Comparison of widely-used models for multifactoral short-term photovoltaic generation forecast

Abstract

The paper presents a comparative analysis of three common methods of forecasting time series for short-term forecasting of the generation of photovoltaic power plants with different horizons. The models were built using real data of the photovoltaic power plant and the local meteorological station. Meteorological data include solar irradiation, ambient temperature, humidity, wind speed and cloud cover. The results of the forecast of gradient boosting, elastic regression and multilayer perceptron for horizons 1 and 24 hours were compared. Sensitivity to input factors was investigated using SHAP value.