A hybrid physical-data approach for solving dynamic optimal power flow considering uncertainties and different topology configurations

The security and economy of power system operation or planning are threatened by various uncertain factors, including the unpredictability of renewable energy, the variation of topology configurations, and significant load fluctuation. It is a challenge to solve dynamic optimal power flow efficiently for the operation and planning of power system considering uncertainties on both Renewable generation and Load, as well as different Topology configurations (RLT-DOPF). For fast solving the RLT-DOPF problem, we propose a hybrid physical-data approach using a physics-driven method — two-stage robust optimization approach based on an improved uncertainty set, and a data-driven method combined the Graph convolutional network with Long short term memory named GL-DOPF. The two-stage robust optimization approach can supply robust solutions for the RLT-DOPF problem under various topology configurations. Then the GL-DOPF model is trained using a training dataset that includes robust solutions. The solutions for RTL-DOPF are available with the trained GL-DOPF model and an AC power flow calculation fast. Simulations on IEEE 30/300-bus systems show that the speedup of the proposed hybrid physical-data approach is x25/29 as compared to a robust optimization method for RLT-DOPF under almost the same accuracy and the solutions for RTL-DOPF are feasible and robust. The training dataset for GL-DOPF generated from the proposed two-stage robust optimization model is less conservative than the original RO models and can cover the prediction errors of uncertainties. The effectiveness and robustness of the proposed hybrid physical-data approach for the RLT-DOPF problem clearly outperform