摘要
利用加速度信号测量位移是油田抽油井光杆位移测量的主要方法,而加速度信号的随机噪声和趋势项是影响测量精度的主要因素,本文提出了一种基于学习的实时消噪和剔除趋势项方法。学习时先获取一段时间的加速度信号,再通过时间序列分析技术得出ARIMA模型及其参数,最后基于FFT变换的Rife-Jane频率估计方法求出加速度信号的周期;在线实时消噪和剔除趋势项方法是基于学习阶段所得模型参数,运用卡尔曼滤波技术消除加速度信号随机噪声;按周期两次积分得到光杆位移,用加窗递推最小二乘法在线消除趋势项。通过抽油机半实物仿真平台测试和分析加速度信号,结果表明,该方法有效地去除了加速度信号中的噪声和趋势项,极大地提高了位移的测量精度。
Acceleration signal-based approach is a main approach used to measure the displacement of polish rod in the oilfield pumping wells.In this study,a learning-based real-time noise immunization and trend term elimination approach is proposed.During the learning,acceleration signal within a period of time are firstly acquired,then corresponding ARIMA model and its parameters are derived,finally the period of the acceleration signal is computed by using FFT transformation and Rife-Jane frequency estimation.The proposed approach bases the model obtained parameters,uses Kalman filtering techniques to remove random noise,computes the polish rod displacement by a quadratic integration of the period and eliminates the trend term by windowed recursive least-squares method.Eventually,an experiment over a pumping unit hardware-in-the-loop plant is carried out,which indicates that the proposed approach can effectively eliminate noise and trend term and obviously improve the measuring precision of displacement.
出处
《电子设计工程》
2013年第14期18-22,共5页
Electronic Design Engineering
基金
陕西省基金项目(2011K06-25)
总装预研项目(2011DA090002C090002)
