Symbolic Reduction of Guards in Supervisory Control Using Genetic Algorithms

Report, 2012

In the supervisory control theory, a supervisor is generated based on given plant and specification models. The supervisor restricts the plant in order to fulfill the specifications. A problem that is typically encountered in industrial applications is that the resulting supervisor is not easily comprehensible for the users. To tackle this problem, we previously introduced an efficient method to characterize a supervisor by logic conditions, referred to as guards, generated from the models. The guards express under which conditions an event is allowed to occur to fulfill the specifications. By exploiting the structure of the given models, some techniques were introduced to simplify the guards and make them tractable. In order to be able to handle complex systems efficiently, the models are symbolically represented by binary decision diagrams and thus all computations are performed symbolically as well. As a consequence, the size of the guards becomes very sensitive to the variable ordering in the BDDs. In this paper, by using genetic algorithms, we aim to find the optimal variable ordering for the BDDs that yields the most compact guard. The approach has been implemented in a supervisory control tool and applied to an academic example example.