The research team, led by engineers at Henan Jiuyu Intelligent Energy Technology, built their approach around a two-stage framework. In the first stage, they generate highly accurate forecasts of wind and solar output using a combination forecasting model. Instead of relying on a single prediction technique, they blend multiple forecasting methods whose weights are optimized by the dung beetle optimizer, a nature-inspired algorithm that mimics how dung beetles navigate and forage. This metaheuristic search automatically tunes the model structure and parameters, squeezing out better accuracy than any individual method could achieve alone.

Forecast accuracy matters enormously in electricity markets. When a scheduler overestimates solar output, expensive fast-response units must ramp up at the last minute; when output is underestimated, cheap renewable energy is curtailed and money is wasted. By feeding sharper forecasts into the second stage of their framework, the researchers give the scheduling model a much more reliable picture of what renewables will actually deliver across each 15-minute interval of the operating day.

The second stage is where shared energy storage enters the picture. Unlike dedicated batteries owned by a single wind farm or solar plant, shared energy storage facilities are independent assets that multiple renewable generators can lease capacity from. This arrangement spreads investment costs across many users, raises utilization rates, and allows the storage operator to serve diverse generation profiles simultaneously. In the scheduling model, the shared storage facility charges when renewable output is abundant and discharges when it is scarce, smoothing the net load curve that thermal plants must follow.

To test their framework, the team simulated a provincial power system containing conventional thermal units, wind farms, photovoltaic plants, and a shared storage station, using real load and weather data from a Chinese province. The results were striking. Compared with single-method forecasting, the combination model reduced prediction error substantially, and the dung beetle optimizer outperformed several competing metaheuristics in tuning the forecast weights. When these improved forecasts drove the scheduling model, total operating costs fell, renewable curtailment dropped, and the thermal fleet operated under less stressful ramping conditions.

The economics of shared storage proved particularly compelling. Because the facility sells capacity leases to multiple generators rather than serving one owner, its effective utilization climbs, and the per-kilowatt cost of storage services falls. The study’s cost-benefit analysis shows that renewable generators leasing storage capacity can lower their overall market participation costs while the storage operator earns steady revenue, creating what the authors describe as a win-win arrangement that supports higher renewable penetration without burdening consumers.

The intra-day scheduling model itself is formulated as a mixed-integer optimization problem. It minimizes total system cost, including thermal generation expenses, renewable curtailment penalties, storage operation charges, and deviation settlement costs under spot market rules. Constraints cover generator output limits, ramp rates, storage state-of-charge bounds, and the market’s deviation assessment mechanism, which penalizes deviations between scheduled and actual output. Solving this problem with accurate renewable forecasts allows the operator to commit resources confidently while respecting all physical and market constraints.

What sets this work apart from earlier studies is the tight coupling of forecasting and scheduling under realistic market conditions. Many previous papers treated prediction and dispatch as separate problems or ignored the deviation settlement rules that govern actual Chinese spot markets. By embedding both within a single framework and testing it on a provincial-scale system, the researchers demonstrate a practical pathway for grid operators seeking to integrate large volumes of variable renewables without compromising reliability or market efficiency.

The implications extend beyond Henan or any single province. As China’s electricity spot markets expand and renewable capacity continues its rapid growth, tools that tame forecast uncertainty while unlocking the economics of shared storage could become standard equipment for system operators. The study suggests that combining nature-inspired optimization with layered forecasting and market-aware scheduling offers a template for building cleaner, cheaper, and more flexible power systems worldwide.

The forecasting stage of the study deserves closer attention, because the choice of component models reflects a deliberate attempt to cover complementary weaknesses. Long Short-Term Memory networks, a recurrent architecture designed to retain information over long sequences, excel at capturing the temporal dependencies that dominate wind and solar output, such as diurnal solar cycles and the persistence of weather systems. Support Vector Machines, by contrast, are grounded in structural risk minimization and tend to perform well on small samples and nonlinear relationships, but they struggle to extract long-term temporal patterns on their own. By fusing the two, the combined model draws on LSTM’s memory of sequence dynamics while using SVM to stabilize predictions where training data are limited or relationships shift abruptly.

The Entropy Weight Method supplies the fusion logic. Rather than fixing weights in advance or averaging them across the whole forecast horizon, it evaluates how much each component model’s errors vary across different time intervals and assigns weights accordingly. When meteorological conditions favor one model during a particular stretch of the day, that model’s influence rises; when conditions change, the weights shift. The equilibrium optimizer then searches the parameter space of the hybrid to minimize overall error. On the study’s test data, the resulting model achieved a coefficient of determination of 0.96, with mean absolute error of 14.45 and root mean square error of 21.60, outperforming the individual forecasting models it was built from.

These error metrics matter because forecast quality propagates directly into market outcomes. In China’s provincial spot markets, generators submit day-ahead schedules and are then assessed against actual output in intra-day and real-time balancing. Persistent deviations trigger economic penalties under deviation settlement rules, so a wind farm that systematically over- or under-declares its expected production faces a compounding financial cost. Sharper forecasts reduce these deviation charges, and they also allow the shared storage operator to plan charge and discharge cycles that genuinely match the shape of renewable output rather than reacting to surprises.

Shared energy storage itself has emerged as a response to a persistent problem in China’s storage sector: low utilization of privately owned batteries. Under the traditional model in which each renewable plant builds and operates its own storage, assets sit idle much of the time because a single plant’s generation profile rarely demands the full capacity of its battery. Centralized construction and unified leasing changes this calculus. Multiple wind and solar operators draw on the same facility at different times, smoothing aggregate demand for storage services and spreading fixed costs across a broader customer base. The study’s case analysis, which produced a total system operating cost of 7,634,351.42 RMB, showed that coordinating the storage operator with renewable producers through this leasing arrangement lowered curtailment and reduced overall costs relative to scenarios without the collaborative mechanism.

The scheduling model’s treatment of uncertainty is also notable. Much of the prior literature on shared storage scheduling treats renewable forecasts as deterministic inputs, effectively assuming the numbers will be right. This study instead builds the forecast error explicitly into the two-stage pipeline, so that the dispatch decisions inherit the statistical properties of the prediction model rather than ignoring them. The authors argue this is what allows the flexible regulation capability of storage to be fully exploited: the battery is positioned in advance for the range of outcomes the forecast suggests, not just for a single point estimate.

The policy recommendations that close the paper point toward market design. The authors suggest that stable operation of provincial spot markets requires rules that properly compensate storage for the flexibility it provides, and that encourage renewable generators to participate in shared rather than isolated storage arrangements. As China’s spot market reforms deepen and renewable penetration rises, the interplay between forecast accuracy, deviation settlement penalties, and storage economics will increasingly determine whether variable generation can be accommodated at reasonable cost. This study’s contribution lies in demonstrating, on a provincial-scale system with real load and weather data, that tightening the link between prediction and dispatch is one practical way to manage that interplay.

Subject of Research: Research on intra-day scheduling model of power system with shared energy storage in the provincial electricity spot market based on renewable energy groups combination forecasting

Article Title: Research on intra-day scheduling model of power system with shared energy storage in the provincial electricity spot market based on renewable energy groups combination forecasting

Article References: Yan, Z., Xue, W., Zhang, Y., Du, X., & Zheng, X. (2026). Research on intra-day scheduling model of power system with shared energy storage in the provincial electricity spot market based on renewable energy groups combination forecasting. Frontiers in Energy Research, 14, Article 1887604. https://doi.org/10.3389/fenrg.2026.1887604

Image Credits: AI Generated

DOI: 10.3389/fenrg.2026.1887604

Keywords: shared energy storage, electricity spot market, renewable forecasting, combination forecasting, intra-day scheduling, power system optimization, energy storage leasing, grid flexibility, Research, intra-day, scheduling, model