What the Data for This Category Looks Like
Data for the basic chemical continuous production category comes from DCS/SIS systems, online analyzers, LIMS laboratory management systems, equipment condition monitoring devices, energy metering devices, and raw material storage tanks. Daily retained documents include operating procedures, process index cards, DCS trend charts, laboratory reports, interlock bypass records, special work permits, and shift logs. Data update rhythms are divided into real-time, per shift, and daily. Most document structures are structured parameter tables and time-series record texts. Fields include process parameter values, equipment status indicators, work approval status, and similar items. Corresponding units include megapascals, degrees Celsius, tons, and others.
Constraints on Knowledge Base Retrieval and Recall
Real-time and high-frequency updated data sources require retrieval and recall to support fast matching of recent data, to avoid interference from stale data during anomaly analysis. Multiple types of structured documents such as process parameter tables and time-series records require retrieval to use a combination of field-level exact matching and full-text semantic recall. This balances parameter accuracy and semantic relevance. Documents cover the full process of process, equipment, and operations. Retrieval scope must cover corresponding business subsets to avoid mixing in off-category irrelevant documents. Large differences in document volume require adapting chunking and recall logic for different text lengths. This ensures short documents such as key interlocks and work permits are not truncated and lose core information.
How to Set Configurations
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Similarity Threshold | 0.75–0.85 | Basic chemical process parameters have strong correlation. Low-related results must be filtered to ensure recall accuracy of core process and interlock information |
| Single Search Limit | Top 8–12 results | This category’s business documents cover multiple links of process, equipment, and operations. Sufficient results must be returned to cover key businesses, avoiding omission of core anomaly records |
| Search Mode | Semantic + keyword hybrid | Adapt to the exact matching requirements of structured process parameters and semantic recall requirements of unstructured operation records |
| Reranking Switch | Enabled, rerank top 20 candidates | After mixed recall of multiple types of documents, reranking must be used to filter irrelevant long time-series trend data, focusing on key process indicators and anomaly records |
| Query Optimization Switch | Enabled | Extending process parameter keywords can improve recall coverage, adapting to multi-parameter correlation scenarios in basic chemical production |
| Collection ID Binding | Bind specified collection IDs by business domain | Classify process, equipment, and operation documents into corresponding collections to avoid cross-domain retrieval interference |
Three Common Configuration Mistakes
- Phenomenon: Table-format documents configured in the knowledge base do not display tables in retrieval result outputs. Cause: Rendering adaptation for structured text is not enabled. Most process index cards and laboratory reports in basic chemical production are in table format. Default recall only extracts plain text, causing table structure loss.
- Phenomenon: Retrieval results return irrelevant documents across business domains, such as mixing raw material storage tank data into retrieval results for process interlocks. Cause: Specified collection IDs are not bound by business domain, leading to retrieval scope covering irrelevant knowledge base collections.
- Phenomenon: Uploading large-volume documents takes too long, and independent selection to prioritize small document uploads or pause upload tasks is not supported. Cause: Document upload segment scheduling and manual control functions are not enabled. The default setting processes documents in upload order.
How to Confirm Correct Configuration
- Upload a process index card table document, initiate a retrieval, and check whether the output result retains the table structure.
- Bind the collection ID of a specified business domain, initiate a query for the corresponding business, and check that returned results only cover documents in this business domain.
- Enable the reranking and query optimization switches, initiate a query containing core process parameters, and check that the sorting of returned results matches business priorities.
- Upload multiple documents of different volumes, confirm that upload priorities can be adjusted or upload tasks can be paused.
The configuration values provided here are common starting points, and should be evaluated against the reader’s own samples.