Real-time Solar Array Data Acquisition and Fault Detection using Neural Networks

Abstract

Continuous real-time solar system monitoring for fault detection and classification can improve solar panel efficiency and overall output. In this study, we developed and implemented a real-time PV fault detection system based on machine learning. The system was implemented on an 18kW testbed facility which consists of 104 solar panels located at the ASU Research Park. Each solar panel is connected to a smart monitoring device (SMD) which obtains real-time voltage and current measurements. SMDs are attached to each panel and transmit all the acquired data to a server that is connected to the internet. We implement fault detection using real-time measurements and various neural network architectures. We train and test both fully connected and dropout neural networks with different dropout regularization. We use both a real-time dataset and a synthetic dataset and present comparative results. We train and classify for the following conditions: soiled panels, shaded and degraded panels, and standard test conditions.