Deep Learning based Real-time Approach for Robust Optimal Power Flow in Three-Phase Unbalanced Distribution Systems

The challenges of growing uncertainties associated with renewable energy generation and flexible loads become pressing nowadays. To effectively address them and enhance intelligent capabilities for various levels of dynamics, power distribution systems are undergoing an development with advanced real-time monitoring and optimization functions. In this paper, we propose a learning-based approach aimed at solving the optimal power flow (OPF) problem in three-phase unbalanced distribution systems. A deep neural network model is developed and applied to learn the mapping relationship between system loading and optimal power generation settings. Then the model is trained by adding linear constraints to find a near-optimal and feasible solution. This method can provide a reference for the power settings of distributed generators (DGs) and phase-changing soft open point (PC-SOP). Compared with traditional methods, this method has a faster solution speed, making it possible to realize near real-time optimization.