Zastosowanie metod uczenia maszynowego do predykcji generowanej energii w małym systemie PV

Abstract

The article presents data analysis for predicting energy production in photovoltaic (PV) power plant systems. The purpose of long-term forecasts is to determine the effectiveness of preventive actions and manage the power system effectively. Climate variables affecting the production of electricity in photovoltaic systems were analyzed. Forecasting methods using machine learning techniques such as Multi-Layer Perceptron (MLP) neural networks and Support Vector Machine (SVM) were compared. In addition, metrics were selected to determine the quality of forecasts. Determining the quality of forecasts was based on the actual varying conditions, not on the weather forecast data. The way of data preparation to create forecasting models were presented and the models with the best metrics were selected. For this purpose, the Scikit-learn library was used to create scripts in Python. The best results were obtained for regression models: MLPRegressor, CatBoostRegressor and Support Vector Regression. Actual measurement data from a system of optimally-positioned panels with a power of 3.0 kWp were used. For the MLPRegressor model, the highest coefficient of determination 0.605 was achieved with the smallest root-mean-square error of 1.79 KWh.