A Review on Machine Learning Models in Forecasting of Virtual Power Plant Uncertainties

The penetration rates of renewable sources and energy storage systems in the energy market have risen considerably due to environmental and economic concerns. In addition, new types of loads such as electric vehicle charging are added to the grid recently. Inherent uncertainty of renewable generation and new type of loads make the power grid more complex and difficult to manage from economic and technical aspects. Virtual power plant (VPP) is a key concept of future smart grid integrating a variety of power sources, controllable loads, and storage devices. VPP environment aims to enhance the stability of the grid and maximize the revenue. Achieving these objectives mostly depends on the precise forecasting of three major uncertainties; renewable generation, load demand and electricity price. On the other side, machine learning (ML) models are quite efficient for complex uncertainties with large scale dataset compared to traditional approaches. In this paper, mostly employed ML models for forecasting VPP uncertainties are analyzed. Firstly, VPP components and operation of the system are explained. Then, preprocessing techniques, ML methods and performance evaluation criteria for forecasting approaches are presented. Contributions and limitations of recent works are critically discussed and separately tabulated. Finally, several future research opportunities are released at the conclusion of this paper.