What the data for this category looks like
Data for this category originates from ERP orders and inventory, BOMs and process routes, equipment capacity, staff shifts, work-in-progress, and supplier deliveries. Updates run daily, weekly, or trigger automatically when orders or exceptions occur. Common documents include sales orders, master production schedules, material requirements plans, weekly and daily scheduling sheets, work orders, material shortage lists, and delivery change notices. Fields include order numbers, material codes, capacity values, shift time periods, work order statuses, and additional relevant fields. Units include basic production units such as pieces, hours, and days.
Constraints on knowledge base retrieval and recall for this category
Multi-source, heterogeneous data for this category requires retrieval to accurately link to the matching knowledge base collection, to avoid including non-production planning unrelated content. Data sources with different update frequencies need paired synchronization strategies to ensure recalled data is timely. Diverse document structures require tailored splitting and parsing rules; for example, scheduling sheets with merged cells need special handling. This category requires tracing quality problems by work order or order, so retrieval must support filtering by business entities. A balance must be maintained between the number of recalled entries and the 4096 token context window limit to avoid exceeding processing capacity.
Configuration Settings
This configuration applies to FastGPT v4.17.0. A single knowledge base supports a maximum of 1000 MB of document storage. Specific parameter settings are listed below:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
collectionIds | List of collection IDs for knowledge bases related to production planning and shop scheduling | Data for this category comes from independent sources such as ERP orders, BOMs, and equipment capacity. Accurate linking to matching knowledge base collections avoids retrieving unrelated content |
similarity | 0.72–0.85 | Documents for this category include structured numerical values and business rules. A threshold that is too low will introduce unrelated scheduling exception records, while a threshold that is too high will fail to recall historical defect data associated with relevant work orders |
limit | Top 8–12 entries | Abnormally associated data for shop scheduling is usually distributed across multiple work orders and scheduling sheets. This range covers relevant dimensions without exceeding the 4096 token context window limit |
searchMode | fullTextRecall | The query examples emphasize identifiers or specialist terms; test this mode with representative data, not as a product default. |
usingReRank | Enabled | This category contains a large number of resource constraints and process rules. Reranking filters low-correlation records and improves recall accuracy |
datasetSearchUsingExtensionQuery | Disabled | Business rules for this category are relatively fixed. No additional query term expansion is needed, to avoid introducing non-production planning unrelated data |
Three Common Configuration Mistakes
- An
Invalid array lengtherror occurs when configuring document splitting. This happens when the structured characteristics of production planning documents are not accounted for; for example, directly splitting a weekly scheduling sheet with merged cells using a fixed length, which causes an abnormal array format during parsing. - The knowledge base can be retrieved normally during debug preview, but retrieval fails after deploying to a custom page. This is caused by failing to correctly specify the
collectionIdsparameter, or failing to include a valid identifier for the associated production planning knowledge base when making the call. - The number of retrieved knowledge base results does not match expectations, with only a small number of scheduling records returned. This is caused by not setting a reasonable
similaritythreshold, or not enablingusingReRank, which allows low-correlation records to overwrite valid recalled content.
How to Confirm Successful Configuration
- Verify that the
collectionIdsparameter exactly matches the collection IDs of the currently bound production planning and shop scheduling knowledge bases. - Initiate a retrieval for a specific work order number, and confirm that the returned results include relevant scheduling, material requirement, and other document content for the corresponding work order.
- Check the relevance ranking of retrieval results, and confirm that the relevance of results after enabling
usingReRankaligns with business expectations. - Initiate a retrieval in a simulated exception scenario, such as "Trace scheduling abnormalities caused by emergency order insertion", and confirm that the recalled content matches business rules.
The values provided here are common starting points, and should be evaluated against the reader's own samples.