Research On The Seismic Phase Identification And The Source Location Based On Oilfield Fracture Microseismic Monitoring

The oilfield water filling and real-time fracture microseismic monitoring arepopular research topic both at home and abroad; with the maturity of theoreticalresearch to microseismic monitoring technologies, microseismic monitoringtechnologies are widely applied in the crack exploration of hydraulic fracturemodification in the low seepage oilfield; therefore, research to the treatment methodsof seismic data is of vital significance. The microseismic monitoring is divided intoground monitoring and well monitoring. Given the stratum absorption andtransmission route complexity, the materials obtained by ground monitoring sufferfrom such defects as less microseismic events, low signal: noise ratio and poorinversion reliability compared with well monitoring; however, the well monitoringsuffers from such problems as equipment precision, complex construction andoperation and rather high cost. All in all, it is particularly important to improve thetreatment methods of seismic data, explore the optimization and improvement route ofthe methods and upgrade the source location and inversion precision.In the noise background, signal inspection is an important part of microseismicmonitoring. Once the microseismic event is automatically and accurately inspected inthe seismic signal, it can greatly upgrade the subsequent data treatment efficiency.However, frequency or identification methods have a rather slow calculation speed inthe frequent seismic phase identification methods and can not meet the demand forreal-time monitoring; the time domain identification methods have a rather lowprecision in the treatment of data with rather low signal: noise ratio. Therefore,research to the microseismic data treatment methods and seismic phase identificationmethods are main contents in this article.In the hydraulic fracture and storage layer monitoring, microseismic sourcelocation is a highly important research topic. The real-time monitoring of growth status of underground rock stratum crack in the hydraulic fracture can raise theoreticalbasis for the construction and adjustment of fractured oil reserve. The existing sourcelocation methods need precise judgment of arrival time of microseismic signal; duringthe monitoring by ground matrix method, it is difficult to identify seismic event in thesignal; the well monitoring is featured as relatively high cost and technical difficulties.Aimed at the features of microseismic signals, research to the seismic source locationmethods is another main content.This article is supported by the national scientific and technological topic:“monitoring technical research and equipment research and fabrication of coal bedgas” and based on microseismic monitoring and aims to launch out the research to themicroseismic signal and seismic phase identification methods and source locationmethods. The concrete research contents are as follows:(1) Discriminative frequency estimate algorithm based on naive bayesianclassificationNaive bayesian classification is a typical method in the classified data excavation.It is featured as simple structure and swift calculation speed. As a generation learningmethod, the basic assumption is irrelevant to all the properties; in the real practice, theassumption is not always reasonable—the article depends on naive bayesianclassification and raises discriminative frequency estimate algorithm. This methodaims to calculate the occurrence frequency of data concentration and differentproperties and thus learn related parameters; at the same time, it has to introduce ajudgment parameter to enhance the association between different properties andclassified target value. This method can not only guarantee classification efficiency,but also greatly enhance classification precision. Test result has revealed thatdiscriminative frequency estimate algorithm raised in this article is much superior tothe traditional decision tree calculation method, naive bayesian classification methodand Boost method in the respect of classification precision and can realizeclassification precision supporting SVM. In the training time, it is much less thanSVM calculation methods; we can conclude from the test that SVM calculationmethods can reach20—50times of the discriminative frequency estimate algorithmin the respect of training time needed for most data set; in the individual propertyquantity and training concentration example, the training time of SVM calculationmethods with numerous quantity and concentrated data is100times of the calculationmethod in this article. (2) Pattern recognition method of seismic phase identification based onSTL/LTASeismic phase identification is a key problem in the microseismic monitoring; atpresent, the most frequently used seismic phase identification method is STA/LTAenergy ratio method—it suffers from the following problem: once the signal is ratherweak, identification ability is rather low; it needs to set up the value of identificationvalve through artificial experience; once the valve value is too high, the realmicroseismic event will be lost; once the valve value is too low, numerousidentification errors will be emerged. Aimed at this problem, the article raised anidentification method of seismic phase mode—it adopts energy ratio sequencecalculated by STA/LTA method as the input parameter and applies discriminativefrequency estimate algorithm for classified identification. This method does not needmanual setup of valve value; it can effectively identify the signal with rather lowsignal: noise ratio. Test result has revealed that this method enjoys95%classificationprecision in the data set of this article; relative to the traditional STA/LTA method, ithas upgraded the precision for10%. Besides, it can implement training and produceclassification device before fracture construction; in the real monitoring, it only needsto input data calculated by STL/LTA method into the classification device withoutincrease of calculation time; it can meet the demand for real-time monitoring. Besides,the classification precision can further upgrade the classification precision with theincrease of training data concentration or addition of other properties as classificationparameters.(3) Research to source location methods based on overlapping of seismicamplitudeMost of the existing microseismic source location methods need to make anaccurate calculation of the arrival time of microseismic event; once the microseismicground monitoring methods are adopted, it is difficult to accurately identifymicroseismic event in the collected data; therefore, the inversion location has a ratherlow precision. Aimed at this problem, the article borrowed the inverted displacementmethod theory in the seismic exploration to raise a seismic source location methodapplicable to low signal: noise ratio. It can displace the collected seismic dataaccording to different position of monitoring area and time difference of each detectorand then overlap data in various channels for calculation. Therefore, it can upgradethe signal: noise ratio of signal; displacement and overlapping can produce a great energy gathering in the real position of microseismic event. This article aims to verifythe calculation methods through simulation data and actual data; test result hasrevealed that once signal: noise ratio of simulation data is0.5, the error of locationprecision is between5—15m; during the location of hammering source, the locationerror is about15m. Therefore, the calculation method still enjoys a rather highlocation precision during the inversion location of seismic signal with rather lowsignal: noise ratio.