What the data for this category looks like
Knowledge base data for refinery and chemical production mainly comes from unit operation manuals, equipment maintenance handbooks, process parameter logs, safety compliance documents, and real-time process alarm logs. Data updates follow unit modifications, regulatory adjustments, or the release of quarterly inspection reports. There are two document structure types. One type is long-text operation guides, which include fields such as equipment number and maintenance cycle. The other type is structured parameter tables, which include fields such as temperature, pressure, and flow rate. Industry standard units like MPa, ℃, and m³/h are used for these parameters.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
The two data types for refinery and chemical production impose multiple constraints on retrieval and recall. Store long-text operation guides segmented by process unit to prevent accidental recall of cross-unit content. Configure structured parameter tables to support field-level matching, ensuring searches for unit-bearing parameters like temperature and pressure associate correct units. Real-time process alarm logs update frequently, so enable incremental sync logic to maintain data timeliness. When recalling mixed multi-source data, assign higher weight to safety compliance documents to prioritize matching compliance-related inquiries.
Configuration Settings
This configuration is built for FastGPT v4.17.0, and meets knowledge base retrieval and recall needs for refinery and chemical production scenarios.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches professional content for refinery and chemical production, filters low-relevance recall results |
limit | Top 8–12 entries | Controls the number of returned results, adapts to the context capacity of single-round inquiries |
searchMode | vector+keyword | Balances keyword matching for structured parameter tables and semantic recall for long-text operation guides |
usingReRank | Enabled | Performs secondary sorting on initial recall results to optimize recall accuracy for long-text operation guides |
datasetSearchUsingExtensionQuery | Enabled | Expands the retrieval scope of professional terms to cover industry-specific vocabulary for refinery and chemical production |
collectionIds | Specified by process unit | Limits the retrieval scope to avoid recall of irrelevant data across process units |
Three Common Misconfigurations
- An error with status code 514 is returned when calling the knowledge base retrieval API. The cause is incorrect configuration of the
collectionIdsparameter, or the specified knowledge base collection has not completed initialization. - Irrelevant operation guides across process units are included in recall results. The cause is failure to limit
collectionIdsby process unit, so the retrieval scope covers non-target scenario data. - Structured parameter table retrieval results match incorrect unit values. The cause is failure to enable the keyword matching mode of
searchMode, relying solely on semantic retrieval which cannot associate fields with unit information.
How to Confirm Configuration Is Correct
- Initiate a parameter inquiry for a specific refinery equipment, verify that the similarity scores of returned results fall within the preset range.
- Call the retrieval interface for the specified process unit
collectionIds, confirm all recall results originate from knowledge base data of the target process unit. - After enabling
usingReRank, compare the initial recall list with the reordered result order, confirm high-matching content is displayed first. - Test inquiries containing industry-specific terms, confirm retrieval results cover the corresponding professional documents.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.