Intelligent prediction Modeling using modified locally weighted regression

The local weighted learning method seems a promising solution to the aforementioned engineering project. This paper has researches a modified locally weighted regression method combing the regression coefficients and bandwidth optimization to learn nonlinear system model for prediction. By means of gradient descent method and weighted distance algorithm, the regression coefficients and the bandwidth are optimized and the performance of the locally weighted regression is modified. The method is useful and applicable for modeling some rigorous nonlinear systems globally, but having linear local characters, which is relatively common in motor manufacturing industrial production process. The implementation of the modified method proves to be effective and superior to the RBF neural network. This proves the implementation of the proposed nonlinear prediction model to be effective and practicable for its industrial application.