<p>Underground substations embedded in mountainous terrains in power plants, urban distribution networks, and subway traction stations typically use large, complex three-dimensional grounding grids. In diagnosing grounding grid corrosion in such scenarios, challenges arise from dimensional expansion, limited measurement ports, and difficulties in cross-layer measurements, leading to issues in modeling three-dimensional grids and solving underdetermined equations. An intelligent decoupling method is proposed. First, sensitivity analysis of branch ports reveals how corrosion states, electrical distances between branches and ports, and port configurations affect resistance variations. Attenuation effects in sensitivity are observed in both single-layer and multi-layer grids. Based on this, a hierarchical decoupling mechanism is proposed, constructing a single-layer grounding grid model focused on the core layer. The objective function minimizes the deviation in equivalent front and rear port resistances, and inter-layer equivalent resistances reduce the three-dimensional network to a two-dimensional plane. Next, the single-layer grounding grid model is combined with particle swarm optimization, considering grounding grid circuit constraints. Equivalent resistance parameters are treated as particle position vectors, with dynamic adjustment of the particle search step size based on the fitness function value for global optimization, resulting in the optimal decoupling solution. Experimental validation with two-layer, three-layer, and real-world power station grounding grids shows that this method maintains core layer topology while reducing dimensionality by 42.9–59.2%. Resistance variations are under 0.01mΩ, and fault branch locations are accurately identified, meeting engineering requirements. This research enables dimensionality reduction in grounding grid corrosion diagnosis without sacrificing accuracy, simplifying the diagnostic process.</p>