Evolutionary parameter optimization of a fuzzy controller which is used to control a sewage treatment plant

In order to meet new environmental standards, sewage treatment plants may need to be redesigned or extended. Instead of reconstructing large parts of a sewage treatment plant, which can be very costly, it is in many cases sufficient to install relatively inexpensive equipment, which controls parts of the plant in a new way. Fuzzy controllers are often used for this task. Use of these controllers often leads to an improved water quality. Such fuzzy controllers contain a number of parameters which are determined by a human expert. With this contribution, a dedicated multi-objective evolutionary algorithm is developed to optimize these parameters. The evolutionary algorithm is based on the successful strength pareto evolutionary algorithm 2 (SPEA2). The fuzzy control parameters, which are optimized are continuous parameters. Therefore, an evolution strategy was employed which uses the multi-objective ranking as used by the SPEA2 algorithm. Optimal parameters were first evolved on simulated sewage treatment plants. One set of parameters was also tested on an actual plant. Owing to the enormous computational demands of simulating a sewage treatment plant, it is only possible to work with small population sizes. Nevertheless, it was possible to evolve parameters which were equally well as those found by a human expert indicating that the parameter tuning can be automized.