Predicting Floating Photovoltaic Platform Multi-Variable Dynamic Responses With Neural Network Learning Methods

Abstract As global regulators have put an emphasis on turning towards renewable methods of energy production, Floating Photovoltaics has emerged as a growing renewable energy source, first establishing itself with inland installations before expanding into more challenging offshore conditions. Floating Photovoltaics platforms experience significant environmental loading in offshore conditions requiring a rigorous design process and increased monitoring requirements once deployed. On-site monitoring of the Floating Photovoltaic platforms (FPVs) can be difficult for several reasons including installation challenges, environmental conditions, difficulties in access, cost-effectiveness and performance of sensors in such harsh conditions. Alternative methods of monitoring the structure with minimal onboard instrumentation would prove beneficial. Here, Machine Learning (ML), i.e., Neural Networks (NN), when adequately trained, can be effective in predicting the dynamic responses of an FPV platform using minimal sensor input. This paper aims to demonstrate how NN can predict dynamic responses of a Froude-scaled FPV platform in a wave flume by training it on wave profile, mooring line tension and platform motion data. Theis study assesses the performance of a NN with the goal of providing insight on its applications in aiding wave flume tests for photovoltaic platforms. The results of this study indicate that a NN trained on a limited experimental dataset can predict the multi-variable responses of an FPV platform across a range of wave conditions, with certain conditions performing better than others. The neural network performs better in predicting larger wave amplitude conditions, with the heave and pitch predictions performing better than the surge predictions.