A Rank-Based Hybrid Model Management Strategy-Driven Two-Stage SAEA for the Inversion of Soil Thermal Resistivity for Power Cable Systems

Accurate soil thermal resistivity is crucial for real-time cable ampacity determination to maximize cable utilization. However, the determination of soil thermal resistivity involves solving a computationally expensive multi-physical field inverse problem where a high-fidelity model (HFM) is used for performance evaluations. Surrogate-assisted evolutionary algorithms (SAEAs) are computationally efficient for such problems; model management strategies (MMSs) are key to SAEAs. Nevertheless, most MMSs struggle to balance the computational cost and the search accuracy due to their reliance on fitness value errors. In fact, maintaining a similar function landscape between the surrogate and the HFM is more essential than achieving precise fitness values on the surrogate. Consequently, a rank-based hybrid MMS-driven two-stage SAEA is proposed. Stage 1 focuses on identifying promising regions. To ensure the similarity between the surrogate and HFM function landscape and thus guide the evolution accurately, a global MMS is proposed. Specifically, a new function landscape similarity metric is proposed to adaptively adjust the surrogate update frequency. A new rank-error-based individual selection strategy selects key individuals for exact evaluations to refine the surrogate similarity. Stage 2 performs a refined local search within the identified promising region, utilizing a local MMS to re-evaluate the optimum of a local surrogate built around the best solution searched in Stage 1. Optimization results confirm the proposed method’s superiority on test functions and a prototype cable.