Evidence constrained agentic retrieval augmented generation for substation civil engineering preliminary design documents in a single project study
Year: 2026
Type: article
Abstract: Substation civil engineering preliminary design documents must integrate project-specific conditions, current regulatory requirements, typical design practices, and historical engineering experience while maintaining consistency across architectural, structural, general-layout, foundation, road, drainage, and technical-indicator sections. Conventional Retrieval-Augmented Generation (RAG) systems remain vulnerable to condition-insensitive retrieval, unequal source authority, incompatible evidence, unsupported engineering claims, and parameter drift during long-document drafting. This study proposes an Evidence-Constrained Agentic Retrieval-Augmented Generation framework, termed EC-ARAG. The framework combines regulation-aware evidence representation, validity-gated hybrid retrieval, source-role-aware conflict resolution, dual planning--retrieval and generation--verification--repair loops, and global project-state memory. The experimental corpus comprised 36 national and industry standards, 12 enterprise standards, 18 typical-design documents, 64 historical preliminary design documents, and nine project-input documents. The held-out evaluation included 120 retrieval queries, 80 applicability cases, 48 evidence-conflict cases, 318 atomic design propositions, and five independent generation runs. EC-ARAG achieved a Recall@5 of 94.1\%, an applicability Macro-F1 of 90.6\%, an evidence support rate of 95.0\%, section completeness of 96.0\%, and cross-section parameter consistency of 97.0\%. Its unsafe-claim rate and character-level manual-edit rate were 3.4\% and 7.6\%, respectively. Although agentic verification increased machine inference time, total operational completion time was 21.4\% lower than that of closed-corpus Corrective RAG. Under the evaluated single-project, text-only conditions, EC-ARAG improved the reliability, traceability, and professional editability of regulation-intensive preliminary design drafting.
Source: Discover Artificial Intelligence
Authors: Yizhang Huang, Xinhui Zhang
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,590
OpenAlex ID: W7204262160