PVHD: Research and Application of PV Query Algorithm Based on Hybrid Retrieval and Dynamic Truncation Mechanism

Efficient and accurate querying of Process Variables (PVs), core parameters in the EPICS control system used in particle accelerators like the China Spallation Neutron Source (CSNS), is vital for real-time monitoring, rapid fault diagnosis, and precise experiment tuning. PVs often include Chinese meanings and descriptions alongside English strings. Traditional query methods struggle with the volume, diversity, and ambiguity of PV information due to incomplete user memory, synonym usage, spelling errors, and the inability to identify semantically similar PVs, impacting efficiency and accuracy. This paper proposes a PV query algorithm based on a multi-way recall weighted mechanism to address these challenges. The algorithm integrates deep semantic understanding recall using semantic similarity based on advanced word embedding models, efficient text feature recall employing keyword matching based on the BM25 algorithm, and intelligent fusion and optimized re-ranking of multi-way recall results through a dynamic weighting mechanism. A dynamic truncation strategy, adaptively filtering results based on the similarity score gradient, further optimizes query result quality and reduces redundancy. Experimental results demonstrate that the proposed method maintains low query latency while improving both recall rate and F1-score compared to single recall models. These results validate the algorithm’s effectiveness in significantly optimizing the efficiency and accuracy of PV data retrieval in particle accelerator environments, providing strong technical support for the intelligent operation and maintenance of large scientific facilities.