基于GA和PSO智能算法的大地电磁一维反演分析

One-dimensional Magnetotelluric Inversion Analysis Based on GA and PSO Intelligent Algorithms

当前,大地电磁勘探方法已逐渐适用于铁路隧道勘察项目中,其实际数据反演计算是重要一环,关于其智能优化算法的反演测试正在逐渐兴起。本文基于大地电磁法相关理论,编写了遗传算法(Genetic Algorithm,GA)和粒子群(Particle Swarm Optimization,PSO)算法反演程序,利用这两种算法对典型一维三层地电模型正演模型数据进行反演测试,并将正演信号加入10%高斯噪声的方式进行了反演抗干扰能力测试。结果表明:两种方法对理论模型模拟的观测数据反演精度都较高,并且PSO算法的迭代收敛速度明显要快于GA算法,但加噪后反演时GA算法相较于PSO算法抗干扰能力更强。然而,由于GA算法采用二进制编码,迭代过程较为复杂,迭代收敛较为耗时,计算所占用的内存更多。两者比较而言各有优劣,应根据实际资料解译情况进行适当选择。

At present,the magnetotelluric exploration method has gradually been applied to railway tunnel surveys,and the inversion calculation of its actual data is of vital importance.The inversion test of its intelligent optimization algorithm is gradually emerging.Based on the related theories of magnetotelluric method,genetic algorithm(GA)and particle swarm optimization(PSO)are programmed in this paper.These two algorithms are used to test typical one-dimensional three-layer geoelectric model,and the anti-jamming ability is tested by adding 10%Gaussian noise to the forward signal.The results show that the two methods have high inversion accuracy for the observation data simulated by theoretical model,and the iterative convergence speed of PSO algorithm is obviously faster than that of GA algorithm,but the latter has stronger anti-interference ability in inversion after noise addition.However,because GA algorithm adopts binary encoding,the iteration process is more complicated,the iteration convergence is more time-consuming,and the calculation takes up more memory.In comparison,the two have their own advantages and disadvantages,and should be appropriately selected according to actual problems.