Daily prediction method of dust accumulation on photovoltaic (PV) panels using echo state network with delay output

Abstract

Dust accumulation over time can be one of the main causes of uncertainty in the output of photovoltaic (PV) systems. In order to better understand these losses, this paper established a daily dust accumulation prediction model for PV panels based on the delay output echo state network (DESN). A pigeon-inspired optimization (PIO) algorithm with adaptive Cauchy (AC) mutation strategy was proposed, which can optimize the reservoir parameters (such as leakage rate, spectral radius, and input scaling) of DESN, shorten the solution time, and improve search speed. Based on typical meteorological and air quality data, as well as daily accumulated dust weight recorded from the experimental platform, model training and testing were carried out. According to the Pearson correlation coefficient, the relationship between the environment parameters (humidity, wind speed, wind direction, PM2.5, PM10, and rainfall) and the dust accumulation was obtained. The results show that the prediction accuracy of AC-PIO-DESN is better than other methods for meteorological and air quality data. The mean absolute percentage error (MAPE) for humidity, irradiance, PM2.5 and PM10 were 4.0743%, 4.4958%, 10.6231% and 12.8402%, respectively. In addition, the proposed daily dust prediction model has good robustness for 10-day samples, with an average relative error ranging from 0.65% to 54%. This method can provide data support for grid scheduling and PV panel cleaning strategy of PV power plants.