An Improved Associative Memory Learning Control System for Industrial Processes with Unknown Dynamics

This paper proposes an improved version of the associative memory learning control system (AMLCS) for industrial processes with almost completely unknown but slowly time-varying dynamics. The main improvements introduced are: 1) Both the predictive process model and the control strategy are in the incremental form, in order to cope with the stepwise disturbances and also to reduce the required memory sizes of the associative memory systems (AMSs) for them; 2) The multi-step ahead subgoal oriented optimization for the current control increment is simplified by introducing a blunting factor.Thus, the original multivariable optimization can be reduced into a singlevariable one; 3) A new addressing mechanism for the AMS is presented so as to reduce the required memory size greatly without the collision problem due to the hash-coding; 4) A local linear extrapolation approach is used in the active learning procedure of the AMS for the predictive process model so as to enlarge its trained region and hence to quicken the convergence of learning without any harmful influence on the control quality. Numerical simulations have shown the feasibility and effectiveness of the new AMLCS proposed.