What the Data for This Category Looks Like
This category of data covers daily documents from production operation, process, and scheduling departments of oil refining enterprises. These documents include operating procedures, production plans, material balance reports, distillate export quality reports, interlock ledgers, shift handover records, and change management documents. Data sources include DCS/APC control systems, LIMS laboratory information management systems, tank farm metering systems, crude oil evaluation databases, equipment management systems, and energy management systems. Update cycles cover real-time operating condition data, shift handover records, and daily production reports. Most documents combine structured reports and text procedures. Fields include unit number, operating parameters, quality indicators, and interlock parameters.
Constraints on Citation and Traceability
The data characteristics of this category impose multiple constraints on the citation and traceability process. Heterogeneous multi-source data (real-time DCS operating conditions, shift handover records, daily material balance reports) requires the ability to associate scenario data from different systems during traceability. For example, process fluctuation handling requires access to both real-time operating data and that day’s quality report. Documents with different update cycles need matching retrieval strategies. Real-time data requires fast response. Historical archived documents need stable indexing. The document structure combining structured reports and text procedures requires distinguishing marking logic for field-level citations and full-text citations during traceability. For refinery processes with multiple coupled units, traceability must also associate data from logistics and energy networks to avoid disconnecting process logic between units.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Similarity threshold | 0.6–0.7 | Matches the mixed characteristics of structured fields and text procedures in refinery process documents, filters irrelevant redundant data |
quoteMaxToken | 2000–3000 token | Covers core citation content of a single process procedure or incident report, avoids truncating critical interlock parameters or quality indicators |
| Single search limit | Top 8 entries | Matches the multi-unit coupled refinery scenario, retrieves associated upstream and downstream unit data and related logs |
responseData | Enabled | Retains source fields of original cited data, facilitating traceability by associating original business data from DCS, LIMS and other systems |
detail | Enabled | Supplements update time and source system information for cited documents, meeting compliance traceability requirements for refinery scenarios |
| Reordering strategy | Sort by citation relevance | Prioritizes retrieving unit operation records and change documents most relevant to the current process issue |
This page is compatible with FastGPT v4.17.0
Three Common Mistakes
- Phenomenon: After setting the similarity threshold to 0.4, the knowledge base retrieves a large number of irrelevant non-refinery process documents. Cause: The threshold is too low, failing to filter historical records unrelated to the current process issue, reducing citation accuracy.
- Phenomenon: After uploading 100 5KB refinery metering reports, approximately half of the files fail during training. Cause: The knowledge base file format verification is not configured correctly, and the field separators of structured reports are not recognized, leading to indexing failures.
- Phenomenon: When selecting a specified large model via the AI chat component, no temperature setting button is found. Cause: The advanced parameter switch is not enabled in the component configuration, so the temperature adjustment option does not load.
How to Confirm the Configuration Is Correctly Set
- Initiate a query for process incident handling, check whether the returned results include the source system and update time of the cited documents.
- Adjust the similarity threshold, compare the number and relevance of retrieved results to confirm that matching accuracy meets expectations.
- Upload a test refinery interlock ledger file, check whether the knowledge base can correctly index and retrieve relevant content.
- View the AI chat component configuration panel, confirm that the advanced parameter switch is enabled and the temperature setting option is visible.
The values provided are common starting points and should be measured against the reader's own samples.