Application of Action Dependent Heuristic Dynamic Programming to Control an Industrial Waste Incineration Plant

In this paper, we describe our application of a neurocontroller based on Action Dependent Heuristic Dynamic Programming (ADHDP) to optimize the combustion-process for an industrial hazardous waste incineration plant. This ADHDP-controller originally was designed for online learning. That implies, that this controller starts with a randomly initialized policy and improves its performance while interacting with the process. This learning scheme could not be used in our case, since the plant operators would not allow long training periods with inevitable poor performance. We describe, how this problem can be solved by a modified training procedure for the action network. Finally, we present first and promising results for the optimization of action net output in order to improve the waste incineration process with respect to the targets defined by the plant operator.