What the data for this category looks like
The core documents for this category include grouping rules, BOM lists, welding and torque records, BMS flashing records, insulation and pressure resistance reports, EOL test reports, and system configuration lists. Data is sourced from cell grading systems, workshop MES systems, welding equipment, torque tools, BMS software version repositories, and EOL test benches. Data updates use a single battery pack or energy storage cabinet as the minimum unit, and are synchronized each time software versions are updated. Document fields include cell parameters, torque values, test voltages, software version numbers, and others. Units include milliampere-hours, newton-meters, volts, and more.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
The data characteristics of this category create multiple constraints for knowledge base retrieval and recall. Dispersed multi-source data sources require the retrieval link to support connecting documents exported from different systems. A cross-data-source knowledge base set must be configured. The high-frequency, small-batch update mode requires enabling incremental parsing first to avoid resource waste from full re-scanning. Documents contain professional engineering parameters and units, so retrieval must retain exact field matching capabilities to prevent parameter misalignment caused by fuzzy matching. Exclusive records bound to a single device require that recall results can be filtered by device serial number or batch number. This ensures that compliant data for the corresponding batch is called when drafting process documents or exception reports.
The following configurations are based on FastGPT v4.17.0.
How to set the configurations
Choose a retrieval mode through representative-query testing; configure category, batch, or entity filters separately because the mode itself does not enforce those boundaries.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Most documents for this category are professional engineering content. A high similarity threshold is needed to avoid irrelevant grouping rules and test records from being included in retrieval results |
limit | Top 8 entries | Balances comprehensiveness and readability for process documents and exception reports. Too many results will increase the workload of context organization |
usingReRank | Enabled | Results recalled from multi-source knowledge bases have varying relevance. Re-ranking can improve the display priority of core documents |
datasetSearchUsingExtensionQuery | Enabled | Process drafting requires associating grouping rules and test records for the corresponding batch. Extended queries can supplement associated context |
collectionIds | Knowledge base set ID corresponding to battery PACK and energy storage system integration | Documents for this category are scattered across multiple knowledge bases. Specifying the set ID can avoid recalling content from unrelated categories |
Three common mistakes
- Phenomenon: Retrieval results are not filtered by preset tags, and documents related to cell manufacturing are mixed in. Cause: No tag filtering logic is configured during the retrieval link, or no classification tags exclusive to battery PACK are added when uploading documents.
- Phenomenon: The number of knowledge base recall results does not match the configured
limit, with too many or too few results. Cause: Thelimitparameter is not set correctly, or the number of valid documents in the knowledge base is insufficient, leading to unexpected recall results. - Phenomenon: Actively pushed documents cannot be retrieved. Cause: The parsing and slicing process for pushed documents is not completed, or the documents are not associated with the specified
collectionIdsset.
How to confirm the configuration is correct
- Upload a dedicated test document for this category, add preset classification tags, initiate a retrieval, and verify that only content with the corresponding tags is recalled.
- Check the knowledge base management interface to confirm that actively pushed documents have completed parsing and slicing and are associated with the specified knowledge base set.
- Adjust the
similarityparameter to verify that the relevance of retrieval results changes as expected with the parameter. - Initiate a simulated call for process document drafting, and verify that the recalled documents include category-specific content such as grouping rules and BOM lists.
The values provided are common starting points and should be evaluated against samples appropriate for the specific deployment.