基于变步长加速搜索的微震源定位方法

微震定位方法是微震监测技术的重要组成部分，其关键是定位震源位置。利用空间网格划分并计算网格交点目标函数值，对微震定位目标函数二维及三维空间分布进行了分析，并据此获取了目标函数连续且极小值唯一、单轴收敛范围逐步减小、各轴收敛范围不一的规律。利用以上规律及模式搜索法、网格搜索法的优缺点，探索出了基于连续比较模块、变步长模块、加速模块的变步长加速搜索法。通过模拟算例与工程数据下收敛稳定性、结果精确度、计算速度以及参数初始值影响程度4个指标的效果对比，结果表明：模拟算例下，对比模拟退火算法、遗传算法，变步长加速搜索法的目标函数值标准差、定位误差标准差、波速误差标准差均为0；该算法的定位误差平均值分别为其余二者的0.7%、1.9%；该算法的计算时间平均值分别为其余二者的6.9%、33.2%。该算法单独更改各参数对定位误差的影响在0.005～0.025 m之间；减小搜索步长下限可有效提高结果精确度，并增加相应的计算时间。在规定初至到时与目标函数模型及检波器位置坐标下，搜索算法对定位精度无实质影响。

Microseismic source locating method based on variable step size accelerated search

Microseismic locating method is an important part of microseismic monitoring technology, the key of which is to locate the hypocenter. We analyze the two-dimensional and three-dimensional spatial distributions of microseismic locating objective functions by using spatial gridding and calculating the objective function values of grid intersections. Accordingly, we find that the objective function is continuous and the minimum value is unique, the convergence range of single axis decreases gradually, and the convergence range of each axis varies. Using the above findings and the advantages and disadvantages of pattern search method and grid search method, we propose the variable step size accelerated search method based on continuous comparison module, the variable step size module and the acceleration module. Through the comparison of four indexes between the simulation example and the engineering data: the convergence stability, the accuracy of the results, the speed of calculation and the degree of influence of initial values of parameters, we get the results that: in the simulation example, compared with that of simulated annealing algorithm and genetic algorithm, standard deviations of objective function value, locating error and wave velocity error of the variable step size accelerated search method are all 0; the average locating error of the variable step size accelerated search method is 0.7% and 1.9% of that of other two algorithm, respectively; the average calculation time of this algorithm is 6.9% and 33.2% of that of the other two algorithm, respectively. The influence of the algorithm changing each parameter individually on the locating error is between 0.005 m and 0.025 m; reducing the lower limit of the step size for search can effectively improve the accuracy of the results but increase calculation time. When the initial arrival time, objective function model and coordinates of the geophone position are specified, the search algorithm has no substantial impact on the locating accuracy.