Fault Diagnosis of Rod Pumping Wells Based on Support Vector Machine Optimized by Improved Chicken Swarm Optimization

Jinze Liu,Jian Feng,Xianwen Gao

doi:10.1109/access.2019.2956221

Abstract

Nowadays, rod pump is widely used in oilfield. Since most oil production equipment like pumping pumps are distributed in the wild, they are usually checked by manual inspection. In the event of a faults, relying solely on labor to observe the indicator diagrams and determine the fault will waste a lot of human and financial resources. If it is not discovered in time, it will cause serious damage to oil exploitation, even shutdown. Indicator diagrams can reflect the working state of the rod pumping well, which can effectively reflect various faults of the pumping well. This paper diagnoses the faults of pumping wells by classifying and identifying the indicator diagrams. Because support vector machine (SVM) has good effect on classification and recognition of small sample data and nonlinear data, this paper uses SVM for classification, and uses the chicken swarm optimization (CSO) to optimize support for the problem that the SVM parameters are difficult to determine. Aiming at the problems of traditional CSO in solving high-dimensional optimization problems, such as premature and rough precision, an improved CSO is proposed. The traditional CSO, particle swarm optimization (PSO) and bat algorithm (BA) are used to compare it. The simulation proves that the improved CSO has good optimization effect and is superior to the other three optimization algorithms.

Highlights

With the continuous development of the petroleum industry, rod pumping has been vigorously developed and widely used in the petroleum industry [1]–[5]
Fault diagnosis for pumping wells is mostly based on analytical indicator diagrams
The research methods for the fault diagnosis technology of rod pumping wells mainly include the application of intelligent systems such as expert system, computer diagnosis, artificial neural network and support vector machine in the pattern recognition of the indicator diagrams

Summary

INTRODUCTION

With the continuous development of the petroleum industry, rod pumping has been vigorously developed and widely used in the petroleum industry [1]–[5]. The research methods for the fault diagnosis technology of rod pumping wells mainly include the application of intelligent systems such as expert system, computer diagnosis, artificial neural network and support vector machine in the pattern recognition of the indicator diagrams. The existing measured indicator diagrams and well acoustic curve were used as samples to train the neural network This algorithm was used to build a rod pump diagnostic expert system. In 2014, Ding and Li [23] mainly used the geometric parameter method and the gray matrix statistic value to extract the features of the indicator diagrams, and applied the selforganizing neural network to establish the model for fault diagnosis.

DACSO ALGORITHM

BASIC THEORY

ALGORITHM IMPROVEMENT

Findings

CONCLUSION AND FUTURE RESEARCHES