Search direction optimization of power flow analysis based on physics-informed deep learning
Baoliang Li, Qiuwei Wu, Yongji Cao, Changgang Li
Abstract
• A search direction optimization scheme based on PIDL is proposed to improve power flow analysis performance. • The higher-order information of Taylor series expansion is employed to adjust the search direction of power flow. • The DBN is used to establish a nonlinear mapping between power flow equations and optimized search direction. • The multi-task learning is utilized to embed the power system physical information into DBN. Power flow analysis is crucial for obtaining power system operation states and optimizing control measures. The increasing integration of renewable energy sources has resulted in a more complex power system, posing challenges to the computational efficiency and convergence of conventional power analysis methods. Based on the physics-informed deep learning, this paper proposes an optimization scheme for the search direction to improve the performance of power flow analysis. The higher-order information originating from the Taylor series expansion of the power flow equation is utilized to optimize the search direction. The deep belief network is used to establish a nonlinear mapping between the power flow equations and the optimized search direction. Additionally, the physical information of the power system is encoded into the deep learning model to meet the real physical constraints. Case study results show that the proposed scheme contributes to improve the computational efficiency and convergence in power analysis, and is feasible for the scenarios of ill-conditioned power flow.