Data-Driven Multiperiod Optimal Power Flow for Power System Scheduling Considering Renewable Energy Integration

Abstract

Integrating renewable energy sources (RESs) into electric power systems is environmentally beneficial; however, it also introduces operational challenges. The multiperiod optimal power flow (MP-OPF) problem becomes increasingly complex owing to the variability and uncertainty in RESs power output and load demand. To ensure system stability and reliability, other conventional generators must quickly adapt their output to counter fluctuations in RESs and mitigate network congestion. In this study, we propose a novel approach utilizing a long short-term memory recurrent neural network (LSTM-RNN) to address the MP-OPF problem. The LSTM-RNN’s capability to handle time-series data enables fast and accurate predictions. By formulating the MP-OPF as a sequence-to-sequence learning problem, we train the LSTM-RNN to map input data (load demand, renewable generation) to output data (generator output, RESs injection) to meet power network constraints while simultaneously achieving economic power dispatch to generators. Additionally, we perform the post-processing on the output of LSTM-RNN results to obtain a feasible power generation schedule at each time step and to analyze network congestion and RESs curtailment. Furthermore, the proposed approach is demonstrated on the IEEE-39 bus system with time-series data, achieving highly accurate OPF solutions. Computation time is approximately 160 times faster than conventional solver.

Keywords

• Deep learning
• generation scheduling
• LSTM-RNN
• optimal power flow
• renewable energy integration