基于并行改进遗传算法的三维电阻率反演方法  

3D resistivity inversion using improved parallel genetic algorithm

作　　者：刘斌[1] 王传武[1] 杨为民[1] 李术才[1] 聂利超[1] 宋杰[1] 

机构地区：[1]山东大学岩土与结构工程研究中心,山东济南250061

出　　处：《岩土工程学报》2014年第7期1252-1261,共10页Chinese Journal of Geotechnical Engineering

基　　金：国家重点基础研究发展计划(973计划)(2013CB036002;2014CB046901);国家重大科研仪器设备研制专项(51327802);国家自然科学基金重点项目(51139004);国家自然科学基金青年项目(41102183);高等学校博士学科点专项科研基金项目(新教师类)(20110131120070)

摘　　要：计算效率极低是阻碍遗传算法用于三维电阻率反演的瓶颈,使得很多对改善反演效果和搜索质量有利但又很耗时的改进方法无法应用到遗传算法中。针对上述问题,基于遗传算法天然的并行计算特性,提出了新的多重主从并行计算策略及其算法。提出了初始群体的严格均布产生方法,以提高初始群体接近最优解的概率;提出了基于交叉个体适应度差异的比例随机算术交叉算法,以保证优良个体的遗传竞争优势;建立了混合变异算法,将传统的随机变异算法与线性反演中确定性搜索优化算法相结合,即保持了变异的随机性又控制了优化方向。最后将并行改进遗传算法用于合成算例和实际应用案例中,发现并行改进遗传算法的计算效率显著提高,且在寻找最优解、压制假异常、提高反演效果方面具有明显优势,为实际工程中电阻率探测的三维成像提供了有效途径。The low calculation efficiency of the genetic algorithm （GA） method is an obstacle to 3D resistivity inversion. Moreover, some improved methods which are time-consuming but beneficial for the inversion effect and the search efficiency can not be used in GA due to their low calculation efficiencies. To solve the above problems, a multi-level master-slave parallel computing strategy for GA is put forward based on the natural characteristics of parallel computing. Through this improvement, a generating method for strictly uniform initial population is proposed, with which the initial generation can be closer to the optimal solution. A random-ratio arithmetical crossover algorithm is proposed based on the differences of fitness values between the cross-individuals, which can keep genetic competition advantages of the better individual. Then the joint mutation algorithm is presented, which is the combination of the traditional random mutation algorithm and the deterministic search optimization algorithm in the linear inversion. It can maintain the randomness of the mutation and optimize the mutation direction. Eventually a 3D resistivity inversion using an improved parallelized GA is formed. The performance of the improved parallel GA is evaluated in synthetic and practical cases. The examples illustrate that the improved parallel GA can enhance the calculation efficiency significantly and has obvious advantages in searching the optimal solution, suppressing the false anomaly and obtaining high-quality inversion results. The improved parallel GA provides an effective way for 3D resistivity inversion imaging in practical projects.

关 键 词：三维电阻率反演成像 并行改进遗传算法 多重主从并行算法 比例随机交叉算法 混合变异算法 

分 类 号：TD853.34[矿业工程—金属矿开采]