What this category of data looks like
This category of data comes from basic hot metal and scrap steel data, L2 process control data, laboratory composition reports, MES systems, energy medium systems, and equipment condition monitoring data. Daily documents include heat plans, batching lists, converter smelting records, continuous casting process records, slab judgment sheets, and more. Fields cover heat numbers, cast numbers, composition content, temperature parameters, equipment operating duration, and more. Units include degrees Celsius, tons, percentages, millimeters, and others. Data updates occur per heat, in real time, and via shift-based archiving. The process runs as high-temperature continuous operation, with tight coupling between heats and casts. Arbitrary shutdown adjustments are not possible.
What constraints do these characteristics impose on knowledge base retrieval and recall?
The data is updated per heat, in real time, and per shift. This requires knowledge base retrieval and recall to support low-latency responses, to avoid using expired ledgers that disrupt real-time scheduling. Data comes from multiple heterogeneous sources. This requires retrieval to support cross-system field association matching, such as linking smelting records and composition test reports corresponding to the same heat number. The high-temperature continuous process has tight coupling between heats and casts. This requires retrieval results to cover associated data from both steelmaking and continuous casting links, to avoid report deviations caused by missing data from a single link. Professional document fields are numerous and parameters are complex. This requires retrieval to accurately match metallurgical professional terminology and filter irrelevant results.
How to set the configuration
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | High metallurgical professional parameter matching accuracy requires filtering low-similarity irrelevant ledger documents, and this aligns with the similarity calculation logic of FastGPT v4.17.0 |
limit | Top 8–12 results | The volume of heat-associated data is moderate. Too many results increase ledger summarization time, while too few may miss key process parameters |
searchMode | mixedRecall | Balances keyword matching for professional terminology and semantic similarity retrieval, and adapts to multiple heterogeneous metallurgical ledger data |
usingReRank | Enabled | Secondary sorting of initial retrieval results by relevance is required to avoid result priority errors caused by mismatches in professional parameters |
datasetSearchUsingExtensionQuery | Enabled | Can expand metallurgical professional synonyms based on original retrieval terms, improving recall accuracy for exclusive identifiers such as heat numbers and cast numbers |
collectionIds | Bind exclusive dataset IDs for ironmaking, steelmaking, continuous casting, quality control, and scheduling departments | Precise restriction of retrieval scope avoids interference from cross-department irrelevant ledgers on summarization results, and aligns with the dataset binding logic of FastGPT v4.17.0 |
Three Common Configuration Mistakes
- Symptom: An error prompt appears after clicking the reading link returned by the API, stating that only .txt, .m and other formats are supported. Cause: Ledger documents uploaded to the knowledge base use incompatible formats, such as scanned PDF files or CAD process files, and have not been converted to text formats supported by the knowledge base.
- Symptom: Retrieval results include a large number of documents unrelated to smelting and continuous casting, such as administrative office documents. Cause:
collectionIdsis not configured correctly, and the retrieval scope covers datasets from irrelevant departments, leading to redundant recall results. - Symptom: The number of retrieval results returned is far lower than expected, and cannot cover all process parameters required for a heat. Cause: The
similarityparameter is set too high, filtering out some qualified low-similarity but critical ledger documents.
How to Confirm Correct Configuration
- Review the dataset binding list, confirm that it only includes exclusive datasets for ironmaking, steelmaking, continuous casting, quality control, and scheduling departments, to verify that the
collectionIdsconfiguration is correct. - Enter a specific heat number as the retrieval term, check whether the returned results include associated documents such as the batching list, smelting record, and composition report corresponding to that heat, to verify retrieval scope and matching accuracy.
- Upload a test metallurgical ledger document, initiate a retrieval, and check the format compatibility of the returned results, to verify that file processing configuration is correct.
- Simulate a real-time data update scenario, initiate multiple retrieval requests, check whether the response time meets the real-time requirements of business scheduling, to verify that the low-latency configuration is effective.
The values provided here are common starting points, and should be evaluated against the reader's own samples.