Abstract: On the basis of research on pump characteristics and pump model, a modeling method based on RBF neural network with K-means clustering algorithm was proposed. With k-means clustering algorithm the center vector and base width parameters, in which lie in the hidden layer, were optimized by input data sample. The weights between the hidden layer and the output layer were optimized by input-output data sample with least squares method. The neural network models of the pump characteristic and pump comprehensive model were separately trained and checked using the detected data. Its results suggest that reasonably choosing the number of hidden layer node and overlap coefficient, the trained neural network can substitute for the classic polynomial equations of the pump characteristics and the pump comprehensive model, and is with high accuracy.

Key words: centrifugal pump, pump model, Q-H characteristic curve, clustering algorithm, RBF neural network