电子测量与仪器学报

卢勇,阎力恒,谢肖肖,陈嘉昕,李陇杰.全局-局部混合算法瞬变电磁反演与接地网检测[J].电子测量与仪器学报,2026,40(3):250-261

全局-局部混合算法瞬变电磁反演与接地网检测

Global-local hybrid algorithm transient electromagnetic inversion andgrounding grid detection

DOI：

英文关键词:transient electromagnetic method grounding grid hybrid optimization algorithm global search-Local correction

基金项目:长安大学青年学者学科交叉团队建设项目（3100104240923）资助

作者	单位
卢勇	长安大学能源与电气工程学院西安710018
阎力恒	长安大学能源与电气工程学院西安710018
谢肖肖	长安大学汽车学院西安710018
陈嘉昕	长安大学能源与电气工程学院西安710018
李陇杰	长安大学能源与电气工程学院西安710018

Author	Institution
Lu Yong	School of Energy and Electrical Engineering, Chang′an University, Xi′an 710018,China
Yan Liheng	School of Energy and Electrical Engineering, Chang′an University, Xi′an 710018,China
Xie Xiaoxiao	School of Automotive Engineering, Chang′an University, Xi′an 710018,China
Chen Jiaxin	School of Energy and Electrical Engineering, Chang′an University, Xi′an 710018,China
Li Longjie	School of Energy and Electrical Engineering, Chang′an University, Xi′an 710018,China

摘要点击次数: 321

全文下载次数: 178

中文摘要:

针对瞬变电磁视电阻率法在接地网缺陷检测中面临的非线性反演精度不足、收敛效率低下等问题，提出一种融合遗传算法(genetic algorithms, GA)全局搜索与牛顿法局部优化的智能混合反演方法。针对传统遗传算法在缺陷检测中收敛缓慢、小尺度缺陷敏感度不足的局限，构建“数据驱动-模型约束”反演框架—通过遗传算法的锦标赛选择、动态交叉变异等反压机制，避免单一数据驱动模型的“黑箱映射”局限，实现初始解空间的可解释性搜索；再以全局搜索所得高质量初值作为牛顿法输入，从根本上解决传统迭代方法的“初始值敏感性”问题，形成“全局预搜索-局部细修正”协同反演策略。实验结果表明，在2 000~10 000组数据场景下，混合算法总体平均总耗时7.2~35.8 s，较传统遗传算法与牛顿法总和耗时（12.0~48.4 s）减少4.8~12.6 s；迭代效率显著提升，达到预设最大迭代次数（100次）终止的案例较牛顿法减少33.6%，迭代次数小于20次的有效解占比提高45.4%；反演精度优势显著，平均误差6.020 7×10-8，较传统迭代法与遗传算法分别降低83.65%和98.95%。最后通过野外缩比模型实验验证，该方法可有效识别接地网拓扑结构及部分断裂、缺口等隐蔽缺陷，在复杂工况下的检测精度与效率较单一方法显著提升。

英文摘要:

To address the challenges of insufficient nonlinear inversion accuracy and low convergence efficiency in grounding grid defect detection using the transient electromagnetic apparent resistivity method, this study proposes an intelligent hybrid inversion approach that integrates the global search capability of genetic algorithms (GA) with the local optimization of Newton’s method. To overcome the limitations of traditional GA, such as slow convergence and low sensitivity to small-scale defects, a “data-driven and model-constrained” inversion framework is established. This framework employs reactive mechanisms in GA, including tournament selection and dynamic crossover and mutation, to mitigate the “black-box mapping” limitations of purely data-driven models and achieve interpretable searches in the initial solution space. High-quality initial values obtained from the global search are then used as inputs for Newton’s method, fundamentally resolving the “initial value sensitivity” issue of traditional iterative approaches and establishing a collaborative inversion strategy of “global pre-search and local refinement.”Experimental results demonstrate that, under scenarios involving 2 000 to 10 000 data groups, the hybrid algorithm achieves an average total computation time of 7.2~35.8 seconds, representing a reduction of 4.8~12.6 seconds compared to the combined time of traditional GA and Newton’s method (12.0~48.4 seconds). Iterative efficiency is significantly improved, with cases terminating at the preset maximum iteration limit (100 cycles) reduced by 33.6% compared to Newton’s method, and the proportion of valid solutions obtained within fewer than 20 iterations increased by 45-4%. The inversion accuracy is notably superior, with an average error of 6.020 7×10-8, reflecting reductions of 83.65% and 98.95% compared to traditional iterative methods and GA, respectively. Finally, field scaled-model experiments confirm that the proposed method can effectively identify grounding grid topologies and detect hidden defects such as fractures and gaps, significantly enhancing detection accuracy and efficiency under complex operational conditions compared to single-method approaches.

查看全文查看/发表评论下载PDF阅读器