A MACHINE LEARNING APPROACH FOR GENERATION SCHEDULING IN ELECTRICITY MARKETS

The problem formulation and the necessary solution tools which have been employed for traditional power system optimisation are unsuitable for modern systems. The gaming aspect in modern power system operation suggests Machine Learning based tools for decision making problems. Machine learning is the subfield of Artificial Intelligence concerned with the design of automatic procedures which are able to learn from available data or experience. This paper presents a Reinforcement Learning (RL) approach for the day ahead scheduling of generators in the deregulated power market where the objective is to maximize the profit of generation companies (GENCOs). RL employs the framework of a Markov Decision Process (MDP) for decision making and is capable of producing good feasible solutions, even if complete information of the problem is unavailable. The decision support tool with appropriate size reduction strategies is suitable for deterministic as well as stochastic scenario. The performance of the proposed approach is tested using a standard 10 unit system in the literature. Analysis of the results shows that an accurate and efficient solution strategy for the problem can be easily developed using the concepts of RL. The development of the solution framework and its implementation is simple, flexible and will be very helpful for the GENCOs for quick decision making in present day competitive scenario