Research On Measurement Of Dynamic Liquid-Level Of Sucker Rod Pumping System Based On Improved SVR Method

With the continuous exploitation of crude oil, energy that keeps oil flowing decreases. At last, the energy can’t be well kept flowing. Then we extract oil by external mining machinery. The sucker rod pumping that is taken as the main equipment is widely used in major oil fields. However, along with the application of the sucker rod pumping, it also brings us a lot of problems; in the oilfield exploitation, oil wells have no enough supplying power. A number of oil wells appear intermittently producing, and even dry pumping phenomenon. If these parts of the wells are mined all day, it could lead to a greater waste of energy and affect the useful life of equipment. In actual production, dynamic fluid is an important parameter. If we get this parameter, we can make ability mining of oil extraction equipment match with reservoir fluid production ability; Real-time closed-loop regulation of the pump parameters could be implemented. Oil wells are improved on low-yielding inefficient economic.The dynamic measurement of liquid level is mainly echo method in engineering practice. But this method has three drawbacks:the echo technique is constrained by well condition; as real-time online measurement is difficult, the promotion is difficult; measure is dangerous. Some of the scholars estimate the value of fluid level, through methods of dynamometer to measure fluid level, method of pressure and method based on support vector machine for non-stationary time series. These are many problems in effectiveness and practicality of the measuring. In this paper, we directly take SVR modeling. We use genetic algorithm (GA) to automatically find the optimal parameters for improving selection efficiency and accuracy. Global kernel function fits to achieve sample correlation distance; local kernel function approximately fits vicinity correlation data fields. Using a hybrid kernel improves the model generalization. After a period of study, in the practical application System, it just has a bigger relevance with the near the working point of the data. The correlation of data that is away from the working point is not very great. Therefore, the model should be constantly updated and changed along with the change of system. As the change of working condition, the old data is to be thrown away and the new data is to be added. Application of a new data model can reflect the current situation, in order to ensure the accuracy of the model. Based on the analysis of theory and actual working condition, we adopt sliding window strategy implementation to weed out the old data and get new data update.Based on the experimental test and theoretical analysis, this paper is in accordance with the principle from simple to complex. In proper sequence, we use the support vector machine, the optimal parameters SVR, the optimal parameters and combination kernel function (MKF) is improved SVR, and the optimal parameters and combination kernel function and sliding window (SW) SVR. Step by step, we get the final optimization algorithm. Combining with the actual project, develop software. Finally it passes through the acceptance of the oilfield in2012.