Structure and parameter optimization of FNNs using multi-objective ACO for control and prediction

Design of a fuzzy neural network (FNN) consists of optimization of network structure and parameters. The objectives are to minimize the network model size with minimum training error at the same time, causing a conflict between the two objectives in the design problem. To address this problem, the multi-objective, rule-coded, advanced, continuous-ant-colony optimization (MO-RACACO) is applied to design FNNs in this paper. The MO-RACACO-designed FNNs are applied to time sequence prediction and nonlinear control problems to verify its performance. Performance of this approach is verified through three simulation examples with comparisons with various multi-objective population-based optimization algorithms and detailed discussions of the results. The results show that the MO-RACACO-based FNN design approach outperforms the multi-objective population-based algorithms used for comparisons in the control and prediction examples.