Theory And Realization Of Torque Soft Sensing System For Motor In Beam Pumping Unit

Beam pumping unit is widely used in oilfield because of its compact structure, long service life and convenient maintenance. With the global energy shortage, its low efficiency becomes increasingly serious. Therefore the improvment of beam pumping unit for saving energy is more and more important. Accurately measuring the running parameters of beam pumping unit such as three-phase voltages, three-phase currents, useful power, power factor,the torque of motor shaft as well as the displacement and load of sucker rod, is the prerequisite of designing energy-saving equipment.At present, among all parameters, the torque to be measured has many problems including high cost, poor reliability and difficult installation. Soft sensing is the best and most effective solution to the problem which the parameter is difficult to be measured directly due to technology or cost. This paper studies the torque measurement by applying soft sensing. Its contributions are as follows:First, the current situation of the running parameters of beam pumping unit measurement is introduced, especially the significance, the current situation and the development of the torque measurement.A novel ideal of soft sensing is proposed to measure the torque. The research and development state of the soft sensing technique is summarized in the preface.Then, the dissertation illustrates the principle and characteristics, design procedure, modeling method of soft sensing. After analyzing the power transition relation and power loss of the motor, the structure and load condition and the operation principle of beam pumping unit, it creates two different torque soft sensing models. The useful power and the displacement of the rod and load of sucker rod were selected as secondary variables by analyzing the measured data used stepwise regression and combining with the mechanism of beam pumping unit. Two models were constructed based on BP neural network (BPNN) and support vector machine (SVM) respectively, and optimized by genetic algorithm. The generalization ability of those models was verified by the measured data and the computational complexity of models was tested in CCS2200 with TMS320LF2407 processor. Its conclusions are as follow:Firstly, all the models have grossly good accuracy; secondly, the mechanism models are simple, their computing time is less than 3ms,but they have less accuracy; thirdly, the accuracy of models based on BPNN is highest and its computing time is 5ms, the accuracy of models based on SVM is close to the former, but its computing time is more than 20ms.Furthermore, the pretreatment methods of the data for establishing soft sensing models were researched.In the dissertation the secondary variable sensors are selected according to the measurement precision requirement,and TMS320LF2407 is employed as soft sensing system microprocessor according to the computational complexity of the system, the basic peripheral circuit of the processor and the interface circuits of keyboard, LCD and sensors are developed. By analyzing software requirement of system, the structure and flowchart of the system software is determined. The real-time operating systemμC/OS-Ⅱwas embedded in the system to manage all the tasks. The application software is programming to acquire and preprocess the data of secondary variables, to calculate out the value of the torque by the previous models, to scan the keyboard, to display measurement data, to modify models, to expand and upgrade soft sensing models and to add other useful functions.The test result of torque soft sensing system for the motor in pumping unit shows that the system can both stably and reliably running and meet the need of measure ment accuracy.