Intelligent Control of Distribution System FACTS Devices Using Deep Learning Models: A Comparative Study

FACTS device created for distribution system is entirely electronic. PI, PID, Series filter, and other controller components are popular for FACTS device design. This work aims to replace FACTS device controller components with deep learning models. To realize proposed FACTS devices, such as the DSTATCOM and DVR, input and output data from distribution system FACTS controllers are collected into a matrix or vector dataset. The developed deep learning-based FACTS device learns controller component behavior using these datasets. Since Deep Belief Nets (DBN) and Convolutional Neural Networks (CNN) learn differently, these models are used to evaluate their performance. The trained deep learning controllers replace the FACTS device in the IEEE 30 bus distribution system and are simulated for verification using PSAT and MATLAB Simulink. The goal is to detect similarities between the proposed and actual devices while putting the FACTS device optimally. The results demonstrated that DBN and CNN-based FACTS devices operate similarly to electronically-based ones. Therefore, deep learning-based DSTATCOM and DVR can intelligently operate the distribution system.