What This Category’s Data Looks Like
Core entities of this category are manufacturers of battery packs, energy storage cabinets, and BMS systems. Daily documents include grouping rules, BOM lists, welding torque records, BMS flashing records, insulation withstand voltage reports, EOL test data, and system configuration lists. Data sources include cell capacity grading data, MES systems, welding equipment, torque tools, BMS software version libraries, and EOL test benches. Each PACK or energy storage cabinet generates a unique record when produced, and documents are updated synchronously each time a BMS software version changes. Document fields include cell capacity parameters, torque values, insulation resistance, BMS configuration items, and other exclusive information for different process links.
What Constraints Do These Characteristics Create for Multi-turn Dialogue and Prompt Engineering
Each PACK or energy storage cabinet generates and updates its own exclusive data independently. Multi-turn dialogue requires binding a unique identifier to distinguish the conversation context of different products, to avoid cross-batch data confusion. Document types cover multiple process links including grouping, welding, and BMS flashing. Prompts need to clearly specify the document type to retrieve, to ensure retrieved content matches the current query’s process step. Data update frequency is tightly bound to production progress. Conversation context should not retain old data beyond the current production batch, to prevent use of outdated parameters. Process questions often involve multi-step verification. Multi-turn dialogue must support breaking down questions step by step and supplementing necessary parameter information.
How to Set Configurations
This page’s configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Generate a unique ID per single PACK/energy storage cabinet | Distinguish the conversation context of different products, avoid cross-batch data interference |
| Maximum context window | 8000–12000 token | Adapt to long text content such as process specifications and test records accumulated during multi-turn dialogue |
| Single search limit | Top 5 results | Match the number of exclusive documents for a single device, avoid redundant retrieval of irrelevant content |
variables | Preset product type (battery pack/energy storage cabinet) and BMS version | Allow prompts to accurately match process rules and version requirements for the corresponding category |
| Question optimization | Enabled | Break down user process questions and supplement necessary parameter verification steps |
messages | Retain the most recent 2 rounds of historical messages | Control context length, avoid impact of old non-current batch data affecting current responses |
Three Common Mistakes
- Phenomenon: After increasing the single search limit, the large language model does not reference any retrieved content. Cause: The prompt does not explicitly require answering based on retrieved process documents, and only general question-answering logic is used.
- Phenomenon: Conversation data from different PACKs or energy storage cabinets is mixed together. Cause: The
chatIdconfiguration is not set to bind the conversation context of a single device, resulting in shared historical messages for different products. - Phenomenon: Streaming output has no response or interrupts early. Cause: The streaming output configuration item is not enabled, or the streaming transmission threshold for large text such as BMS flashing records is not adapted.
How to Confirm the Configuration Is Correct
- Initiate a query about insulation withstand voltage test for a single energy storage cabinet, check if the response includes the insulation resistance parameters corresponding to that device.
- Switch to different
chatIdto initiate conversations, confirm that process query results for different products do not interfere with each other. - Adjust the single search limit, verify that the number of retrieved results matches the set range.
- Enable the streaming output function, verify that the response is returned step by step in segmented form.
The values given are common starting points and should be measured against the reader's own samples.