Optimal Switch Reduction in Distribution Networks: A Reinforcement Learning Approach for Dynamic Reconfiguration

The continuous fluctuations in electricity demand across different locations in the distribution network led to variations in power flow in the lines and, consequently, changes in power losses and voltage drop. With technological advancements and restructuring of the electricity industry, these changes have become more serious. So, maintaining a single network configuration at all hours may not be optimal. To address this, researchers have proposed and implemented dynamic reconfiguration of distribution networks as an effective solution. This study investigates the dynamic reconfiguration of radial distribution systems, using reinforcement learning to minimize the number of switches. The primary objectives of dynamic reconfiguration are to minimize power losses and enhance bus voltage profiles. A case study is conducted on the IEEE 33-bus standard network. Initially, it is assumed that remotely controllable switches are installed on all lines, allowing full participation in the reconfiguration process. Subsequently, the proposed algorithm for reducing the number of switches identifies critical switches that should be eliminated at each stage, followed by another round of dynamic reconfiguration. For each stage of switch reduction, an economic analysis over 20 years is performed, considering installation and annual maintenance costs alongside power loss costs. Finally, a sensitivity analysis determines the optimal number of switches based on switch costs and electricity tariff rates. The results indicate that lower electricity tariffs and higher switch costs improve the effectiveness of the proposed switch reduction method.