Modelling soil deposition predictions on solar photovoltaic panels using ANN under Malaysia’s meteorological condition

Abstract

Solar photovoltaic (PV) panels performance is influenced by various external factors such as precipitation, wind angle, ambient temperature, wind speed, transient irradiation, and soil deposition. Soiling accumulation on panels poses a significant challenge to PV power generation. This paper presents the development of an artificial neural network (ANN)-based soil deposition prediction model for PV systems. Conducted at a Malaysian solar farm over three months, the research utilized power output data from the inverter as model output and meteorological data as input variables. The model employed the Levenberg-Marquardt backpropagation method with Tansig and Purline activation functions. Performance assessment via statistical comparison of experimental and simulated results revealed a coefficient of determination (R2) value of 0.68073 for the ANN architecture of 5 input layers, 30 hidden layers, and 1 output layer (5-30-1). Sensitivity analysis highlighted relative humidity and wind direction as the most influential parameters affecting PV soiling rate. The developed ANN model, combined with sensitivity analysis, serves as a robust foundation for enhancing the efficiency of smart sensors in PV module cleaning systems.