Study of Markov decision process-based optimal switching algorithm performance for small power systems

In this paper, the author studies the effect of load flow analysis type (aka. full AC vs. decoupled), horizon (aka. number of steps beyond which discount factor drops to zero) and system stress (aka. magnitude of aggregate load) on the accuracy of optimal power system switching studies implemented via application of dynamic programming to a Markov Decision Process. The objective is to determine whether general guidelines may be established that allow a power system operator to obtain maximum benefit from such studies with minimal computational effort, guidelines that may help future engineers utilize more complex operating techniques. This paper addresses derivation of an innovative solution algorithm, development of appropriate simulation files and test cases, as well as data analysis and discussion of results. For this work, the author performed 240,000 optimal switching studies, in Matlab, for 100 randomly-generated 14-bus power systems derived from the IEEE Standard.