混合计算架构与知识库增强的资源受限财务报表分析方法OA

Enterprise financial statement analysis is traditionally characterized by a heavy reliance on manual labor and low efficiency.Further-more,small-parameter Large Language Models(LLMs)deployed in resource-constrained environments often suffer from numerical hallucina-tions,knowledge deficits,and cross-chapter semantic confusion.To address these challenges,this paper proposes a novel method for enterprise financial statement analysis based on hybrid computing architecture and knowledge base augmentation.This approach decouples numerical indi-cator calculation from LLM reasoning,employing structured query language to execute deterministic computations.Furthermore,this paper constructs an"indicator-industry-rule"knowledge base,parameterizing threshold,trend,and comparison rules into anomaly detection func-tions.Additionally,a"chapter-wise reasoning with content isolation"mechanism is designed,establishing exclusive prompts for individual sub-chapters to maintain context integrity.Experimental results demonstrate that this method effectively resolves issues regarding numerical ac-curacy and professional reliability in financial analysis under resource constraints,which offers a viable solution for lightweight financial state-ment analysis.