Fault diagnosis of photovoltaic system based on machine learning model fusion

Abstract

With the reduction of non-renewable resources, the energy issues become more and more important and urgent. Distributed photovoltaic power stations not only provides more energy with flexible deployment, but also proposes a possibility of how to fully use the solar energy with a better efficiency. Although a distributed photovoltaic power station has many advantages and great potential, it is not easy to be operated and maintained. In recent years, the applications of machine learning in multiple fields make it possible to do the fault diagnosis of photovoltaic systems more efficiently. This paper designs a full processing system to realize the function of real-time fault diagnosis specially for distributed photovoltaic power stations, which includes the data processing, an online system to give the result of real-time diagnosis, an offline system to retrain models, and a fusion model of several machine learning algorithms. The data processing and performances of different models in this problem are introduced in detail. The fusion model that proposed in this paper acts much better than single models. Hence, the fault diagnosis system could be used in the operation and maintenance of distributed power stations.