Pattern Recognition: an Alternative to Dynamics Description

For a class of complex production process systems, difficult even impossible to construct exact model, we give a pattern recognition method to describe the system dynamics. Different from traditional pattern-based control methods, we use the categorical characterization of the temporal trends of working condition pattern to capture the system dynamics. First, the actual run status data is collected and a statistical space mapping clustering algorithm is given to partition these data into some pattern classes. Then a variable we called pattern class variable is defined to describe the variation law of pattern class over time. A new Petri nets is constructed as the prediction model based on pattern class variable rather than state variable or output variable. Simulations are given to show that the proposed method might provide the satisfied results for the practical applications without having the exact mathematical models.