Research On The Detection Method Of Oil Well Dynamic Liquid Level Based On Echo Signal

Dynamic fluid surface depth detection is an important topic in the field of oil industry.Its research plays a crucial role in understanding real-time oil storage and making production management regulations.It is of important value to capture oil wells' production status on line and enhancing the ability of safety production.Therefore,in order to solve the problem of dynamic fluid surface depth's calculation and prediction,this work focuses on how to decide the position and depth of dynamic liquid surface and also study the prediction models of dynamic liquid surface depth.It involves some comparative discussions of multi-layer perceptron and linear regression prediction models,and meanwhile the gradient boosting decision tree model is used to study the prediction model of dynamic fluid surface depth in terms of an improved genetic algorithm and the acoustic wave method.The main work and the achievements can be summarized as follows.A.The acoustic wave method is difficult when applied to determining the location of oil wells' fluid surface,due to serious environment disturbance in the oil well and weak acoustic signals at the location of fluid surface.Thus,an estimation method is developed to calculate the dynamic liquid surface depth of oil well with the help of multiple liquid surface candidate positions and the average acoustic velocity.More precisely,several hidden candidate positions of liquid surface included in the acoustic signals are extracted by designing an improved short-time energy over-zero function.On the other hand,the average velocity of acoustic signals is calculated through constructing an improved three-center clipping function model.Additionally,a LSTM-based prediction model of dynamic liquid surface depth is developed,in which the acquired candidate liquid surface positions and average velocity are taken as the input of the LSTM neural network.Comparative experiments show that the estimation method can effectively detect the dynamic liquid surface position and calculate the liquid surface depth,while the obtained prediction model can initially perform the short-term prediction of dynamic liquid surface depth.B.The multi-layer perceptron and linear regression models are taken as the representations of dynamic fluid surface depth prediction,by which four prediction models are acquired for predicting the dynamic fluid surface depths of single and multiple oil wells with the help of an improved gradient learning algorithm,the least squares method and the fastest descent method.The experimental analysis shows that the multi-layer perceptron and linear regression models can obtain different prediction results when used for single oil wells' dynamic fluid depth prediction,in particular the multi-layer perceptron model can perform well.However,the linear regression model is weak when used for predicting the dynamic fluid surface depths of multiple oil wells.C.When the acoustic sensor sends acoustic wave signals to an oil well continually,a time series of dynamic liquid surface depth values can be obtained and used for constructing a training sample set.This can guide us to design a decision tree prediction model so as to predict multiple oil wells' dynamic liquid surface depth.Herein,a gradient boosting decision tree model is taken as the prediction model of dynamic liquid surface depth,and later an improved genetic algorithm is used to optimize the structure of the model.The comparative experimental analysis shows that the improved genetic algorithm-based decision tree model can effectively predict each oil well's dynamic fluid surface depth.