设计了一种求解一维大地电磁测深反演问题的实数编码混合遗传算法,它是通过单纯形搜索与遗传算法结合而成。针对传统的遗传算法在优化应用中存在局部搜索能力弱、计算量大、对较大空间适应能力弱和早熟收敛,而基于局部线性化的单纯形法易使解陷入局部极小值,严重依赖初始模型的选择等问题,在遗传算法中加入一个改进的单纯形搜索算子,并采用最优群体保留策略。该新算法既具有遗传算法的全局收敛性,又具有单纯形法的快速收敛性。对各种类型的大地电磁测深理论曲线进行计算,结果表明:采用实数编码混合遗传算法进行反演具有收敛速度快、解的精度高和避免出现早熟等优点,可用于大地电磁资料解释。
The inverse problem of MT for 1-D was studied by using a real coded hybrid genetic algorithm, which was based on the combination of simplex method and genetic algorithm. Due to the fact that the standard genetic algorithm has poor local search ability, large amounts of calculation, and adaptability to large space, and that simple method based on local linearization is usually lost in local minimum values, a new method was put forward through a promoted simple operator embedded into genetic algorithm and the strategy was adopted to keep the group of best individuals. The new algorithm has not only the global of genetic algorithm, but also the fast convergence of the simplex algorithm. The inversion results of synthetic magnetotelluric sounding data are ideal, which indicates that the algorithm possesses advantages of expediting convergence, avoiding earliness and improving precision, and can be used in MT data analysis.