Forecasting of virtual power plant generating and energy arbitrage economics in the electricity market using machine learning approach.

Over time, the importance of virtual power plants (VPP) has markedly risen to seamlessly incorporate the sporadic nature of renewable energy sources into the existing smart grid framework. Simultaneously, there is a growing need for advanced forecasting methods to bolster the grid's stability, flexibility, and dispatchability. This paper presents a dual-pronged, innovative approach to maximize income in the day-ahead power market through VPP. On one front, forecasting VPP generation units, including solar photovoltaic, wind power, and combined heat and power, employs a novel Adam Optimizer Long-Short-Term-Memory (AOLSTM) machine learning technique. Conversely, estimating the revenue's superior frontier is accomplished by integrating energy storage and Monte-Carlo optimization. The proposed method effectively synergizes the concepts of VPP, energy storage, and AOLSTM to yield more substantial income in the day-ahead electricity market. Notably, the introduced AOLSTM approach demonstrates minimal error metrics compared to conventional methods such as persistence, Gradient Boost, and Random Forest.