Optimal Productivity in Solar Power Plants Based on Machine Learning and Engineering Management

Abstract

The complexity of solar power plants is constantly increasing. This sophistication includes the increasing number of solar panels installed and the technologies that are employed in the energy conversion systems. The new solar plants require advanced methods to ensure the availability of all the panels. This paper proposes a recurrent convolutional neural network algorithm for detecting failures, reducing the costs and the time of the inspections. The method is aimed to analyze the data provided by an unmanned aerial vehicle fitted with a thermographic camera. This system provides thermographic data and telemetry. A region-based recurrent convolutional neural network is trained by a previously created dataset. Once the neural network is trained, it is used as a hot spot detector. This detector will have employed the telemetry in order to identify the real panel that can be affected.