Structure Optimization of Locally Linear Model Tree Using Extermal Optimization

Locally Linear Model Tree (LOLIMOT) algorithm proposed by Nelles deals with local linear nearo-fuzzy models that is based on divides-and-conquer strategy that a complex modeling problem is divided to a number of smaller and thus simpler sub problems. So the characteristic of such a neuro-fuzzy model depends on division strategy for the original complex problem. For finding the best output the algorithm divides the problem to a number of local linear models (LLMs), then continues with finding the worst LLM and dividing it. LOLIMOT splits the local linear models into two equal halves with an axis-orthogonal decomposition strategy. In this paper a new approach based on extremeal optimization (EO) is used to optimize the structure of LOLIMOT. Simulation results show the effectiveness of the enhanced LOLIMOT to have a higher precision with optimal number of neurons.