Online Optimization of Fuzzy Controller for Coke-Oven Combustion Process Based on Dynamic Just-in-Time Learning

To guarantee the control performance of a fuzzy control system for the combustion process in a coke oven, the parameters of the fuzzy controller need to be optimized so that the controller can handle large changes in the operating state of the oven. This paper describes an online optimization method for this purpose. In this method, the distance and angle of the trend of the change are used to select data, and just-in-time learning is used to create a dynamic sample base and to build a radial-basis-function neural-network model of the process. A variable-universe fuzzy logic controller controls the process, and an adaptive differential evolution algorithm optimizes the universe parameters. This enables the controller to adapt to changes in the operating state in a timely fashion. Simulation results demonstrate the effectiveness of the method.