Flexibility Forecast and Resource Composition Methodology for Virtual Power Plants

As more distributed generation, of different sizes, is being integrated into the electricity network, energy systems become more decentralized and therefore need to adapt to the changes to balance electricity and simplify trading. The concept of aggregating distributed generation resources has become commercial with the formation of Virtual Power Plants (VPP). VPP is a software-driven solution for aggregating and controlling multiple distributed energy resources to provide flexibility services to the grid operator. We define VPP flexibility as the capability of increasing or decreasing power and modulating available energy stored in the VPP upon request. This paper investigates forecasting of a VPP day-ahead hourly flexibility profile by means of training deep learning models with historical flexibility and weather data to predict day-ahead profiles. Accurate forecasting of the flexibility profile is required in order to avoid penalties for not meeting the contracted capacity and to minimize the risk of failing to provide reserve due to overestimation of production. The decomposition of the VPP flexibility profile in terms of different categories of resources is analysed following a similar deep learning approach. In general, the knowledge of flexibility composition can be used for long-term forecasting, optimization of daily operation, generation and network planning activities.