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

After many years of production, most of the yielding wells with jet capability in the early time have been developed into low-yielding wells relying on artificial lift now. The contradiction between the continuous declining of productivity and the increasing consumption becomes more and more obvious, how to reduce losses and increase the degree of automation of oil pumping system is an important issue in the oil fields to be resolved for a long time. In the actual production, the dynamic level is an important parameter, which directly reflects the ability of the fluid supply of the oil well. Adjusting the pumping time and pumping parameters in time, which need to achieve real-time monitoring of the dynamic level and master the production status of the oil well, is the key to improve the well productivity. As the existing detection method of dynamic liquid level, such as echo method and indicator diagram method, has the problem of non-real-time detection or low measurement accuracy, therefore, this paper established the detection model of the dynamic liquid level using support vector regression method.In this paper, the theory of support vector regression, the kernel function of support vector machine and its parameters selection are discussed in detail. On the basis of the exhibited characteristics of different kernel functions, mixed kernel function that is a linear combination of the radial basis function and the polynomial kernel function, is selected as the kernel function of support vector machine. The global kernel function is used to fit the correlation of the distant sample data, while the partial kernel function to fit the correlation of the data in neighboring fields in order to improve generalization performance of the model. Through comparing the simulation results of the single kernel SVR model with the mixed kernel function SVR model’s prove that the latter model has better generalization performance.The model parameters have a significant impact on its performance, the particle swarm algorithm is used to choose out a optimum combination of parameters in order to improve the generalization performance of the support vector regression model in this paper. Particle swarm optimization algorithm can easily fall into local optimal solution, against that drawing variation thought of genetic algorithm to improve PSO. The improved PSO has a better global search capabilityWith the continuing process of oil production, the parameters of oil well will change, therefore, it is necessary to update the model. Incremental algorithm can solve the bottleneck problem of computing, due to the data sets too large. So, this paper intends to use the incremental algorithm to update the model when the working conditions changed, so that the model can be adjusted in time to ensure its predictive ability.Finally, this paper establishes a sensor model of dynamic level for rod pumping system and made a simulation, using the above methods. From the viewing angle of the model predicting effects, the predicting effects of this method for dynamic liquid level of rod pumping wells is well and meet the project requirements fully.