Complexity reduction of explicit MPC based on fuzzy reshaped polyhedrons for use in industrial controllers

The explicit model predictive control (EMPC) generates the rules of control defined for a set of polyhedral regions. Online EMPC calculations consist of searching a look-up table to find the appropriate control law according to a particular state. This paper discusses the complexity of online computation and the memory required to store data in an EMPC implementation. Therefore, a new reshaping method is applied to the active regions so that the definition of the polyhedron has regular boundaries. This approach has made some improvements. First, the usable memory will be a lot less for the actual implementation compared to the traditional EMPC approach. Second, the small number of new clusters reduces search time in explicit lookup tables and speeds up overall implementation. To this end, fuzzy clustering is used to introduce a novel method of transforming polyhedrons in the context of fuzzy explicit model predictive (FEMPC) control, followed by a new fuzzy-based piece-wise affine (PWA) explicit formulation for control law calculations. The stability of the proposed method is investigated using the Lyapunov stability criteria. The proposed algorithm has been tested on a nonlinear continuous stirred tank reactor (CSTR) benchmark system and simulation tests show that the proposed approach involves a compromise between storage space requirements and online efficiency.