State-space approximate dynamic programming for stochastic unit commitment

It is known that unit commitment problems with un certainties in power demands and the outputs of some generators can be represented as factored Markov decision process models. In this paper we propose a state space approximate dynamic programming algorithm to solve such models. The algorithm features a method to generate representative system configurations (states) and a functional metric to measure the similarity among system configurations. Experimental results show that the algorithm outperforms two deterministic approaches in resulting in both lower risks and operational costs, and that it can solve larger problems than a stochastic approach based on decision space approximate dynamic programming.