An immunity-based optimization algorithm for tuning neuro-fuzzy controller

Immune evolutionary algorithm is proposed based on the evolutionary principle in the immune system. In the algorithm, two new parameters of expansion radius and mutation radius are defined to construct a small neighborhood and a large neighborhood. Then expansion and mutation operations are designed to perform local and global search respectively by using the two neighborhoods, thus, two-level neighborhood search mechanism is realized. The results of multi-modal function optimization show that the algorithm has nice global and local searching performances. The algorithm is employed to design a neuro-fuzzy controller for real-time control of an inverted pendulum. In order to avoid the combinatorial explosion of fuzzy rules due to multivariable inputs, state variable synthesis scheme is suggested to reduce the number of fuzzy rules. Experimental results show that the designed controller can control actual inverted pendulum successfully.