Application of Classical Machine Learning Methods for Forecasting Energy Generation Volumes Using Photovoltaic Power Plants.

This work is dedicated to the application of classical machine learning methods for forecasting energy generation volumes using photovoltaic power plants (PVPPs). The primary focus of this work is on using machine learning algorithms such as SVM and random forest to model and predict energy production based on data collected from PVPP sensors and inverters. Accurate energy generation forecasting is crucial for efficient energy system management, particularly for demand balancing, maintenance planning, and resource optimization. The choice of machine learning algorithms is motivated by their ability to handle large volumes of data and detect complex patterns, which is essential for accurate prediction of energy metrics. The main challenges encountered in the study included data irregularities caused by weather conditions and the technical characteristics of PVPPs, as well as the need to select optimal features for the models. The relevance of the study is driven by the growing role of alternative energy sources in global energy and the need to improve the efficiency and reliability of photovoltaic power plants. The work includes analysis and comparison of the effectiveness of different machine learning algorithms for forecasting tasks based on time series characteristics derived from data collected from PVPPs. One of the study's goals is to identify the optimal features for building forecasting models and their impact on prediction accuracy. The obtained results show that machine learning algorithms, while effective, require careful tuning of parameters and features to achieve high forecasting accuracy. The application of machine learning methods has demonstrated significant improvements in forecasting accuracy compared to traditional approaches, enabling more effective integration of PVPPs into energy systems and ensuring their operational stability. This research underscores the importance of modern data analysis technologies for optimizing the operation of photovoltaic power plants, contributing to their efficient use in the energy sector. It is demonstrated how the application of machine learning methods can significantly improve the accuracy of forecasting the volumes of electricity generation of the FES, which, in turn, contributes to the effective management of energy resources and maintenance planning. For this, data was collected from various power system devices, including sensors and inverters. The data underwent pre-processing, which included cleaning, normalization and selection of key features such as temperature, solar radiation intensity. The models were trained on the training sample and their performance was evaluated on the test sample using metrics such as Accuracy, Precision, Recall, F1-score and ROC curve. The ROC curve showed high performance of the models, with AUC close to 1, indicating high accuracy of predictions.