A neural network model for short-term forecasting of electricity generation by solar power plants

Abstract

Today, energy consumption in the world is growing and it is becoming urgent to solve the problem of replacing traditional energy sources with alternative ones. The solution to this problem is impossible without a preliminary data analysis and further forecasting of energy production by alternative sources. However, the use of alternative energy sources in the conditions of the wholesale electricity and capacity market currently operating on the territory of the Russian Federation is impossible without the use of short-term predictive “day ahead” models. In this article, the authors perform a brief analysis of the existing methods of short-term forecasting which are used when making forecasts for the generation of electricity by solar power plants. Currently, there are already a fairly large number of predictive models built within each of the selected methods of short-term forecasting, and they all differ in their characteristics. Therefore, in order to identify the most promising method of short-term forecasting for further use and development, the authors used a previously developed classification. In the course of the study, a preliminary processing of initial data obtained from the existing solar power plants using spectral analysis was carried out. Further, to build a predictive model, a correlation analysis of the initial data was carried out, which showed the absence of a linear relationship between the components in the retrospective data. Based on the results of the correlation analysis the authors made a decision to select parameters empirically in order to build a predictive model. As a result of the study, a mathematical model based on an artificial neural network was proposed and a learning sample was generated for it. In addition, the architecture of an artificial neural network was determined, the result of which is a short-term forecast of electric power generation in the "day ahead" mode, and calculations were performed to obtain numerical values of the forecast. From the results of the study, it follows that the developed predictive model in the predicted interval has a mean absolute error of about 13.5 MW. However, at some intervals, the peak discrepancies can reach up to 200 MW. The root mean square error of the model is 27.8 MW.