Research On Measurement Of Dynamic Fluid Levels In Oilfield Based On Soft Sensor

Dynamic fluid level data is one of the most important factors in oilfield. Bottom-holeflowing pressure calculated by dynamic fluid level，it can make reasonable pumping stateby controlling the frequency of pumping unit. However, it is very difficult to measure thedynamic fluid level. There are four common methods after now, such as buoy,dynamometer card, pressure measurement and echo method. The methods mentionedabove are all mechanism methods and can’t change the huge workload caused by handoperation and safety issues. Therefore, we use soft sensor to gradually replace manualmeasurement to avoid the above problems.The problem of dynamic fluid level has been searched and discussed in this paper:LSSVM is selected as the most important algorithm of the modeling for prediction ofdynamic fluid level.Chaotic characteristics is found in dynamic fluid level time series by analyzing beforethe system is established to make the dynamic fluid level prediction system qualified forthe job. Soft sensor can’t accurately and long-termly predict the chaotic time series basedon the related theories research at the present stage. In this paper, the dynamic fluid leveltime series are reconstructed in phase space, because of the time series with chaoticcharacteristics needs analysis and modeling in phase space. It could provide a reliable basicguarantee for building the system by mining the potential data information of dynamicfluid level time series.It is the black hole algorithm instead of the artificial assignment that is proposed tosolve the parameter assignment problem of LSSVM in this paper. The black hole algorithmchanges the parameter assignment problem into the optimization problem. The establishingof the prediction model of dynamic fluid level based on BH-LSSVM is not only improvethe automation level of the prediction system, but also reduce the bothers from theartificial parameter selection and debugging. Through the simulation experiments of the phase space reconstruction of the dynamicfluid level using LSSVM, PSO-LSSVM, BH-LSSVM. The results of the experiments showthat BH-LSSVM has higher accuracy for predicting the dynamic fluid level of oilfield.Three kinds of auxiliary parameters are added into the prediction model, and the resultof simulation experiment proves that the new prediction model is better than traditionalmethod. The new prediction model which with three kinds of auxiliary parametersimproves the prediction accuracy of the phase space reconstruction of the dynamic fluidlevel and makes the prediction errors meet the oilfield’s error requirement. So the newmodel can be applied to the measurement of the dynamic fluid level in oilfield.Saturated data has been a big problem for updating the soft sensor model. Therefore,sliding window method and forgetting factor method are being used in this paper to solvesaturated data problem, respectively. Each of these two methods has its own merits, so youcan select one of them by your actual requirements.In this paper, the dynamic fluid level predicting system could make short-termlyprediction and successfully reduce the workload of the workers in oilfield. Otherwise, theusing of the dynamic fluid level predicting system avoids the potential safety hazards andimproves the automation level of oilfield as well.