Multi-objective energy dispatch with deep reinforcement learning for wind–solar–thermal–storage hybrid systems

Artikel i vetenskaplig tidskrift, 2025

With the intensification of environmental pollution and energy shortage, wind–solar–thermal–storage hybrid systems have been widely considered in the advancement of sustainable energy sources. To reduce the fuel cost and carbon emissions while tracking the demanded load power, this paper proposes a novel energy dispatch strategy based on deep reinforcement learning to achieve the multi-objective task for wind–solar–thermal–storage hybrid systems. Firstly, considering the huge amount of annual load data, a characteristic day acquisition model is constructed to reduce the computational burden. To deal with the uncertainties of wind and solar data, a kernel density estimation method is proposed to obtain the seasonal power generation expectation. Then, a deep Reinforcement Learning method based on Deep Deterministic Policy Gradient is developed to achieve the multi-objective energy dispatch strategy by considering the carbon trading and fuel cost. Finally, simulation results verify that the designed method can effectively achieve multi-objective task and have better robustness performance compared with other methods.