What the data for this category looks like
The core data for this category comes from concentrate and auxiliary material analysis reports, cell voltage and current data collected by DCS systems, composition test results issued by laboratories, energy and environmental protection monitoring ledgers, and manual records from smelting and electrolysis processes. Daily generated documents include batching lists, smelting logs, electrolytic cell condition records, anode and cathode quality reports, test reports, and environmental protection ledgers. Data updates proceed in real time alongside raw material batches, single-furnace or single-tank operations. Individual records contain fields such as raw material composition, operating parameters, quality indicators, and pollutant emission values, covering multiple types of professional units.
Constraints on knowledge base retrieval and recall
This category of data has strong real-time performance and is split by operation batches. Knowledge base retrieval must quickly respond to real-time condition queries for single furnaces or single tanks. Multi-source heterogeneous data includes professional process parameters such as concentrate impurity composition, cell voltage and current, and metal recovery rate. The retrieval model must adapt to specialized terminology for non-ferrous metal smelting and avoid matching irrelevant general process content. Individual records are associated with multi-dimensional operating indicators. Recall results must cover the context of associated fields to avoid fragmented information affecting process judgment. Operational errors will directly lead to reduced metal recovery rates or exceedance of environmental protection standards. The matching accuracy of retrieval results must meet the rigor requirements of process decision-making.
Configuration settings
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the professional accuracy requirements for matching process parameters of this category. Too low will introduce irrelevant results, too high will miss valid matching content |
limit | Top 8–12 entries | Single process records have many associated fields. Too many entries will exceed the context window limit, too few will fail to cover complete information about associated operating conditions |
searchMode | embedding | Matches based on semantic and numerical characteristics of process parameters, adapting to multi-source heterogeneous professional data scenarios |
usingReRank | Enabled | Associated features of multi-source data require re-ranking to optimize relevance sorting, improving the accuracy of process decision-making |
datasetSearchUsingExtensionQuery | Enabled | Process questions often include specialized abbreviations or term variants. Extended queries can cover more associated historical operating documents |
collectionIds | Bind the knowledge base collection ID corresponding to the business | Isolate process documents from other categories to avoid retrieval scope exceeding the non-ferrous metal smelting and electrolysis business scope |
Three common configuration mistakes
- Symptom: An abnormal parameter prompt is returned when the workflow calls knowledge base search. Cause: The
collectionIdsparameter is not specified correctly, or the bound knowledge base collection does not include the required smelting and electrolysis process documents, resulting in a verification error during retrieval. - Symptom: An execution error is returned during the knowledge base question answering splitting stage. Cause: The field association characteristics of long documents of this category are not adapted, and the process context across fields is truncated during chunking, triggering a format verification failure.
- Symptom: A 503 error is returned when uploading process documents via the dialog box. Cause: Reasonable upload thresholds are not set for large-volume batch documents of this category, or the knowledge base indexing service is experiencing temporarily high load, resulting in request timeout.
How to verify successful configuration
- Manually enter a known process question and check whether the retrieval results include matching business documents.
- View the actual values of the configuration items and verify that they comply with the preset professional matching and recall rules.
- Trigger a workflow call and confirm that the number of returned results and matching degree meet business requirements.
- Upload a single electrolytic cell condition record and verify that the upload and indexing processes are completed normally.
The values provided here are common starting points and should be evaluated against the reader's own samples.