基于CGAN的抽油机电参数反演示功图研究
Study on the Inversion of Dynamometer Cards for Beam Pumping Units Based on Conditional Generative Adversarial Networks
李翔宇 1邓昱航 1袁春华1
作者信息
- 1. 沈阳理工大学 自动化与电气工程学院,沈阳 110159
- 折叠
摘要
示功图的精准测量在有杆抽油系统故障诊断中非常重要,针对载荷传感器直接测量法维护成本高、稳定性差,电参数间接测量法精度低、应用性不强的问题,提出一种基于条件生成对抗网络(CGAN)的电参数反演示功图混合模型.首先建立将电参数和机构参数转化光杆位移和负载的机理模型,生成粗糙的示功图样本数据;然后利用CGAN在图像转换领域的应用,建立粗糙示功图数据细化器,实现粗糙示功图与实测示功图之间的图像转化,使粗糙示功图与实测示功图更加相似;此外,为了使CGAN能更好地提取示功图轮廓,在生成器中加入自注意力机制进行改进.通过现场实测的电参数和示功图历史数据进行验证,结果表明该方法对比纯机理模型反演示功图的精度有显著提高.
Abstract
Accurate measurement of dynamometer cards is crucial in diagnosing faults in rod pum-ping systems.For the high maintenance cost and poor stability of direct load sensor measurement methods,and the low precision and weak applicability of indirect electrical parameter measurement methods,a hybrid model of electric parameter inversion dynamometer card based on conditional generative adversarial networks(CGAN)is proposed.Firstly,a mechanism model is established to transform electrical parameters and mechanical parameters into polished rod displacement and load,generating rough dynamometer card sample data.Then,leveraging the application of CGAN in the field of image transformation,a refiner is created for rough dynamometer card data to achieve image transformation between rough and actual dynamometer cards,making them more alike.In addition,in order to allow CGAN to better extract dynamometer card contours,a self-attention mechanism is incorporated into the generator for enhancement.Finally,this method is validated with field-measured historical electrical parameters and dynamometer card data.The results show that this method signifi-cantly improves the accuracy compared to the pure mechanism model inversion dynamometer card.
关键词
示功图测量/电参反演/条件生成对抗网络/图像转化Key words
dynamometer card measurement/electrical parameter inversion/conditional generative adversarial networks/image transformation引用本文复制引用
基金项目
国家自然科学基金项目(62173073)
辽宁省教育厅高等学校基本科研项目(LJKMZ20220618)
辽宁省本科教改优质教学资源建设与共享项目(SBKJGYZ-2021-06)
出版年
2024
NETL
NSTL
维普
万方数据