Intelligent Fault Detection Based on Reinforcement Learning Technique on Distribution Networks

The incorporation of distributed energy resources in the distribution networks changes the fault current level and makes the fault detection be more complex. There are several challenges brought by these heterogenous energy systems including power quality, voltage stability, reliability and protection. In this paper, a fault detection based on reinforcement learning approach is proposed. The heart of this approach is a Q learning approach which uses a non-adaptive multi-agent reinforcement learning algorithm to detect and identify nonlinear system faults, and the algorithm learns the policy by telling an agent what actions to take under what circumstances. Moreover, the Discrete Wavelet Transform (DWT) is utilized to extract coefficient values from the captured one-fourth cycle of the three-phase current signal during fault which occurs during the transient stage. The simulations and signal analysis for different faults are used to validate the proposed fault detection method in MATLAB environment. The simulation results show that different types of faults such as CA, AB, ABC and ABCG can be detected and the best correlation coefficient achieved is 0.87851.