Abstract: Gray matrix-extreme learning machine (GM-ELM) was proposed to solve the fault diagnosis of sucker rod pumping wells. Firstly， the gray matrix method was applied to extract the fault features of sucker rod pumping wells. Secondly， the mathematical method was applied to establish eigenvectors of gray matrix， and the eigenvectors were used as the input value of the fault diagnosis model. Finally， the GM-ELM model was established to diagnose the fault of sucker rod pumping wells. The simulation results indicate that GM-ELM method has higher accuracy of fault diagnosis than GRNN(general regression neural network)， LS-SVM(least squares support vector machine)， BPNN(back propagation neural network).

Key words: ELM, feature extraction, fault diagnosis, gray matrix, dynamometer card