Research On Micro-seismic Inversion Based On BP-GA Mixture Algorithm

As a geophysical exploration technology, Micro-seismic monitoring technologywhich can real-time monitoring the reservoir fracture and drive has developed rapidlyin recent years. And also it has great significance in the exploration and developmentof oil and gas fields. Micro-seismic monitoring operations designed to received theproduction activities induced micro-seismic events by the detectors array laid inmonitoring wells or ground, to inverse the fracturing parameters such as sourcelocation to apply them in productive activities to monitor and guide. Most of thegeophysical inversion is an objective function optimization problem withmulti-extreme value, so it can easily fall into local extreme which solved with lineartheory, and the results are not reliable. Following the trends that the life engineeringscience mutual penetration, cross, as well as to promote the development of modernscience, it fully digest and absorb domestic and foreign scholars to research onnonlinear optimization problem in this paper, and proposed the combined hybridalgorithm of BP neural networks and genetic algorithms, and applied into themicro-seismic monitoring the key issues—micro hypocenter.The first part describes the methods of micro hypocenter, such as classic Geigerposition method、and S-wave with the same wave different method and nonlinearinversion algorithm. Mainly introduces the basic concepts of nonlinear inversionalgorithm genetic algorithm and BP neural network algorithm, elements, theoreticalbasis and operational mechanism and characteristics of the shortcomings andlimitations of traditional algorithms, also made the proposed improvement measures.The second part designed a forward modeling with variety of different mediamodel according to the ray path tracking method. The media models including layeredhomogeneous medium model, layered curved medium model and complex mediummodel. The forward algorithm applied dichotomy in the shooting method and grid method based on the grid thinking of slightly changed.The third part is the focus of this article. First reviewed the development andapplication status of Genetic and BP neural network algorithm in micro-seismicinversion. With the in-depth analysis and study of the advantages and disadvantagesof two algorithms in the inverse problem, propose the strategies and programs of thehybrid algorithm with GA and BP algorithm mutual penetration and integration,which objective to improve the search performance of nonlinear global optimizationmethod and improve the efficiency and inversion accuracy of its operations in themicro-seismic parameters inversion.The fourth part designed the BP-GA mixed optimization algorithm with the GAand BP optimization toolbox integration on the matlab platform, and apply it to thesimulation of micro-hypocenter inversion simulation experiments. Main steps: First,the establishment of the layered homogeneous medium model under the forwardmodeling, get the time difference data samples of the sources assumed to the detectorarray, then use the sample training neural network optimized by the GA, to enablethem to forecast micro focal position coordinates of optimal weights and thresholdsconstraints.