Fault diagnosis of sucker rod pumping system using modified extreme learning machine assisted by gravitational search algorithm

Sucker rod pumping system is a kind of important oil production unit. How to diagnose its fault conditions based on the electric power data is a challenging task. Aiming at this problem, this paper builds a sucker rod pumping system fault diagnosis method based on the modified extreme machine learning. Two aspects, including the feature extraction and the classifier construction, are involved in the proposed method. Firstly, the electric power data are collected and data features are developed by applying wavelet analysis. Furthermore, a modified extreme learning machine assisted by gravitational search algorithm (GSA-ELM) is proposed to overcome the influences of random initialization of the input weights. In comparison with the basic ELM method, GSA-ELM can provide better generalization performance with a simple model structure. Lastly, the real industrial data are used to test the proposed method and the results demonstrate its effectiveness.