Requirements and Constraints of Forecasting Algorithms Required in Local Flexibility Markets

The increasing use of renewable energy sources, combined with the increase in electricity demand, has highlighted the importance of energy flexibility management in electrical grids. Energy flexibility is the capacity that generators and consumers have to change production and/or consumption to support grid operation, ensuring the stability and efficiency of the grid. Thus, Local Flexibility Markets (LFMs) are market-oriented mechanisms operated at different time horizons that support flexibility provision and trading at the distribution level, where the Distribution System Operators (DSOs) are the flexibility-demanding actors, and prosumers are the flexibility providers. This paper investigates the requirements and constraints of forecasting algorithms required to participate in LFMs. The paper analyses the adequacy of current load forecasting algorithms to fulfill the requirements of LFMs. The work extracts the forecasting requirements for data granularity, forecasting horizon, participants aggregation, and their relevance for market operation; highlighting the implications of data availability at both training and forecasting stages related to the different local market actors (i.e., DSO, aggregator, prosumer) and market operation timing. The analysis evidences the relevance of load aggregation and forecasting horizon in the performance of forecasting algorithms and their impact on the accuracy, depending on the actors and stages during market operation. It evaluates how data volume, forecasting horizon, and participant aggregation affect the performance of forecasting models. Key findings show that aggregating participants and reducing the forecasting horizon considerably improve forecasting accuracy. The accuracy of DSO forecasting is usually better due to the availability and completeness of aggregated data at the system level (i.e., feeder, transformer, substation). Main findings show that increasing training data further than half a year does not keep improving forecasting accuracy, using a next-hour time horizon achieves around 29% better accuracy than a next-day time horizon, aggregating LFM participants can increase forecasting up to 100% depending on the aggregation number. The findings are discussed in the context of LFM operated with current data infrastructures and provide recommendations for improving the integration of forecasting algorithms to enhance flexibility management.