What the Data for This Category Looks Like
Knowledge base data for petroleum refining scenarios comes from DCS/APC control systems, LIMS laboratory systems, tank farm metering systems, crude oil evaluation reports, equipment management systems, and energy management systems. Daily documents include operating procedures, production plans, material balance sheets, distillate export quality reports, interlock ledgers, shift handover records, and change management documents.
Data update cycles cover real-time, per-shift, and daily. Document structures include multiple types of fields such as process parameters, metering data, and ledger records, with no unified standardized format. Multiple continuous production units are coupled through logistics and energy networks, and the fluctuation range of raw material properties is large.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
Multi-source and heterogeneous data sources require retrieval to adapt to document formats and field rules of different systems. This prevents irrelevant equipment management documents from being recalled during process quotation preparation.
Real-time, per-shift, and daily update cycles require retrieval to prioritize recalling the latest crude oil evaluation and material balance data. This ensures the timeliness of bid quotations.
The business logic of multi-unit coupling requires retrieval to associate load data and logistics information of different units. This avoids quotation deviations caused by recalling isolated single-unit data.
The lack of unified document formats requires retrieval to cover both semantic matching and keyword matching. This accurately locates process parameters and quality indicators.
Configuration Settings
The following configurations are formulated based on FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Petroleum refining data includes clear process parameter thresholds. This range filters low-match irrelevant documents and ensures the relevance of retrieval results |
limit | Top 8–12 entries | Bid documents need to cover business data from multiple units and links. This value range balances the comprehensiveness of retrieval results and the context length limit of 500–1000 characters per document on average |
searchMode | mixedRecall | Adapts to multiple types of fields in refinery documents, matching precise keywords of process parameters and semantic associations of business logic |
usingReRank | Enabled | Multi-source data has similar expressions. Reranking prioritizes recalling core process and quality indicator documents related to bid quotations |
datasetSearchUsingExtensionQuery | Enabled | Bid requirements often include associated queries about crude oil properties and unit loads. Optimized queries cover more relevant shift handover records and change management documents |
collectionIds | Bind corresponding collection IDs by production, laboratory, and equipment modules | Refinery enterprise knowledge bases are split by business departments, accurately recalling documents from corresponding business links and avoiding mixing irrelevant data |
Three Common Misconfigurations
- Symptom: Calling the FastGPT chat interface with a specified knowledge base returns a large number of irrelevant documents. Cause: The
collectionIdsparameter is not configured correctly, and cross-business module knowledge base collections are bound, resulting in equipment management documents being mixed into process quotation retrieval. - Symptom: Importing knowledge base content from web links fails, returning a 401 status code. Cause: No web crawling login credentials are configured, making it impossible to access internal permission-protected LIMS system document links at the refinery.
- Symptom: Retrieval request latency exceeds 60 seconds. Cause: The value of the
limitparameter is not reasonably restricted, and too much cross-unit historical data is recalled, exceeding the interface processing limit.
How to Verify Successful Configuration
- Call the FastGPT knowledge base retrieval interface, pass a query term related to crude oil evaluation, and check whether the returned results include documents corresponding to crude oil properties.
- View the knowledge base import log to confirm that protected internal web links have been successfully crawled and imported into the knowledge base.
- Verify that when calling the API, the
collectionIdsparameter is specified, and confirm that only documents from the bound business module are returned, with no cross-module irrelevant content. - Test the sorting of retrieval results after enabling
usingReRank, and confirm that documents with core process parameters are ranked at the top of the retrieval results.
The values provided here are common starting points and should be measured against the reader's own samples.