Optimizing Planning Strategies: A Machine Learning Forecasting Model for Energy Aggregators and Hydropower Producers

Abstract

The global push for higher renewable energy production is driven by concerns about climate change, pollution, and diminishing fossil fuel reserves. Governments, businesses, and communities worldwide prioritize cleaner energy sources like solar, wind, and hydroelectric, over traditional fuels. Technological advancements enhancing efficiency and cost-effectiveness have made renewables more competitive, catalyzing their growing dominance in the energy market. In this context, renewable energy forecasting models are fundamental for both operators of the energy market called energy aggregators, and prosumers for different reasons like planning, decision-making, energy sales optimization, and investment evaluation. Therefore, the present work aimed to develop a machine learning model designed for multi-step hydropower forecasting of plants integrated into Water Distribution Systems (WDSs). The Alcantara 1 Hydroelectric Plant, situated in Italy, was utilized as the case study. This plant generates electricity from the water flow utilized for municipal water supply, which is then sold to the medium voltage network, resulting in substantial remuneration. This innovative approach utilizes previously unused architectures like TCN and N-Beats, to provide multi-step hydropower forecasting for WDS-integrated plants, a special category of systems for which models have not yet been developed. Results indicate TCN as the most accurate model for addressing the proposed task.