Abstract:

With the expansion of industrial production scale and the increasing complexity of production process,demands  for process simulation are increasing. In the paper, we propose an improved clustering analysisbased neural network ensemble method.Firstly,according to the density distribution of data,an improvement in Kmeans method is made to overcome the disadvantage in the selection of the original central point，and then the samples are classified and the differences among the samples are enlarged.Secondly,aiming to the different samples, the general regression neural network (GRNN) with fast learning ability is used to construct and train individual neural networks.Thirdly,a compensation network is constructed for all the samples by GRNN so as to eliminate the output errors due to false selection.Finally,the obtained clustering center is utilized to make numerical analysis for the input samples and to select the individual neural network.The output of the selected individual neural network is compared with the output of the compensation network,and the neural network ensemble is realized.The verification on artificial data Sinc illustrates that the model accuracy is enhanced by the proposed method.Thus,it provides a new way for increasing the accuracy of process simulation.

Key words: K-means;clustering analysis;general regression neural network;neural network ensemble;process simulation