A function approximation approach to Reinforcement Learning for solving unit commitment problem with Photo voltaic sources

Abstract

Day ahead scheduling of power generation resources has turned out to be complex and challenging optimization problem with the integration of Renewable energy sources to the power grid. Photovoltaic sources are the most significant and abundant renewable energy source being integrated to the power system. Due to the stochastic nature of the solar irradiance and the associated power, scheduling of the various generating sources including solar power turns to be a stochastic optimization problem. Reinforcement Learning is an optimization method which can accommodate the uncertain nature of the problem environment very effectively. When the solution space becomes very large, simple reinforcement Learning solution through look up table approach becomes more complex and time consuming. In this paper we propose a Neural Network based Reinforcement learning method for the optimal commitment in the presence of Photo voltaic sources. The algorithm is found to be effective for day ahead solution in a smart grid with stochastic renewable energy sources.