The results were consistent across datasets and evaluation sets. Networks trained with cross-entropy loss produced better-approximated Pareto fronts than those trained with the weighted binary cross-entropy variant, and the evolutionary searches matched or beat their non-evolutionary counterparts on front quality while producing sets of networks with lower total complexity. Parallel coordinate plots of per-class F1 scores made the conflicting nature of the objectives vividly visible: the line connecting a classifier’s best performance on one class typically dipped on others. Crucially, when the authors picked the single best network from each evolutionary run’s set, measured by overall validation accuracy, those networks matched the accuracy of the unpruned master architecture while being substantially smaller, demonstrating that the many-objective formulation finds leaner models without sacrificing performance.

The benchmark comparisons are perhaps the most striking part of the study. When the networks discovered by the proposed method were plotted against published reference results for ResNet-18-based classifiers on the same datasets, no published result dominated the new networks on the combined axes of accuracy and complexity. The full ResNet-18 carries roughly 11.5 million parameters relevant to the comparison, whereas the evolved networks came in at around 2.5 million, meaning the reference architecture is about 360 percent more complex. That dramatic reduction in size cost less than 8 percent in accuracy, and on the DermaMNIST skin-lesion dataset the evolved classifiers actually surpassed the published benchmark. For a method that never fine-tuned its feature extractor and trained each candidate for only five epochs, this is a remarkable showing.

The practical implications extend beyond raw benchmark numbers. Because the output of the search is a set of classifiers with exposed trade-offs, users gain flexibility that single-model pipelines cannot offer. A deployment with tight computational budgets can select the smallest acceptable network from the set, while an application where a particular class matters most, say a rare disease category in a medical screening tool, can pick the member that favors that class. Alternatively, the trade-off structure can guide the construction of ensembles, an avenue the authors identify as a priority for future work. They also flag several open questions: whether fine-tuning the shared feature extractor would push accuracy higher, how the approach handles imbalanced data and noisy labels common in real-world problems, and how alternative pruning schemes, such as removing whole layers or blocks, might perform.

The study also fills a conceptual gap in the literature. Although Pareto-based multiobjective machine learning has existed for years, formal treatments of many-objective supervised learning, where the number of objectives exceeds three, have been scarce, and no prior work had combined such a formulation with neural architecture search and a transferred feature extractor. Most existing multiobjective NAS studies juggle only two or three objectives, typically accuracy, model size and inference cost, rather than one objective per class. By demonstrating a working pipeline at ten objectives, backed by fifteen-run statistical analysis per configuration, the Tel-Aviv and Braude team has shown that class-aware, many-objective architecture search is not only feasible but competitive. As automated machine learning matures, approaches like this suggest a future in which the question is no longer simply how accurate a network is, but which compromises it makes, and whether those are the compromises its users actually want.

Subject of Research: Class-based many-objective neural architecture search for deep classifiers using evolutionary optimization and transfer learning

Article Title: Class-based many-objective neural architecture search for deep classifiers

Article References: Mardiks, G., Salih, A., & Moshaiov, A. (2026). Class-based many-objective neural architecture search for deep classifiers. Neural Computing and Applications, 38(19), Article 800. https://doi.org/10.1007/s00521-026-12510-4

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12510-4

Keywords: neural architecture search, many-objective optimization, Pareto front, evolutionary algorithms, transfer learning, deep learning, multi-class classification, MOEA/D, pruning, ResNet-18, AutoML, machine learning

The problem the team set out to solve is a generalization of the classical job-shop scheduling problem, which is already computationally brutal. In a flexible job shop, each operation of each job can be processed on one of several candidate machines, but not every machine is compatible with every operation. Machines differ in their processing characteristics, operations cannot be interrupted once started, jobs must respect precedence constraints that dictate the order of their operations, and each machine can handle only one operation at a time. Scheduling such a system means simultaneously deciding which machine performs each operation, in what sequence the operations run on each machine, and when each operation starts. The number of possible schedules grows explosively with problem size, which is why factories have historically relied on simple dispatching rules that are fast but leave significant efficiency gains on the table.

What makes this study distinctive is its explicit treatment of three competing objectives at once. The first is makespan, the total time needed to complete all jobs, which is the traditional measure of production efficiency. The second is processing-related energy consumption, which reflects the total energy used by machines as they carry out their assigned operations. The third is workload imbalance, characterized conceptually by the standard deviation of machine utilization, so that schedules in which some machines are overloaded while others sit idle are penalized. Balancing workload matters for practical reasons: it extends equipment life, reduces bottlenecks, and makes maintenance planning more predictable. Because these three goals pull in different directions, a schedule that minimizes completion time may waste energy, and an energy-optimal schedule may leave the factory running far longer than necessary. There is no single best answer, only a set of trade-off solutions known as a Pareto front.

To model the problem rigorously, the researchers formulated a mixed-integer linear programming model, or MILP, that captures all the constraints of the system, including machine-operation compatibility, precedence relations, non-preemptive processing, and machine-capacity limits. One mathematical subtlety arose with the workload-balancing objective: the standard deviation of utilization is inherently a nonlinear quantity, which would make the MILP model intractable. The authors therefore used a linear absolute-deviation surrogate in the model, which preserves the spirit of the balancing goal while keeping the formulation linear and solvable. The MILP simultaneously minimizes makespan, energy consumption, and this workload-imbalance measure. Exact models of this kind can certify how close a solution is to the theoretical optimum, but on large realistic instances they can take prohibitive amounts of time, which is precisely why the team did not rely on the MILP alone.

Instead, the proposed framework works in two phases. In the first phase, an adaptive hyper-heuristic takes over. A hyper-heuristic is essentially a method that chooses among lower-level heuristics rather than manipulating the schedule directly. The adaptive version developed here dynamically selects among constructive heuristics, which build schedules from scratch; improvement heuristics, which refine existing schedules; and perturbation heuristics, which shake up a schedule to escape local optima. Throughout this search, the algorithm maintains an archive of non-dominated solutions, meaning solutions that are not outperformed by any other known solution across all three objectives. This phase is designed to explore the vast solution space quickly and to collect a diverse set of good candidate schedules without requiring the exhaustive reasoning that an exact method demands.

The second phase then hands selected solutions from the archive to a normalized weighted-sum MILP formulation for refinement. Crucially, the hyper-heuristic solutions are supplied as warm-start incumbent solutions rather than as fixed decision variables. This means the mathematical solver is free to modify machine assignments, operation sequencing, and start times as it searches for improvements, using the heuristic solution only as a starting point. The warm-start approach accelerates the solver and, importantly, allows the MILP to provide optimality-gap information, telling decision-makers how far the refined solution sits from the best possible bound. This two-phase design marries the exploration strength of heuristics with the precision and provable guarantees of exact optimization, a combination that neither approach achieves on its own.

The evaluation was deliberately rigorous. The framework was tested on industrial and synthetic flexible job-shop scheduling instances and compared against NSGA-II, one of the most widely used multi-objective evolutionary algorithms; MOEA/D, a decomposition-based evolutionary method; classical dispatching-rule heuristics; a standalone version of the adaptive hyper-heuristic; and a single-objective MILP baseline. All methods were evaluated with the same objective definitions and the same final non-dominated filtering procedure, ensuring a fair comparison. For the multi-objective methods, the final external archives were scored using three standard quality indicators: hypervolume, which measures the size of the objective space covered by the Pareto front; inverted generational distance, which measures how close the obtained solutions are to the true best front; and spacing, which measures how evenly distributed the solutions are.

The numbers are striking. Compared with NSGA-II, the proposed method reduced makespan by 4.3 percent, energy consumption by 8.1 percent, and workload imbalance by 8.0 percent, while improving hypervolume by 18.1 percent and cutting inverted generational distance by 63.4 percent. Relative to the best single-phase baseline, the improvements reached 6.5 percent for makespan, 21.6 percent for energy consumption, and 13.4 percent for workload imbalance. The Pareto-quality metrics indicated better convergence, coverage, and distribution of the non-dominated solutions, meaning the method did not merely find a few lucky schedules but mapped out the trade-off landscape more thoroughly and evenly. Statistical analysis confirmed that these improvements are significant rather than artifacts of random variation, lending weight to the claim that the two-phase architecture is genuinely superior for this class of problems.

The practical implications extend well beyond the benchmark instances. Energy-aware scheduling is becoming a cornerstone of sustainable manufacturing, as factories worldwide face rising electricity costs, tightening emissions regulations, and corporate commitments to reduce carbon footprints. Because the framework improves energy consumption by more than a fifth relative to the strongest single-phase baseline, adopting such methods could translate into meaningful savings without any capital investment in new machinery; the gains come purely from running existing equipment more intelligently. The workload-balancing objective adds further value by reducing uneven wear on machines and smoothing maintenance demands. The authors also emphasize reproducibility, presenting the framework as an effective and reproducible approach for balancing productivity, energy efficiency, and resource utilization in sustainable Industry 4.0 manufacturing systems.

For the broader field, the study offers a template for how hybrid intelligence can be engineered into industrial software. The adaptive hyper-heuristic demonstrates that letting an algorithm learn which heuristic to deploy at each moment outperforms committing to any single strategy, while the warm-started MILP shows that exact optimization need not be abandoned in the era of metaheuristics; it can be deployed surgically, where it adds the most value. As digital twins and real-time factory data become standard, scheduling engines of this kind could be re-run continuously as orders arrive, machines go offline, or energy prices fluctuate, turning the abstract mathematics of Pareto optimality into a live operational tool. The research, published open access with a permanent DOI, arrives at a moment when the intersection of artificial intelligence, operations research, and energy management is rapidly becoming one of the most consequential arenas of applied science.

Subject of Research: Multi-objective optimization for energy-aware flexible job-shop scheduling in Industry 4.0 manufacturing

Article Title: Multi-objective optimization for energy-aware job scheduling in industry 4.0

Article References: Danach, K., Haidoura, H., Mansour, A., & Nasser, A. (2026). Multi-objective optimization for energy-aware job scheduling in industry 4.0. Discover Sustainability. https://doi.org/10.1007/s43621-026-04771-9

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04771-9

Keywords: flexible job-shop scheduling, Industry 4.0, energy-aware scheduling, multi-objective optimization, hyper-heuristics, MILP, NSGA-II, Pareto front, sustainable manufacturing, workload balancing, hypervolume, warm-start

The core difficulty that CMOEA-TSD targets is deceptively simple to state: in many constrained problems, the set of solutions that satisfy all constraints — the feasible region — is tiny, fragmented, or shaped like a narrow winding corridor through a vast infeasible expanse. Classical multi-objective evolutionary algorithms, which mimic natural selection by evolving a population of candidate solutions toward a set of optimal trade-offs known as the Pareto front, tend to stumble in such terrain. Populations converge prematurely onto whatever small pocket of decent solutions they first encounter, losing the genetic diversity needed to explore elsewhere. Worse, when the feasible region is sparse, a single population must simultaneously explore the broader landscape, satisfy constraints, and maintain a well-spread approximation of the Pareto front — three demands that pull the search in incompatible directions and often lead to stagnation.

The new algorithm, developed by Weiye Bai, Cunyan Liu, and Wei Wang, attacks this problem by refusing to make one population do everything. Instead, CMOEA-TSD runs three cooperating subpopulations in parallel, each with a distinct mandate. The first is a global search component that approximates the unconstrained Pareto front — the optimal trade-off surface one would obtain if constraints were ignored entirely. This may seem like a detour, but the unconstrained front provides a valuable map of where the best objective values lie, even if those points violate constraints. The second subpopulation performs a diversity-enhanced search: it reformulates the original problem into a relaxed multi-objective form using reference vectors and a dynamically shrinking constraint threshold, effectively treating constraint violation as an additional objective that is gradually tightened over time. The third subpopulation is feasibility-oriented, focusing squarely on driving solutions into the constraint-satisfying region.

This division of labor is orchestrated through a two-stage structure. In the early stage of the search, the algorithm emphasizes exploration: the diversity-enhanced and unconstrained-front populations roam widely, gathering information about the landscape’s structure and locating promising regions that a purely feasibility-driven search would never visit. As the run progresses, a stage-switching mechanism — driven by the overall population’s convergence rate — adaptively adjusts mating strategies, shifting the emphasis toward exploitation. In practical terms, the algorithm first learns where the interesting parts of the search space are, then concentrates its reproductive effort on refining solutions within and near the feasible region. The convergence-rate trigger is crucial: rather than switching stages at a fixed generation count, the algorithm monitors how quickly the population is actually improving, allowing the transition to happen at the right moment for each individual problem.

The technical machinery behind the diversity-enhanced subpopulation deserves particular attention. Reference vectors, a concept popularized by decomposition-based evolutionary algorithms, partition the objective space into directional cones, ensuring that the search maintains coverage across the entire Pareto front rather than clustering on a few segments. By pairing these vectors with a constraint threshold that shrinks dynamically, the algorithm implements a gradual tightening strategy: solutions with moderate constraint violations are tolerated early on because they may carry useful genetic material that points toward the feasible region, but the tolerance is progressively reduced until only fully feasible, well-distributed solutions remain. This approach echoes a growing body of evidence in the evolutionary computation literature that infeasible solutions are not merely waste — they can be essential stepping stones, particularly when the feasible ratio of a problem is vanishingly small.

To validate the design, the authors subjected CMOEA-TSD to an unusually thorough experimental regimen. The algorithm was tested on four benchmark test suites — standardized collections of constrained multi-objective problems engineered to expose specific weaknesses, such as large infeasible regions, deceptive landscapes, and low feasible ratios. It was then applied to nine real-world engineering problems, the acid test for any optimization method, since industrial problems bring irregular constraint structures and noisy objective landscapes that synthetic benchmarks often fail to capture. Across these evaluations, CMOEA-TSD was compared against seven state-of-the-art algorithms, and the results showed significant performance advantages for the new method, both in converging toward the true Pareto front and in maintaining a well-spread set of solutions along it.

Just as important as the headline results are the ablation studies, in which the authors systematically removed components of their algorithm to verify that each part earns its place. These experiments confirmed that both the three-population co-evolutionary structure and the diversity enhancement mechanisms contribute measurably to performance. This kind of component-level validation matters in a field where new algorithms sometimes owe their wins to a handful of clever tricks whose individual contributions are unclear. By demonstrating that the multi-population architecture and the relaxed-search strategy each pull their weight, the study offers a clearer picture of why the method works, not just that it works — a distinction that matters for researchers hoping to build on the approach.

The significance of this work extends beyond the evolutionary computation community. Constrained multi-objective optimization sits at the heart of countless practical applications: flow-shop scheduling for energy savings, industrial copper burdening, optimal power flow in electrical grids, and cleaner production planning have all been framed in these terms in recent literature. When an algorithm fails to find feasible, well-distributed trade-off solutions, engineers are left either violating constraints or accepting inferior designs. A method that reliably handles sparse feasible regions could therefore translate into tangible gains — lower energy consumption, better resource allocation, and more robust engineering designs — across sectors where multiple objectives and hard constraints collide.

The study also reflects a broader trend in the field: the move from single-population, single-phase algorithms toward co-evolutionary frameworks in which specialized populations exchange information and adapt their roles over time. Earlier work explored two-population and dual-archive designs, push-and-pull search strategies, epsilon-constraint handling, and helper-problem-assisted multitasking. CMOEA-TSD pushes this line of thinking further with its tri-population structure and convergence-based stage switching, suggesting that the future of constrained optimization may lie less in any single clever operator and more in architectures that coordinate multiple search behaviors intelligently. The authors, who received no external funding for the study, report no conflicts of interest, and the algorithm’s components — reference vectors, shrinking thresholds, and adaptive mating — are all implementable within standard evolutionary computation platforms.

For practitioners, the message is that the hardest constrained problems — those with narrow, disconnected feasible regions and strongly conflicting objectives — are becoming tractable with the right architectural choices. For researchers, the paper offers a template: divide the search burden, protect diversity as a first-class citizen rather than an afterthought, and let measured convergence rather than arbitrary schedules dictate when to shift from exploring to exploiting. As computational optimization continues to permeate engineering, logistics, and machine learning system design, algorithms like CMOEA-TSD illustrate how a deeper understanding of search dynamics — when to spread out, when to converge, and how to let specialized populations cooperate — can turn previously intractable problems into solvable ones. The full study is available in Cluster Computing, and its benchmark code and experimental protocols provide a foundation for the next generation of constrained multi-objective solvers.

Subject of Research: A multi-population evolutionary algorithm for constrained multi-objective optimization

Article Title: A novel multi-population evolutionary algorithm based on two-stage division and diversity enhancement for constrained multi-objective optimization

Article References: Bai, W., Liu, C., & Wang, W. (2026). A novel multi-population evolutionary algorithm based on two-stage division and diversity enhancement for constrained multi-objective optimization. Cluster Computing, 29(12), Article 732. https://doi.org/10.1007/s10586-026-06515-w

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06515-w

Keywords: evolutionary algorithms, constrained multi-objective optimization, Pareto front, diversity enhancement, co-evolution, constraint handling, reference vectors, stage switching, benchmark test suites, engineering optimization, Cluster Computing, Xi'an Polytechnic University

The core problem the team attacked is flow maldistribution. In a classic parallel flow field, each channel experiences a different hydraulic resistance, so gas preferentially rushes through the outer channels while the central ones starve. The consequences cascade through the entire cell: oxygen-deprived dead zones form in the middle of the active area, local current density collapses, and water droplets accumulate in the sluggish regions, partially flooding the porous gas diffusion layer. In the conventional design simulated by the researchers, dead zones showed oxygen concentrations more than 30 percent below the average, and localized hotspots reached volumetric heat generation rates of 3.6 times 10 to the ninth watts per cubic meter, roughly 2.6 times the catalyst layer average.

The modified design solves this with two cooperating mechanisms. First, the main channels taper: they are wide at both the inlet and outlet manifolds and progressively vary in width between them, defined by a tapering ratio that compares the width difference to the end width. Counterintuitively, a larger tapering ratio produces a lower pressure profile along the main channels, consistent with Bernoulli’s principle, because the cross-section does not simply shrink. Tapering also rearranges the sub-channel connections, redistributing hydraulic losses and suppressing the preferential edge-channel flow. In the simulations, mid-channel flow speeds rose from 0.13 to 0.77 meters per second, and velocities across different channels became far more uniform, reducing central flow starvation by more than a factor of four.

Second, the sub-channel walls are shaped into smooth, connected circular arcs that repeat every wavelength, creating a converging-diverging wave pattern with no sharp corners. As gas flows through these undulations, centrifugal forces drive it outward and generate pairs of stationary, counter-rotating vortices, akin to the Dean vortices familiar from curved pipe flow. The enhanced design exhibited vortex pairs with a vorticity magnitude of 3000 per second inside every sub-channel, whereas the simple parallel design showed almost none. These vortices perform double duty: they mix fresh oxygen from the channel core toward the gas diffusion layer, and they sweep along the layer surface, repeatedly stripping away the stagnant boundary layer that would otherwise throttle diffusion.

The payoff is visible in the species and water maps. Oxygen concentration at the gas diffusion layer-catalyst layer interface rose 38 percent, from 0.0045 to 0.0062 kilomoles per cubic meter, and the spatial variation dropped from more than 30 percent below average in the dead zones to under 5 percent. Under limiting-current conditions, where concentration losses dominate, the limiting current density climbed 40 percent, from 1.5 to 2.1 amperes per square centimeter. Water saturation fell from 0.13 to 0.11 while membrane water content stayed constant, because higher local gas speeds and vortex shear detach droplets and push them downstream, and the diverging wave sections provide sudden pressure drops that aid water removal without demanding much extra compressor power.

Simulating every conceivable design with a full three-dimensional, multiphase computational fluid dynamics model would be prohibitively expensive, so the researchers built a surrogate. They ran a Central Composite Design of 81 simulation cases, spanning five operating variables: anode and cathode stoichiometries, pressure, wave amplitude, and tapering ratio, each evaluated at three voltages of 0.3, 0.5, and 0.7 volts, plus 20 random unseen cases for testing. The key innovation is the Physics-Informed Response Surface Methodology, or PIRSM. Conventional response surfaces fit a single low-order polynomial to the whole polarization curve, but a fuel cell’s voltage-current relationship spans three physically distinct regimes, activation, ohmic, and concentration losses, and a lone polynomial cannot capture them all. In one demonstration, a conventional fit deviated by 17.7 percent at open-circuit voltage.

PIRSM instead builds three separate second-order response surfaces, each trained only on data from one loss regime, then aggregates them into a third-order polynomial that transitions smoothly between regions and correctly predicts zero output at open-circuit voltage. For the parasitic power ratio, a cube-root transformation motivated by the nonlinear scaling of pressure drop with velocity prevents non-physical predictions such as negative pressures. The resulting models achieved R-squared values of 0.99 against the simulation data for both objectives, with root mean square errors of 0.017 watts per square centimeter for power density and 0.005 for the parasitic ratio, and an R-squared of roughly 0.96 on the held-out test set. The underlying CFD model itself was validated against experimental polarization data with a root mean square error of 0.0222 amperes per square centimeter, and a grid convergence index of 0.58 percent confirmed numerical reliability.

On top of the surrogate sits the second major novelty: an ensemble optimization framework. Rather than relying on a single metaheuristic, the team ran five multi-objective algorithms from distinct families, three evolutionary methods (NSGA-II, SPEA II, and MOEA/D) and two swarm-based methods (multi-objective grey wolf optimization and multi-objective particle swarm optimization). Their Pareto fronts, each reflecting a different search bias, were pooled and filtered by a dominance criterion, in which one solution dominates another only if it is no worse on every objective and strictly better on at least one. Statistical testing over 30 runs per algorithm, using pairwise Wilcoxon signed-rank tests with Bonferroni correction, established a clear hierarchy: NSGA-II was the superior performer, winning 86.7 percent of rank-1 placements with corrected p-values below 0.001, while SPEA II occupied a consistent second tier. The ensemble’s pooled front captured a wider range of trade-offs than any single algorithm, and because the algorithms run in parallel, the wall-clock cost is the maximum of individual run times rather than their sum.

The winning configuration, selected from the ensemble Pareto front using the TOPSIS method, features a tapering ratio of 2.23, a wave amplitude of 0.47 millimeters, a pressure of 4 atmospheres, stoichiometries of 3 on both anode and cathode, and an operating voltage of 0.47 volts. It achieves an oxygen concentration of 0.01396 kilomoles per cubic meter at the catalyst layer interface, roughly 210 percent higher than the simple parallel design, and an oxygen uniformity index of 0.8174, up 26.5 percent. The limiting current density reaches 2.35 amperes per square centimeter, a 56.7 percent increase over the baseline, and the pressure drop falls by a factor of 6.5 relative to the simple parallel layout. The parasitic power ratio of 0.0186 represents an 84.5 percent reduction compared with the conventional design, although the simpler tapered variant without waves retains slightly lower pumping losses.

The researchers caution that their predictions rest on simulations within the tested parameter ranges, and that experimental validation of the waved-tapered geometry remains an essential next step. The smooth arcs of the wave pattern avoid sharp corners and should benefit mechanical integrity, but repeated converging-diverging sections may introduce cyclic stresses, so accelerated stress tests, thermal cycling, and degradation experiments are recommended to confirm durability. Even so, the study demonstrates a compelling template for fuel cell engineering: couple geometric innovations that exploit real flow physics with surrogate models constrained by electrochemistry, and let a diverse committee of optimizers, rather than any single algorithm, chart the trade-off frontier. As hydrogen economies scale up, designs like this one, squeezing more power from every square centimeter while spending almost nothing on parasitic pumping, could help make fuel cells not just clean, but genuinely competitive.

Subject of Research: Multi-objective optimization of a tapered parallel flow field with wavy sub-channels for proton exchange membrane fuel cells using physics-informed response surface methodology

Article Title: Ensemble multi-objective optimization of modified tapered parallel flow field PEM fuel cell using physics-informed response surface methodology

Article References: Ghasabehi, M., & Shams, M. (2026). Ensemble multi-objective optimization of modified tapered parallel flow field PEM fuel cell using physics-informed response surface methodology. Results in Engineering, 32, Article 113366. https://doi.org/10.1016/j.rineng.2026.113366

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113366

Keywords: PEM fuel cell, flow field design, tapered channels, wavy sub-channels, response surface methodology, multi-objective optimization, NSGA-II, Pareto front, computational fluid dynamics, power density, parasitic losses, hydrogen energy

The core problem pyMOODS addresses is one that optimization researchers have long acknowledged but rarely solved. Multi-objective optimization does not yield one optimal design; it yields an ensemble of Pareto-optimal candidates, each representing a different compromise among objectives such as cost, generation capacity, and environmental impact. Choosing among them is a human judgment call, informed by policy goals and stakeholder priorities. Yet the tools available for this post-optimization stage are fragmented. Algorithmic toolkits like pymoo, PlatEMO, and Borg generate Pareto fronts but offer little for the decision-maker. Multi-criteria decision analysis libraries such as pymcdm and pyDecision rank alternatives once objectives are fixed, but expose no exploratory interface. Visualization tools like PAVED and Parasol support interactive exploration but stop short of preference-weighted ranking. What has been missing, the PNNL team argues, is a single interactive platform that connects the visualization layer to a complete decision-support workflow on fronts that have already been computed.

pyMOODS fills that gap with a decoupled software architecture that separates heavy computation from interactive exploration. The frontend is a React and MUI application rendering linked interactive plots built with React Plotly, while a Python backend handles the analytics. User actions issue asynchronous requests to a Flask REST API, which mediates all communication between the two. Because interaction state lives in the frontend while analysis runs in the backend, computation never blocks the interface. A data access module loads, validates, and caches datasets held as CSV and JSON files, and a data processing module performs transformations and filtering before invoking the analytical routines. Notably, the dashboard evolved from early Plotly Dash prototypes into the current React-based implementation, gaining dynamic rendering of parameters, plots, and objective functions, along with better responsiveness under high-dimensional workloads.

One of the framework’s most elegant design decisions is its declarative input schema. A problem formulation is supplied as a single JSON configuration file that links to the data files and assigns each column to one of five categories: input parameters, hyperparameters, decision variables, objective functions, and control inputs. This means a single deployment can serve formulations whose objectives, decision variables, and hyperparameters differ in number and meaning, without any code changes. That flexibility matters because the framework targets real infrastructure applications, such as offshore wind co-design, MTDC-ESS system design, and load-shedding-constrained network expansion, where problem structures vary dramatically across use cases. It is a practical requirement that benchmark-oriented tools, typically evaluated on synthetic test problems like ZDT and DTLZ, simply do not address.

The analytical heart of pyMOODS combines two very different classes of methods. For structuring the solution space, the framework turns to unsupervised machine learning: a Uniform Manifold Approximation and Projection (UMAP) embedding projects the high-dimensional objective and decision space into two dimensions, and density-based HDBSCAN clustering groups similar solutions within that latent space. This gives users a navigable map of the entire front, where clusters of comparable candidates are visible at a glance and outliers stand out immediately. Users can also plot any pair of objectives or decision variables directly, or restrict the working set to solutions sharing a chosen hyperparameter configuration, such as a particular technology option.

For trade-off interpretation, pyMOODS takes a deliberately different path: a deterministic, parameter-free rank analysis. Every solution is ranked on each objective, and a solution’s generalizability is defined as its worst rank across all objectives. A solution is said to specialize in an objective when its rank there is strictly better than that of every more generalizable solution. Because this characterization derives entirely from the rank matrix, it introduces no trained model, no tuned threshold, and no random seed; identical inputs always produce identical generalizer and specializer assignments. This answers a question that ranking alone cannot: not merely which solutions score well under a given weighting, but which specific compromise each one represents. Solutions can additionally be ordered by an aggregate score reflecting user-specified objective weights, giving stakeholders a familiar mechanism for expressing their priorities, and the ranking is computed from the stored front rather than by re-solving, so stakeholders with different priorities can interrogate the same results interactively.

The framework’s workflow comes alive in its demonstration on the MoCoDo dataset for offshore wind farm planning with battery energy storage. The problem involves six objectives, including battery cost, cable material cost, day-ahead revenue, real-time revenue, and two reserve revenue streams, along with two decision variables: cable capacity and battery rated power. A conventional approach would aggregate all revenue streams into one total and all costs into another, reducing the problem to a simple two-way trade-off, but that aggregation obscures how individual solutions perform across specific revenue mechanisms. pyMOODS instead guides the user through four linked views: a scatter plot offering a first overview of the solution space, a ranking table for selecting a candidate, a parallel coordinates plot distinguishing generalizers from specializers across all six objectives, and beeswarm plots showing the distribution of all candidates per objective and decision variable. Selection propagates across views, so a candidate can be examined from several perspectives without losing context.

Performance testing suggests the architecture delivers on its responsiveness promise. On a local deployment running on an Apple M2 Pro with 16 GB of RAM, the team profiled three representative user tasks across two built-in formulations, Cameo_datacenter and MoCoDo_v2, repeating each measurement three times. Formulation initialization took roughly five milliseconds in both cases, while full dashboard load sequences averaged 3.838 seconds for the larger formulation and 0.475 seconds for the smaller one, with filter and selection interactions showing nearly identical latencies. The release also includes automated frontend tests using Playwright, which verify that dashboard components load correctly, the solution table sorts properly, and the generalizer-specializer controls update results, using simulated backend responses to validate the interface without requiring the full backend to run.

A capability comparison against representative open-source tools underscores what makes pyMOODS distinctive. Front-generation frameworks and metaheuristic platforms are not designed to be driven by a decision-maker; multi-criteria decision analysis libraries rank alternatives but expose no exploratory interface; and interactive visualization tools support exploration but stop short of preference-weighted ranking. In the published comparison, pyMOODS is the only tool that simultaneously offers an interactive user interface, operates on precomputed fronts from any solver, supports user-weighted ranking, provides automatic trade-off characterization, and accepts schema-agnostic problem formulations. The authors are candid about the release’s limitations, however: performance has been characterized on only two formulations on a single machine, rank-and-weight aggregation cannot recover non-convex regions of a Pareto front, and no usability evaluation has yet been conducted. Planned features include additional MCDM methods for comparison and an AI chatbot assistant.

The implications reach well beyond software engineering. As power systems worldwide absorb rising penetrations of onsite energy sources and face intensifying extreme weather, the co-optimization of generation, transmission, and storage investments is becoming one of the defining computational challenges of the energy transition. Tools like pyMOODS aim to ensure that the enormous computational effort poured into optimizing these systems does not end in an unreadable list of trade-offs. Released under the BSD 3-Clause License and archived on GitHub and Zenodo, version v0.0.3 puts a complete post-optimization decision-support pipeline into the hands of the planners, engineers, and researchers who must ultimately decide which compromises our future infrastructure will embody. Whether it changes real-world decisions, the authors acknowledge, is a question that future work, and future users, will have to answer.

Subject of Research: An open-source visual analytics framework for multi-objective decision support in large-scale infrastructure planning

Article Title: pyMOODS: An open-source framework for multi-objective decision support in large-scale infrastructure planning

Article References: Mattoo, P., Pham, J. N., Jain, M., Arendt, D., Avila, P., Ramachandran, T., Adetola, V., Yun, J. Y., & Wenskovitch, J. (2026). pyMOODS: An open-source framework for multi-objective decision support in large-scale infrastructure planning. SoftwareX, 36, Article 103095. https://doi.org/10.1016/j.softx.2026.103095

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103095

Keywords: pyMOODS, multi-objective optimization, Pareto front, visual analytics, infrastructure planning, offshore wind, UMAP, HDBSCAN, decision support systems, open-source software, energy systems, Pacific Northwest National Laboratory