Piecewise affine power flow model for distribution network optimization using CIM aided data-driven approach

Abstract

Power flow (PF) analysis serves as the basis for power system operation that is extended to the distribution levels integrated with more distributed generations (DG). Solving optimal power flow (OPF)-resembled problems for practical networks is still challenging due to unavailability of accurate PF equations adaptive to systems variations. To solve this issue, we first set up the PF simulations for complex and realistic power networks based on the common information model (CIM). This automates the recognition of network topologies and parameters in the reduced bus-branch form for system simulations and data generations. The non-convexity of OPF imposes an additional challenge, where solving the problem is still difficult for scenarios with intertemporal couplings considering energy storage. We further employ a multi-layer neural network that provides piecewise affine approximations of the PF equations with desired accuracy. The corresponding OPF-resembled problem is then reformulated as a mixed-integer optimization that facilitates fast and optimal coordination of large-scale DGs for various network operational purposes. We demonstrate the proposed framework using real CIM files from system operators, and compare against typical approaches with improved performance for voltage regulation and loss minimization.