Research on Photovoltaic Power Prediction Using an LSTM Recurrent Neural Network

The chapter presents the main parameters, characteristics and classification of photovoltaic (PV) power plants. A description and comparative analysis of existing methods and models for forecasting solar irradiation, temperature and photovoltaic power are provided, which allows to identify factors that affect the improvement of the accuracy of software calculation methods. As a result of the research, it was found that the peculiarity of systems with the use of PV systems is the inconsistency of the generated electricity in time, caused by the partial or complete eclipse of the solar battery (SB), which is caused by the accumulation of water vapor, small solid particles in the atmosphere and the daily course of the Sun across the celestial sphere. To ensure the required capacity of energy sources using solar technologies, excess power of the SB is installed. A model of the LSTM network has been built for the insolation forecasting, which has 300 hidden layers and has passed 250 iterations, and for temperature forecasting model, which has 300 hidden layers and has passed 200 iterations. The obtained graphs of the neural network’s learning processes show the degree of its learning. Comparative graphs of measured and predicted insolation were constructed; measured and predicted air temperature, where it is determined that the RMSEins = 56.8616 and RMSEtemp = 0.20283, respectively. Simulations were carried out using the PV power plant model power_PVarray_grid_det and a detailed schedule of electric power for 5 days in advance was obtained. The results obtained can be used for planning, dispatching and improving the safety of the electrical network.