| Dynamic fluid level is the liquid level in annular space of the well formedbetween the sucker rod and the sleeve, also one of the most important parameters ofoil well’s working condition for comprehensive analysis. Presently, acoustic reflectionmethod is usually used to measuring the dynamic liquid level depth value. Because ofthe complex state of annular space of the well, in the transmission process, soundwaves will be diffused and interfered by various kind of noises, making the liquidsurface reflection wave and coupling reflection wave hardly to identify, and in the endcan not make accurate identification of dynamic liquid level depth values from theacoustic reflection signal obtained.In view of the problems faced in the sound wave reflection method during theprocess of dynamic liquid level measurement, this article summarizes the law ofsound waves transmission in annular space and reasons of the noises occurred duringthe process of dynamic liquid level measurement. The research and development ofdigital automatic measuring device and the related features of is discussed. in theaspects of generation of acoustic, acquisition of dynamic liquid level reflected signal,design of controlling circuit, remote wireless communication module and so-on, theworking process of the automatic measurement device and the method of use aredescribed in detail.Studied the theory of blind source separation, and the realize of blind sourceseparation method and requirement of the condition that the number of observedsignals is less than the number of source signal is discussed, enumerated several kindof common mixing matrix estimation and source signal estimation method; Choosesparse component analysis(SCA) which is widely used in underdetermined blindsource separation as the main method of separating the working fluid level reflectedsignal from mixed signals with noise. Wavelet analysis is used in pretreatment processand noise reduction in order to improve the success rate of separation; Use clusteringmethod based on potential function to estimate the mixing matrix and use a linearprogramming method that converted from a posteriori probability method to estimatethe source signals. By using the dynamic liquid level reflection signal from the fieldtest, successfully extracted more clear liquid surface reflection wave signal fromobserved signals mixed with the Gaussian white noise, proved the acceptable performance of the algorithm. |
