Affiliations: [a] School of Instrument Science and Opto-Electronics Engineering, Beijing Information Science and Technology University, Beijing, China
| [b]
Beijing Yusheng Zhengchuang Technology Co., Ltd, Beijing, China
Correspondence:
[*]
Corresponding author. Haisha Niu, School of Instrument Science and Opto-Electronics Engineering, Beijing Information Science and Technology University, Beijing, China. E-mail: niuhs@buaa.edu.cn.
Abstract: Traditional mechanical models and sensors face challenges in obtaining the dynamometer diagram of the sucker rod pump system (SRPS) due to difficulties in model solving, high application costs, and maintenance difficulties. Since the electric motor powers the SRPS, its power output is highly correlated with the working state of the entire device. Therefore, a hy-brid method based on electric motor power and SPRS mechanical parameter prediction is proposed to predict the dyna-mometer diagram. First, a long short-term memory neural network (LSTM) is used to establish the LSTM-L model for predicting the dynamometer load based on electric motor power. Then, a mathematical and physical calculation model (FLM-D) of the dynamometer diagram displacement at the hanging point is constructed by combining the four-bar linkage structure of the sucker rod pump. Finally, the experimental production data of oil wells are collected through an edge computing device to verify the prediction performance of the LSTM-L&FLM-D hybrid model. Experimental results show that the proposed LSTM-L&FLM-D model has a high fitting degree of 99.3%, which is more robust than other models considered in this study, and exhibits better generalization ability.
