Data-driven Optimal Dynamic Dispatch for Hydro-PV-PHS Integrated Power Systems Using Deep Reinforcement Learning Approach

To utilize electricity in a clean and integrated manner, a zero-carbon hydro-photovoltaic (PV)-pumped hydro storage (PHS) integrated power system is studied, considering the uncertainties of PV and load demand. It is a challenge for operators to develop a dynamic dispatch mechanism for such a system, and traditional dispatch methods are difficult to adapt to random changes in the actual environment. Therefore, this study proposes a real-time dynamic dispatch strategy considering economic operation and complementary regulatory ability. First, the dynamic dispatch of a hydro-PV-PHS integrated power system is presented as a multi-objective optimization problem and the weight factor between different goals is effectively calculated using information entropy. Afterwards, the dispatch model is converted into the Markov decision process, where the dynamic dispatch decision is formulated as a reinforcement learning framework. Then, a deep deterministic policy gradient (DDPG) is deployed towards the online decision for dispatch in continuous action spaces. Finally, a case study is applied to evaluate the performance of the proposed method based on a real hydro-PV-PHS integrated power system in China. Simulations show that the system agent reduces the power volatility of supply by 26.7% after hydropower regulating and further relieves power fluctuation at the point of common coupling (PCC) to the upper-level grid by 3.28% after PHS participation. The comparison results verify the effectiveness of the proposed method.