Abstract
In the modern world, effective management of electric grids is impossible without achieving an energy balance between electricity production and consumption. The stability of power grids directly depends on the ability of network operators to accurately predict energy production, especially when it comes to photovoltaic installations. These energy sources, which are actively integrated into electrical networks, are influenced by many factors, including changeable weather conditions, which makes the task of forecasting especially difficult. As part of this study, an attempt was made to improve the accuracy of forecasting the output power of a 20 kW photovoltaic system by collecting and analyzing data on weather conditions and energy production for a typical year covering all four seasons. Special attention was paid to the evaluation of two machine learning methods: long-term memory (LSTM) and multilayer perceptron (MLP), which were chosen because of their potential in processing and analyzing time series of data. The results of the study demonstrate the high efficiency of both methods in predicting the output power of photovoltaic systems, which opens up new opportunities for improving the management of electric networks. The use of LSTM and MLP makes it possible not only to accurately predict energy production, but also helps to optimize network utilization, minimize losses and increase the overall efficiency of the electric power system. It is important to note that the success of these models depends on the quality and volume of the data collected, as well as on the thorough preparation of the data for analysis. The study highlights the importance of integrating advanced machine learning technologies into the field of energy system management, offering practical recommendations for energy companies and power grid operators.