Structurally Optimized Neural Fuzzy Modeling for Model Predictive Control

This paper investigates the local linear model tree (LOLIMOT), a typical neural fuzzy model, in multiple-input-multiple-output predictive control (MPC). In conventional LOLIMOT, structural parameters including centres and variances of its Gaussian kernels are set based on equally dividing input data space. this paper, after initially obtained from space partition, they optimized by gradient descent search, which partitions further adjusted. makes it better for structure to fit statistics, leading improved modelling performance with small size. The MPC proposed structurally LOLIMOT is then implemented verified both numerical diesel engine plants. Validation results show that has significantly controlling than making an attractive solution practice.