Operational Planning of Thermal Generators with Factored Markov Decision Process Models

We describe a method for creating conditional plans for controllable thermal power generators operating together with uncontrollable renewable power generators, under significant uncertainty in demand and output. The resulting stochastic sequential decision problem has mixed discrete and continuous state variables and dynamics, and we propose a discretization method for the continuous part of the model that unifies all variables into a large discrete Markov decision process model. Although this model is way too large to be solved directly, its state transition probabilities can be factored efficiently, and a reduction of all continuous variables to one net demand variable makes it tractable by dynamic programming over a suitably constructed AND/OR tree. The proposed algorithm outperformed existing non-stochastic solvers on several problem instances, resulting in both lower risks and operational costs.