Deep integration between machine learning approaches and renewable energy optimization

Over the decades, numerous optimization models have been developed to maximize the use of limited resources in power systems. With the rapid advancement of machine learning (ML) technologies, integrating ML with these optimization models has emerged as a promising trend. Unlike traditional optimization methods that repeatedly solve similar problems without learning from past experiences, ML leverages historical data and previous decisions to significantly enhance optimization performance. This project aims to develop knowledge enabling deep integration between ML approaches and the short-term hydro-dominant unit commitment (UC) problem in a deregulated power system. The UC problem is the foundation for production scheduling. The goal is to optimize the operation of generators, primarily hydro-turbine generators, over a finite planning horizon under various operational and market conditions. New renewable energy sources like solar and wind power will also be considered when necessary. The ML-based optimization model developed in this project will significantly improve the computational efficiency of generating daily production schedules for Norwegian hydropower producers, making them more efficient and adaptable to market conditions. Successful implementation of ML in the optimization model will address current challenges faced by industrial partners, such as reducing the time resolution from one hour to 15 minutes, extending the model’s application from an operational tool to an investment analysis tool over longer time horizons (weeks to years), effectively incorporating uncertainties in inflow and prices, and providing bidding strategies for participating in multiple markets. This project exemplifies the synergy between academia and industry, both domestically and internationally. Comprehensive case studies conducted under different market structures will lead to more robust solutions and broader dissemination of knowledge and experience.

During the past half-century, numerous optimization models have been developed to optimize the utilization of the limited resources in the power system. With the booming development in machine learning (ML) technologies, deep integration between ML approaches and optimization models has become the most promising technical trend. The conventional optimization methods repeatedly solve similar problems without accumulating any experience. In contrast, ML efficiently gains experience from historical data and previous decisions. Numerical studies have shown that ML can significantly boost optimization performance. The renewable energy optimization addressed in this project is the unit commitment (UC) problem, constituting the foundation for production scheduling. It aims to optimize the operation of generators over a finite planning horizon under operational and market conditions. The generators are mostly hydro-turbine generators. New renewable energy, such as solar or wind power, will also be considered when necessary. The ambitious project aims to realize an ML-based optimization model that enhances the computational efficiency of generating daily production schedules for Norwegian hydropower producers, accurately represents their physical watercourse, and supports their bidding strategies in energy and capacity markets. The successful implementation of this project will address current challenges faced by industrial partners, such as reducing the time resolution from one hour to 15 minutes, extending the model’s application from an operational tool to an investment analysis tool over a longer time horizon (weeks to months or even years), and effectively incorporating uncertainties in inflow and prices. This project represents a domestic and international complementarity of both academia and industry. Comprehensive case studies conducted under different market structures will lead to more robust solutions and broader dissemination of knowledge and experience.