Forecasting of Electricity Generation by Solar Panels Using Neural Networks with Incomplete Initial Data

Abstract

An increase in the number of alternative energy sources, in particular solar power stations in the generation structure of the modern energy systems, makes it difficult to control such energy systems due to the inconsistency in the amount of electricity production that depends on external factors. It also leads to an increase in risks when concluding electricity supply agreements in a modern energy market. This problem may be solved by forecasting electricity by modern means, such as neural networks. In this context, due to the flexibility and nonlinearity of neural networks there is no need for source data that directly affects the generation value, and there is a possibility of using indirect data available to most alternative energy producers. This has resulted in the creation of the solution based on neural networks that allows forecasting energy production using solar panels with high accuracy on the basis of conventional meteorological data.