该文针对最小二乘支持向量机(LSSVM)中核函数参数与正则化系数的选择对高程异常拟合精度影响较大问题，通过引入混沌映射、非线性调整收敛因子及莱维飞行方法改进灰狼优化算法，提出改进灰狼优化LSSVM模型(IGWO-LSSVM)的高程异常拟合方法，为常规LSSVM提供最优核函数参数与正则化参数。分别以线状工程、面状工程两种典型工程的GNSS水准重合数据进行实验验证，结果表明：IGWO-LSSVM拟合方法的外符合精度较GWO-LSSVM拟合方法分别提高33%、19%,可为相关GNSS高程异常拟合工程提供一种参考。

In response to the significant impact of the selection of kernel function parameters and regularization coefficients on the accuracy of height anomaly fitting in Least Squares Support Vector Machine(LSSVM)models, this paper proposes an improved Grey Wolf Optimization LSSVM(IGWO-LSSVM) algorithm for elevation anomaly fitting by incorporating chaotic mapping, nonlinear convergence factor adjustments and Lévy flight methods. This model optimizes kernel function parameters and regularization coefficients for conventional LSSVM. Experiments are conducted using overlapping GNSS leveling data from two typical engineering projects: linear and planar engineering.Results show that the external accuracy of the IGWO-LSSVM fitting method improves by 33% and 19% respectively than that of the GWO-LSSVM fitting method, providing reference for relevant GNSS elevation anomaly fitting engineering projects.