This document is formulated based on the interface and parameter logic of FastGPT v4.17.0.
What the data for this category looks like
Petroleum refining business data covers production operations, process, scheduling, equipment, and laboratory teams. Daily documents include operating procedures, production plans, material balance sheets, distillate export quality reports, interlock ledgers, shift handover records, and change management documents. Data sources include DCS/APC systems, LIMS laboratory management systems, tank farm metering systems, crude oil evaluation databases, equipment management systems, and energy systems. Updates occur in real time for operating condition data, after each shift for handover records, and daily for material balance reports. Fields include process tag numbers, distillation range values, interlock trigger conditions, shift information, and some data includes clear physical units.
Constraints on vector models and indexing
Multiple heterogeneous data sources require indexes to support mixed-format document parsing. This includes structured LIMS reports and unstructured operating procedures. Real-time updated DCS data requires indexes to support incremental synchronization. This avoids performance loss from full reindexing. Coupled relationships across multiple continuous units require vector retrieval to associate process documents across units. Indexes must retain association metadata between documents. Large fluctuations in raw material properties require vector matching thresholds to adapt to text features of different operating conditions. Frequent updates to shift handover records and change documents require controlling the number of documents and processing time per batch of indexing.
Configuration settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapt to the single-paragraph content length of refining process documents, avoid losing process association information across chunks |
| Overlap Length | 100–150 characters | Retain association of process parameters and operation steps across chunks, ensure complete process logic is covered during retrieval |
| Similarity Threshold | 0.75–0.85 | Match the rigor requirements of refining processes, filter low-confidence irrelevant documents |
| Single Search Limit | Top 20 entries | Control the amount of information in search results, adapt to the associated display needs of multi-unit data |
| Maximum Context | 4000 token | Accommodate associated content from multiple process documents and real-time data, meet context requirements for quality traceability |
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. |
Three common configuration errors
Phenomenon 1
A shift handover record file always appears in the index without completing indexing. Cause: The file contains a large number of meaningless blank lines and repeated line breaks, triggering an abnormal loop during chunk processing.
Phenomenon 2
A 60-second timeout error occurs when switching the knowledge base index. Cause: The knowledge base contains real-time DCS data from multiple units, and the data volume exceeds the processing threshold for a single index.
Phenomenon 3
A prompt about mismatched embedding dimensions is returned when calling bge-large-zh-1.5. Cause: The vector index parameters are not configured to match the model's embedding dimensions, or the dimensions returned by the third-party API do not match the platform requirements.
How to verify correct configuration
- Upload a typical distillate export quality report, check whether the index completion status updates within a preset reasonable time.
- Initiate a quality traceability query for an abnormal operating condition of a specific unit, check whether the returned search results include the operating procedures, interlock ledgers, and LIMS laboratory data of the corresponding unit.
- Adjust the similarity threshold, verify whether the matching degree of search results conforms to the business judgment logic of refining processes.
- Test the
defaultTokenconfiguration parameter, confirm that the embedded vector dimensions match the storage requirements of the index library.
The values provided are common starting points and should be measured against the reader's own samples.