Genetic reinforcement learning in neurofuzzy control systems

Fuzzy controllers are knowledge based and for many real world processes it is possible to design a fuzzy controller which provides bounded regulation using only a heuristic approach. However, in order to achieve satisfactory performance it is always necessary to carry out complicated procedures of fine tuning. In this paper, a fuzzy controller is implemented in a neural structure which then provides for automated tuning using a learning algorithm. Learning is achieved through reinforcements using genetic algorithms. It is also shown that the provision of initialization of the fuzzy controller greatly improves the learning task. The technique is demonstrated on control of a gas turbine jet engine.