Design of a CMAC-Based PID Controller Using Operating Data

In industrial processes, PID control strategy is still applied in a lot of plants. However, real process systems are nonlinear, thus it is difficult to obtain the desired control performance using fixed PID parameters. Cerebellar model articulation controller (CMAC) is attractive as an artificial neural network in designing control systems for nonlinear systems. The learning cost is drastically reduced when compared with other multi-layered neural networks. On the other hand, theories which directly calculate control parameters without system parameters represented by Virtual Reference Feedback Tuning (VRFT) or Fictitious Reference Iterative Tuning (FRIT) have received much attention in the last few years. These methods can calculate control parameters using closed-loop data and are expected to reduce time and economic costs. In this paper, an offline-learning scheme of CMAC is newly proposed. According to the proposed scheme, CMAC is able to learn PID parameters by using a set of closed-loop data. The effectiveness of the proposed method is evaluated by a numerical example.