Physics-Informed Distributionally Robust Multi-Agent Reinforcement Learning for Coordinated New-Type Power System Operation

Deep Reinforcement Learning-Based Energy Management for Hydrogen-Enabled Community Microgrids Under Uncertainty

Physics-Guided Reinforcement Learning for Reliability-Aware Gate Driving in Renewable-to-Hydrogen High-Power Converters

: High-power renewable-to-hydrogen conversion systems impose stringent and dynamically coupled constraints on semiconductor switching behavior, thermal cycling, and electrolyzer degradation. Conventional IGBT gate driving strategies rely on fixed or heuristically tuned parameters that fail to explicitly account for nonlinear electro-thermal dynamics, parasitic interactions, and downstream electrochemical aging mechanisms under stochastic renewable input. This paper reformulates gate driving as a constrained multi-objective optimal control problem and proposes a physics-guided reinforcement learning (PGRL) framework for adaptive gate trajectory morphing in megawatt-scale hydrogen converters. A unified electro-thermal–electrochemical model is constructed to capture nonlinear switching transients, parasitic inductive–capacitive effects, junction temperature evolution, Miner-based fatigue accumulation, DC-link ripple propagation, and ripple-induced electrolyzer degradation. Physics consistency is enforced through differentiable safety projection, residual regularization against governing dynamic equations, and structured policy parameterization reflecting device topology. The learning objective simultaneously minimizes switching energy, voltage overshoot, electromagnetic stress, thermal cycling amplitude, and stack degradation rate. Case studies on a 1.2 MW PEM electrolyzer system demonstrate up to 20% reduction in peak junction temperature rise, 50% ripple suppression during renewable gust events, and extension of projected electrolyzer lifetime beyond 10,000 operating hours under uncertainty. The proposed framework establishes a cross-domain bridge between microsecond-scale semiconductor control and multi-year

Machine-Learning-Assisted Multi-Energy Coupling and Battery–Grid Coordination for Deep Decarbonization of Smart Integrated Energy Systems: Modeling, Optimization, and Applications

In grid-connected smart integrated energy systems with high shares of renewable generation, source-side variability and inadequate coordination among battery storage, other energy carriers, and the external grid limit local renewable-electricity utilization and impede deep decarbonization. This study proposes a machine-learning-assisted, renewable-driven framework for multi-energy coupling and scenario-based multi-objective optimization of electricity–heat–hydrogen–storage systems. Historical meteorological and load data are processed using K-means clustering and Latin hypercube sampling to construct representative operating scenarios across multiple volatility regimes and characterize source–load uncertainty. The equipment model includes photovoltaic arrays, wind turbines, heat pumps, electrolyzers, fuel cells, grid-interactive battery energy storage, thermal storage, and hydrogen storage; cross-carrier conversion dynamics and emissions from purchased electricity and natural gas are embedded in the energy-balance constraints. A mixed-integer linear programming formulation then co-optimizes battery charging and discharging, grid exchange, and other multi-energy flows with respect to operating cost, carbon emissions, and renewable-energy curtailment. At 95% renewable-energy penetration, the proposed method achieves a renewable-energy absorption rate of 91.6% and a curtailment rate of 8.4%. Across the carbon-price cases, annualized operating cost ranges from 126.5 × 104 to 141.2 × 104 USD yr−1, while carbon-emission intensity ranges from 26.4 to 38.5 gCO2/kWheq. Under the specified high-risk grid disturbances, the coordinated strategy limits load shedding to 1.8%—73% below deterministic scheduling and 79% below the heuristic benchmark—and maintains 92.6% hydrogen self-sufficiency. These results provide a data-driven modeling and decision framework for battery–grid coordination and deep decarbonization in smart integrated energy systems.

Beyond Reactive Operation: Sensor-Informed Model Predictive Control for Renewable Hydrogen Storage

Renewable hydrogen storage can absorb surplus wind and solar generation, reduce curtailment, and provide a flexible energy carrier. Its value depends not only on the electrolyzer, storage tank, and fuel cell, but also on the operational strategy used to coordinate these components. This study develops a sensor-informed model predictive control (MPC) framework for renewable hydrogen storage and evaluates its performance against a no-hydrogen configuration and a reactive hydrogen storage controller under identical operating conditions, component parameters, and evaluation metrics. The MPC uses receding horizon optimization with measured hydrogen storage and safety states, together with short-term forecasts of renewable generation, load, electricity price, and hydrogen service demand. Using synthetic but physically motivated time-series profiles, the results show that hydrogen hardware under reactive rule-based operation eliminates renewable curtailment and reduces operating cost; however, it frequently depletes the tank to the reserve boundary and fails to reliably meet subsequent hydrogen demand. In contrast, the sensor-informed MPC achieves full hydrogen accessibility, eliminates curtailment, reduces total operating cost to 71.2% of the no-hydrogen baseline, and lowers cost by 46.3% relative to the rule-based controller. The MPC imports more grid electricity than the rule-based case, showing a clear trade-off between hydrogen service reliability and grid dependence. These results indicate that reliable renewable-hydrogen operation depends not only on storage hardware, but also on predictive supervisory control that coordinates renewable use, hydrogen service, reserve security, and grid interaction.

Safety-Constrained Deep Reinforcement Learning for Source–Load–Storage Coordinated Operation of Green Low-Carbon Data Centers

A safety-constrained deep reinforcement learning framework for source–load–storage coordinated operation of a grid-connected green data center and shows how separating reward learning, cumulative safety pricing, and one-step engineering projection changes low-carbon dispatch within the specified model.

Dynamic Interval Forecasting-Based Demand Response Scheduling for Hydrogen-Battery Microgrids Under Renewable Uncertainty

High penetration of wind and photovoltaic generation introduces significant uncertainty into microgrid operation, particularly when renewable resources are coordinated with hydrogen storage, battery storage, grid interaction, and flexible demand. This paper proposes a demand-response-assisted chance-constrained scheduling framework for a grid-connected wind/PV/hydrogen/battery microgrid. A dynamic interval forecasting model is first developed by integrating a hybrid iTransformer–LSTM–KAN architecture with Monte Carlo Dropout and an adaptive confidence-level mechanism, enabling time-varying uncertainty bounds for wind and PV generation. The resulting prediction intervals are embedded into a chance-constrained optimization model that jointly schedules renewable dispatch, battery charging/discharging, electrolyzer operation, fuel cell generation, grid power exchange, and flexible load shifting. The proposed framework is evaluated on a microgrid consisting of 2 MW wind generation, 3 MW PV generation, 2 MWh battery storage, a 1 MW electrolyzer, a 0.5 MW fuel cell, and 20% flexible load participation. Results show that the proposed strategy reduces total operating cost by 60.1%, renewable curtailment by 65.3%, and grid purchase energy from 18.39 MWh to 10.49 MWh compared with deterministic scheduling. These findings demonstrate that coupling dynamic renewable uncertainty with demand-side flexibility can enhance renewable accommodation, reduce grid dependence, and improve the economic operation of hydrogen battery microgrids.