Real-time autonomous dynamic reconfiguration based on deep learning algorithm for distribution network

Abstract

In this work, a novel real-time autonomous dynamic reconfiguration (ADR) method is proposed to reduce the cost of power loss and switch action of distribution network based on the deep learning (DL) algorithm. The proposed ADR method can be decision-making from the historical control dataset and the real-time system state. To obtain the historical control dataset efficiently, the heuristic rule is first introduced to improve the mixed integer convex programming method for making reconfiguration strategies to support grid operate in cost effective manner. To learn the reconfiguration strategy from the historical dataset, a long-short term memory network (LSTM) is presented to provide the effective training on the historical control dataset, and a switch action function considering real-time differences of operation cost is formulated, which can be combined with the trained LSTM model to achieve ADR. Case studies on the IEEE 33-bus system and a realistic Taiwan power company (TPC) 84-bus system verify that the ADR can obtain an ideal reconfiguration solution in the order of milliseconds and has high robustness.