Chaotic fire hawk and pufferfish algorithm targets energy-aware resource allocation in B5G IoT networks

Researchers in India have proposed a new hybrid optimization algorithm, called the Hybrid Iterative Chaotic Fire Hawk–Pufferfish Optimization (HICFHP) algorithm, designed to allocate communication and computing resources in Internet of Things networks that operate over Beyond 5G (B5G) infrastructure. The study, published in the journal Cluster Computing, was authored by Pretty Singla and Jitendra Kumar Samriya and appeared online on 25 October 2026 as part of Volume 29 of the journal. The work is a simulation study: the algorithm was tested in a modeled B5G-enabled IoT environment rather than in a deployed network, and all reported performance figures come from that simulation.

The motivation behind the work lies in the growing demands that smart city applications place on wireless infrastructure. The authors note that 5G, first deployed in 2019, can connect up to one million devices within 0.38 square miles, achieves delays as low as one millisecond, and reaches estimated peak speeds of 20 Gbps. Yet even these capabilities may fall short for emerging smart city services such as holographic communications, which demand both very large bandwidth and very low latency. B5G networks, according to the paper, are expected to reach data throughput roughly 50 times greater than 5G at around 1 Tbps, reduce air-link delay to about one tenth of the roughly 100 microseconds that 5G achieves, and operate in the terahertz frequency range between 0.1 and 10 THz, where link speeds from 100 Gbps to the terabit range are anticipated.

Against this backdrop, the researchers argue that existing resource allocation methods struggle with the changing requirements of B5G networks, producing poor energy management, delays, and declining network performance. Their review of prior approaches covers deep Q-learning based resource allocation, game theory-based allocation, non-orthogonal multiple access, combinatorial auction methods, matching theory, and convex optimization. The authors contend that while these methods are effective, they regularly struggle with the high complexity, dynamic changes, and multi-objective nature of B5G IoT networks, often focusing on single objectives or requiring high computational power that limits real-time usability in large-scale deployments.

The proposed HICFHP algorithm combines two bio-inspired metaheuristic techniques: the Pufferfish Optimization algorithm (PuFO) and Fire Hawk Optimization (FiHwO). PuFO, which draws on pufferfish territorial defense behavior, serves as the global exploration phase, maintaining population diversity and preventing premature convergence to local optima. FiHwO, which converges quickly with few control parameters, handles the exploitation phase, refining promising candidate solutions and accelerating convergence. The key innovation is the injection of sine-based iterative chaotic maps into the PuFO initialization stage. These chaotic maps generate deterministic but non-repetitive sequences that spread candidate solutions uniformly across the search space, increasing initial population diversity and postponing premature convergence, which the authors say helps the algorithm avoid local optima and attain improved Pareto-optimal solutions.

The optimization problem is formulated as a weighted combination of seven objectives: energy consumption, latency, throughput, cost, makespan, network usage, and bandwidth utilization. Because these objectives have different numerical ranges, the framework applies min–max normalization to each objective before fitness evaluation, preventing any single metric from dominating the calculation. The system model describes devices generating tasks with unique priority levels, and defines mathematical expressions for total energy consumption, total delay including both communication and computation components, total cost with coefficients for power, bandwidth, and computational resources, makespan as the maximum completion time across tasks, computational overhead, residual energy, and throughput measured as packets successfully delivered per unit time. Notably, the algorithm requires no training phase, unlike learning-based methods such as deep reinforcement learning, which the authors say suffers from state-action space explosion as device numbers grow.

For the evaluation, the framework was implemented in Python and tested in a simulated B5G-enabled IoT environment consisting of 100 IoT nodes connected to a single base station. The device behavior was modeled using a publicly available IoT smart home sensor dataset from Kaggle, containing environmental sensing attributes such as temperature, humidity, and appliance state. The authors are explicit that this dataset serves as a behavioral proxy: raw sensor attributes were transformed into network-relevant traffic characteristics including packet size, arrival rate, and quality-of-service demand levels, and the reported metrics are aggregated network-level performance figures rather than per-sensor physical measurements. The proposed algorithm was compared against the Pelican Optimization Algorithm, the Osprey Optimization Algorithm, and the Hybrid PufferFish Osprey Optimization algorithm, among others.

The simulation results show the HICFHP algorithm achieving lower total energy consumption than competing models across varying maximum transmit power levels. The algorithm reached 0.04106 J at -4 dB and 0.00112 J at 4 dB, while the best existing model consumed 0.04842 J at -4 dB. Total execution time for the HICFHP procedure was 89.77 seconds at -4 dB and 1.43 seconds at 8 dB, compared with 134.77 seconds for WaOA at -4 dB. When the number of nodes was varied from 10 to 50, the algorithm’s total energy usage rose from 0.00057 J to 0.00872 J, and it completed operations with an average duration of 50.09 ms at 50 nodes, against 151.15 ms for WaOA. The algorithm also showed reduced latency, achieving an overall delay of 29.54 ms compared with 90.57 ms for WaOA.

Under varying bandwidth conditions from 10 to 50 Mbps, the HICFHP algorithm recorded total energy use of 13.422 J, 12.723 J, 10 J, 9.344 J, and 7.834 J respectively. At 50 Mbps, it consumed 7.834 J, outperforming HPFOO at 11.324 J, POA at 15.87879 J, OOA at 19.21185 J, ESO at 21.27262 J, and WaOA at 23.0908 J. Operation at 50 Mbps took the proposed algorithm 91.26 milliseconds, while WaOA required 124.4856 ms. A convergence analysis presented in the paper suggests that the chaotic initialization improves population diversity in early iterations and increases global exploration capacity. The authors also report a computational complexity of O(I·P²·D), where I is the number of iterations, P the population size, and D the decision variables, with the quadratic term arising from population-based interactions during exploration; because the population size is kept small and fixed, they argue the algorithm remains suitable for large-scale problems.

The authors position the algorithm’s scalability as a central advantage. Because the population size is fixed and the method involves no training phase, they state that computational complexity remains independent of the number of IoT nodes, and that chaotic initialization preserves solution diversity in large-scale environments. A comparative scalability table in the paper contrasts the algorithm with existing methods on memory consumption and suitability for large-scale B5G IoT networks. The authors also note that the hybrid design does not rely on external archiving, which they say reduces memory overhead compared with buffer-dependent evolutionary algorithms.

The authors are careful to frame the work as preliminary with respect to real-world deployment. They state that future research will focus on validating the HICFHP framework in real-world B5G-enabled IoT environments, investigating practical deployment challenges including dynamic network conditions, user mobility, communication overhead, scalability in ultra-dense IoT scenarios, and computational constraints of edge devices. Planned work includes experimental validation using realistic network traffic models, hardware testbeds, and software-defined networking platforms, along with integration of Multi-Access Edge Computing frameworks and clustering with deep learning methods for real-time optimization. Until such validation is carried out, the reported energy, latency, and throughput gains remain findings from a simulation environment using a smart home sensor dataset as a behavioral proxy, and their transferability to live B5G networks has yet to be demonstrated.

Subject of Research: A hybrid bio-inspired optimization algorithm for multi-objective resource allocation in simulated B5G-enabled IoT networks

Article Title: Multi-objective resource allocation in B5G-enabled IoT networks using a hybrid iterative Chaotic Fire Hawk–Pufferfish Optimization Algorithm

Article References: Singla, P., & Samriya, J. K. (2026). Multi-objective resource allocation in B5G-enabled IoT networks using a hybrid iterative Chaotic Fire Hawk–Pufferfish Optimization Algorithm. Cluster Computing, 29(16), Article 897. https://doi.org/10.1007/s10586-026-06656-y

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06656-y

Keywords: B5G, Internet of Things, resource allocation, metaheuristic optimization, Pufferfish Optimization, Fire Hawk Optimization, chaotic maps, smart cities, energy efficiency, latency, simulation study, Cluster Computing