Optimal Control for Complex Systems under Multi-granularity Based on Data Mining

This paper proposes a control scheme for a kind of complex systems whose dynamics follow statistical law. First, inspired by data mining technology, the samples are clustered into several classes reflecting the working pattern by modified ISODATA method. Second, based on the clustering result, the cell state space is constructed within global and local target cells respectively by Cartesian product, and the cell mapping further is created via the analysis of pattern time series. Third, by the reverse searching algorithm, global optimal paths to the target cell are found and the information provided by the searching paths constitute the optimal controller at a coarse-grained level so as to bring the system into the target cell quickly, once enter into the target cell, another controller which consists of the subcell mapping with the optimal cost increment in refined granularity level is applied in order to reduce steady-state error. Finally, a simulation in the dynamic trajectory of a sintering process is given to demonstrate the feasibility of the proposed approach.