Development of an IoT Based Photovoltaic Monitoring System Using Hybrid Modeling

This research outlines the design, implementation, and testing of a low-cost Internet of Things (IoT) based photovoltaic (PV) monitoring system which hybridizes physical modeling and machine learning. The physical models solve for preliminary expectations for PV power generation, while the machine learning model is applied to account for additional system inefficiencies. The proposed monitoring system consists of a Raspberry Pi as the central computing unit and several sensors to collect environmental data. In this research, a hardware system was created, including a mock PV system and the proposed monitoring system. A software system was developed as well, to implement the hybrid modeling of the expected power and compare it with the actual generated power to detect anomalies. The proposed monitoring system can accurately compute the expected power generation of PV systems and effectively detect anomalies which cause power losses. Results from the system find that the monitoring system with hybrid modeling has good performance, as the residual difference between the expected and actual measured power is within 2%, regardless of weather conditions. The average residual difference is 0.74% and the average Mean Absolute Error between the expected and actual measured power is 0.31 for the testing period, which is more accurate than either physical or machine learning modeling alone.