What this category of data looks like
Data for this category originates from concentrate and auxiliary material analysis reports, DCS systems, cell voltage and current monitoring data, laboratory output reports, and energy and environmental monitoring devices. Daily stored documents include batching lists, smelting records, electrolytic cell condition records, test reports, anode and cathode quality records, and environmental protection ledgers. Data updates trigger when each batch of raw materials, each furnace, or each cell job completes. Some cell condition and energy consumption data updates in real time. Document fields include raw material component proportions, furnace temperature parameters, cell voltage and current values, and metal recovery rate indicators. Common units are percentage, degrees Celsius, volts, amperes, and similar units.
Constraints on vector models and indexes
Data for this category is multi-source and updates frequently. Some cell condition and energy consumption data updates in real time. This requires vector indexes to support high-frequency incremental synchronization and low-latency retrieval. Documents contain multiple fields such as raw material component proportions, furnace temperature, and cell voltage and current, with diverse unit types. This requires vector models to adapt to mixed encoding logic for multiple fields. Document structures vary significantly. Batching lists are short field lists. Smelting records and cell condition records are long text sequences. This requires unified chunking rules to adapt to content of different lengths. Core indicators such as metal recovery rate and impurity content have high matching accuracy requirements. This ensures key abnormal data is not missed during index retrieval.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapt to long text content in smelting records and cell condition records, avoid single chunk content exceeding model input limits |
| Overlap Length | 100–150 characters | Cover key indicators across chunks, ensure continuous fields in batching lists and test reports are not truncated |
| Similarity Threshold | 0.72–0.80 | Match the accuracy requirements for core indicators such as metal recovery rate and impurity content, filter low-correlation historical records |
maxToken | 4096 token | Adapt to the input length limit of mainstream vector models, comply with the parameter specifications of FastGPT v4.17.0 |
defaultToken | Set from representative sample documents | defaultToken is the default token count for direct text chunking; tune it on representative source documents. See the v4.17.2 model-field reference. It is not a vector dimension, embedding-model input cap, or retrieval/LLM context. |
| Single Search Limit | Top 8 entries | Cover the number of associated records for a single furnace or cell, avoid retrieving too many irrelevant data |
Three Common Errors
- Attempting to connect a self-hosted Qdrant vector database results in a connection error, or the knowledge base fails to load existing data. Cause: The connection address and permission parameters of the custom vector database were not correctly configured on the FastGPT platform, and existing document vectors in the database were not synchronized.
- After enabling the Tencent Hunyuan vector model, search results are empty or have very low matching degrees. Cause: The API key and access endpoint of the model channel were not correctly configured, and the index enable switch for the vector model was not turned on.
- After configuring the index model, the system prompts a timeout or the number of retrieved results is abnormal. Cause: The index refresh cycle was not adjusted based on the real-time data update frequency, and the single search limit was set too low, resulting in key records not being retrieved.
How to Confirm Proper Configuration
- View the vector database connection logs, confirm there are no connection timeout or permission error messages, and check that the number of synchronized documents matches the actual stored quantity.
- Enter core indicator keywords such as "copper recovery rate" and "abnormal cell voltage", test that the number of retrieved search results falls within the range set by the single search limit.
- Upload a recent smelting record, verify that the system can correctly retrieve associated historical documents from the same batch of raw materials.
- Check the model configuration page, confirm that the channel parameters of the vector model and index model have been correctly filled, with no empty values or format errors.
The values given here are common starting points, and should be measured against the reader's own samples.