Soft sensor hybrid model of dynamic liquid level for sucker rod pump oil wells

Abstract

The measurement of the dynamic liquid level is of great importance during the oil-producing process of rod pumping wells. It may affect the production of oil fields and motor efficiency. However, the dynamic liquid level is difficult to detect because of the circumstances in the downhole of rod pump wells. In this paper. Firstly, according to the working principles of the sucker rod pump, the mechanical models of the pumping motor and four-bar linkage mechanism are respectively built. Secondly, for the underground frictions, a mechanical model based on the energy conservation equation is built and then the mechanism model is built between the dynamic liquid level and power of the motor. To improve the accuracy of the mechanism model, a novel method based on an artificial fish swarm algorithm optimization Gaborc-kernel extreme learning machine is used to establish a soft sensor dynamic liquid level error compensation model. The mechanism model is paralleled with the soft sensor model to establish a hybrid model of dynamic liquid level. Eventually, the AFSA-Gaborc-KELM soft sensor hybrid model is verified by using the oil dataset collected from the electrical parameter acquisition equipment. This hybrid model is compared with some other models. In the comparison, the proposed hybrid model has better performance and prediction accuracy for the dynamic liquid level than the BP hybrid model, GA-ELM hybrid model, and LSSVM hybrid model.