Research of software solutions for forecasting electricity generation and consumption in Ukraine that are based on machine learning methods

• Дата: 2023
• Автори: Sinitsyn, I.P., Shevchenko, V.L., Doroshenko, А.Yu., Yatsenko, O.A.
• Формат: Стаття
• Мова: Ukrainian
• Опубліковано: Інститут програмних систем НАН України 2023
• Теми: electricity; algorithm; machine learning; forecasting; regression; bilateral contracts market; day-ahead market; intraday market; balancing market; UDC 658.26; 004.421; 004.89
• Онлайн доступ: https://pp.isofts.kiev.ua/index.php/ojs1/article/view/586
• Назва журналу: Problems in programming

The problem of security in energy sector is an important aspect for Ukraine. The purpose of monitoring in this area is to optimize the flow of electricity between market participants, between European partners and Ukraine. It is critically important to maintain a balance between producers and consumers of energy. Both over and undersupply of energy represent risks to infrastructure. The previously available wholesale electricity market model with single buyer has been replaced by a model based on bilateral, day-ahead and intraday markets, as well as balancing and ancillary services markets. Now the participants can freely trade electricity and energy companies can provide services that provide stability of the energy system and supply electricity to the final consumer. The demand forecasting in electricity markets is one of the components that must be implemented for successful business operations and optimization of business processes. Based on the model of the Institute of problems of modeling electricity engineering of NANU, the paper sets out the task of developing a software system for forecasting threats in the energy sector of Ukraine using machine learning methods. Experiments were conducted on the application of regression methods to restore a column with data from bilateral contracts for the task of forecasting electricity generation and consumption. The results of the application of machine learning algorithms on peacetime data demonstrated that it is possible to predict market volumes and tariff plans one hour in advance with a good accuracy which allows to go beyond one-day planning in the future.Prombles in programming 2023; 3: 99-108