The application discloses a question and answer
system and method based on retrieval enhancement generation, and relates to the technical field of industrial visual question and answer.The
system comprises a reflection filing module, a local reflection mapping matrix is constructed, and a reflection sensitive operation domain is generated; a semantic tracing module, which maps an external target real track to a predicted virtual track by using the matrix, compares abnormal behavior events, and obtains a reflection source evidence
score; a
label enhancement module, which combines direct operation evidence, determines a semantic
source type as a direct source, a reflection source, a mixed source or an undetermined source to generate a reflection source enhanced
label; and a retrieval question and answer module, which generates context according to the enhanced
label, and verifies the consistency of a factual conclusion and a semantic
source type when an answer generation model outputs.The application effectively blocks the
pollution of optical illusions on the
knowledge retrieval link, and guarantees the reliability of power
supply and distribution safety event attribution.