What Data for This Category Looks Like
The data for this category comes from design BOM, silicon steel sheet and insulating material procurement data, vacuum drying equipment logs, insulating oil test reports, and high-voltage test station records. Data is updated per single product or each batch of raw materials. Documents associated with a single product include winding and core assembly records, vacuum drying records, insulating oil reports, partial discharge and withstand voltage test reports, and nameplate parameter tables. Document fields include insulation structure parameters, cleanliness inspection items, lead connection information, tap position data, high-voltage test values, and some fields have dedicated testing units.
Constraints Imposed on Multi-turn Dialogue and Prompt Engineering
Single products are associated with multiple exclusive documents. Dialogue context must accurately match parameters of the current batch or single product to avoid mixing cross-batch data. High-frequency high-voltage test and insulation detection data have numerous professional fields. Prompts must clearly limit retrieval scope to the single document corresponding to the current task, to prevent irrelevant data from interfering with analysis. Data update frequency follows the production progress of single products. Dialogue processes must support dynamic loading of the latest test records, while limiting context length to avoid excessive token usage from old batch data. Additionally, there are many professional terms for insulation and high-voltage tests. Prompts must fix term interpretation rules to reduce ambiguity during analysis.
Configuration Settings
Based on FastGPT v4.17.0, the following configuration recommendations apply to multi-turn dialogue and prompt engineering for this category:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Maximum Context | First 20 historical dialogue turns + current task documents | Matches the volume of documents associated with a single product and standard dialogue rounds, to avoid context overflow |
| Single Search Limit | Top 3 matching documents | The number of documents associated with a single product is limited. Excessive results will interfere with professional analysis |
| Question Optimization | Enabled | Adapts to professional terms related to insulation and high-voltage tests, automatically completes implied inspection conditions |
| Knowledge Base Association Count | Configured per exclusive document group for a single product | Prevents irrelevant cross-batch knowledge bases from being included in the current task |
| Context Timeout | 120 seconds | Matches the loading and analysis rhythm of high-voltage test data, avoids long document parsing timeouts |
| Session Binding | Bind single product via chatId | Ensures messages history records only associate dialogue data for the current product, and variables stores exclusive nameplate parameters |
Three Common Configuration Errors
- Debug preview operates normally but errors occur in the formal environment, with a prompt that the knowledge base cannot be loaded. The root cause is that the knowledge base path configuration in the formal environment does not match the local deployment path, and the associated single product document group is not synchronized.
- The JSON-formatted test report entered during dialogue exceeds the character limit and cannot be parsed. The root cause is that the context character upper limit configuration is not adjusted, and the default limit does not match long-text test reports for transformers.
- Associating too many knowledge bases leads to deviations in analysis results. The root cause is that knowledge bases are not grouped and associated per single product, and insulating oil or withstand voltage test data from other batches is mixed into the current task.
How to Verify Proper Configuration
- Import all associated documents for a single transformer, verify that the number of associated knowledge bases matches the product's document list.
- Initiate a dialogue including partial discharge anomaly analysis, verify that returned results only include test data and assembly records for the current batch.
- Enter a long-text insulating oil report, verify that the dialogue can be parsed normally without character overflow errors.
- Verify that the bound
chatIdmatches the current product's identifier, and confirm thatmessageshistory records do not include data from other batches.
The values provided here are common starting points, and should be evaluated against the reader's own samples.