What the Data for This Category Looks Like
Data for the iron and steel smelting and continuous casting scenario comes from hot metal and scrap steel collection, L2 process control systems, laboratory testing, MES production management systems, energy medium monitoring, and equipment condition monitoring systems. Daily generated documents include heat plans, batching lists, converter smelting records, composition reports, continuous casting process records, and slab judgment sheets. Data updates occur per heat, real-time, or per shift. Most documents carry unique fields such as heat number, cast number, composition content, and process parameters.
What Constraints Do These Characteristics Impose on the "Citation Sources and Traceability" Link
The multi-source heterogeneous data, high-frequency update rhythm, and strong production-binding attributes of this category bring multiple constraints to citation traceability. First, data originates from multiple independent systems. Unique identifiers such as heat number and cast number must be used to bind traceability, preventing incorrect cross-heat and cross-cast associations. Second, the real-time or per-heat update frequency requires that traceability citations must be synchronized with the latest data from production links. Otherwise, judgment errors based on outdated data will occur. Third, unlike other process industry categories, this category uses a high-temperature continuous process, with strong coupling between heats and casts. Arbitrary shutdown adjustments are not possible. Therefore, traceability configuration must be lightweight, and must not occupy production system resources.
How to Set the Configuration
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
quoteMaxToken | 8000–12000 token | Single document data volume for iron and steel smelting and continuous casting is not large, but after multi-source splicing, the context must be controlled within the model's upper limit. A reasonable range is set based on the high-frequency citation characteristics of this category |
| Similarity Threshold | 0.75–0.85 | Document fields for this category are precise. An overly high threshold will miss matching key process parameters, while an overly low threshold will introduce irrelevant heat data. Matching logic must be optimized for unique fields |
responseData | Return original document metadata | Unique fields such as heat number, detection time, and source system must be retained to meet traceability and compliance requirements for production links |
| Single Search Limit | Top 3–5 entries | The number of documents related to a single heat is limited for this category. Too many citations will increase the model inference load, adapting to the compact production rhythm |
detail | Enable detailed traceability | Specific document sources and update times must be displayed to allow production personnel to quickly locate data from corresponding production links |
Three Common Misconfigurations
- Phenomenon: The number of citation sources returned by knowledge base search is fixed and cannot be adjusted, failing to cover all relevant documents for a single heat. Cause: The single search limit or
quoteMaxTokenparameter is not configured correctly, or the default fixed citation limit is used, without adapting to the characteristics of multiple documents per single heat for this category. - Phenomenon: Target composition reports or smelting records still cannot be matched after setting a low similarity threshold. Cause: Mandatory matching conditions for unique fields such as heat number and cast number are not added, and only general text similarity matching is used, leading to irrelevant heat data interfering with matching results.
- Phenomenon: No cited document sources are displayed in AI responses, making it impossible to trace back to specific converter smelting records or slab judgment sheets. Cause: The
responseDataconfiguration item is not enabled, and the switch to return original document metadata is not configured, resulting in missing traceability information.
How to Confirm the Configuration Is Correct
- Initiate a query for a known heat number, check whether the returned citation sources include exclusive documents for that heat, to verify matching accuracy.
- View the citation section of the AI response, confirm whether metadata such as document source system and update time is included, to verify complete traceability information.
- Adjust the value of
quoteMaxToken, observe whether the returned context length meets expectations, and confirm that the parameter takes effect. - Test the query scenario for multiple documents related to a single heat, confirm that the number of returned citation sources matches the configured single search limit.
The values provided are common starting points and should be measured against the reader's own samples.