Surrogate-assisted evolutionary algorithms (SAEAs) are well-suited for optimizing computationally expensive black-box problems in diverse real-world scenarios. The sample efficiency of SAEAs depends largely on the configuration of the surrogate model and sampling criteria. However, configuring these core components requires substantial manual effort and expert knowledge, limiting the broader applicability of SAEAs. To address these challenges, we propose CoE-SAEA, a novel paradigm that co-evolves large language models (LLMs) and configuration strategies to enhance SAEAs. Specifically, the paradigm consists of three populations with distinct roles: one evolves LLM prompts to generate robust configuration strategy instructions, another optimizes the configuration strategies, and the third solves the optimization problem using the selected algorithm configuration. Additionally, an exploration-exploitation module is incorporated to decide whether to explore new configuration strategies via LLMs or exploit existing ones. We empirically validate the efficacy of CoE-SAEA by comparing it to state-of-the-art algorithms across various benchmark problems and a real-world traffic signal optimization task. The source code of the proposed CoE-SAEA is publicly available at: https://github.com/ForrestXie9/CoE-SAEA.

Dynamic distribution network reconfiguration (DDNR) is a widely used technique for the secure and economic operation of power distribution networks (PDNs), especially in the presence of high-penetration renewable energy sources (RESs). DDNR is realized by controlling the on/off status of remotely controlled switches (RCSs) equipped at power lines in PDNs to optimize power flows. Thanks to the enhanced data availability of PDNs, data-driven solutions to DDNR, such as deep reinforcement learning (DRL), have gained growing attention recently. However, DDNR solves a sequence of combinatorial problems featuring a vast and sparse action space incurred by a so-called "radiality constraint," which is highly challenging for DRLs to handle. Existing DRL methods are either unscalable to large-scale problems or potentially restrict optimality. Hence, we propose a sequential masking strategy to decompose its complex action space into a sequence of maskable sub-action spaces. A gated recurrent unit (GRU)-based agent and an adapted soft actor critic (SAC) algorithm are designed accordingly, producing a data-efficient, safety-guaranteed, and scalable DRL solution to the DDNR problem. Comprehensive comparisons with existing data-driven methods and model-based benchmarks are conducted via various case studies, demonstrating the advantages of the proposed method in both algorithmic performance and scalability.