Machine Learning Based Prediction of Solar Power Plant Performance Under the Impact of Natural Dust Accumulation

Abstract

In this paper, machine learning based prediction of solar PV power plant performance has been done in MATLAB/Simulink environment. The accuracy of the predicted data is analyzed for three different sets of field input datasets such as; local tilt angle, vertically, and horizontally positioned low-iron glass sheets used as the surface of solar panels. For studying the impact of dust accumulation optically, a set of three low iron glass samples as a surface of commercial solar panels are set up at the roof top beside the solar panels. The data from the three orientations provide with the information to realize the impact of natural dust on solar PV plant optical performance in terms of transmittance and thereby further estimation of power output. Long Short-Term Memory (LSTM) and Artificial Neural Network (ANN) have been used to predict the optical performance of solar PV plant. Both the models show the highest accuracy in predicting data for the horizontally placed panels. As a validation, the ANN model offers an accuracy of over 99%, whereas the LSTM model provides with an accuracy of over 92%. These prediction study will help in predicting the performance of the location specific solar power plants.