Harmony Search Algorithm for Fuzzy Cerebellar Model Articulation Controller Networks Optimization

The general learning algorithm of Fuzzy Cerebellar Model Articulation Controller networks usually applies the gradient-descent type methods. However, these gradient-descent methods cause the high possibility to converging into local minima. To cope with the local minimum problem, we instead propose to apply harmony search algorithm to achieve better performances. The harmony search algorithm optimizes not only Fuzzy Cerebellar Model Articulation Controller network’s weight values, but also optimizes network receptive field’s centre positions and width parameters. To find the best optimized network, the weight values, centre positions, and width parameters are transformed to three data strings. In addition, an improved version of harmony search algorithm is used to search the best combination within data domains. The network’s performances are verified by approximating four non-linear formulae. The experimental results show that the improve harmony search algorithm performs very fast convergence speed.