What the Data for This Category Looks Like
Daily documents for this category include operation procedures, production plans, material balance reports, distillation outlet quality reports, interlock ledgers, shift handover records, and change management documents. Data sources cover DCS/APC systems, LIMS laboratory systems, tank farm metering systems, crude oil evaluation databases, equipment management systems, and energy management systems. Data update frequencies include real-time, per shift, and daily. Most documents are long-form PDFs or structured reports. Fields include unit number, material composition, control parameters, interlock thresholds, and other items. Units use standard industrial units such as temperature, pressure, flow, and component content.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
Multiple continuous units coupled via logistics and hydrogen energy networks require retrieval to support cross-unit process parameter correlation matching. This avoids returning one-sided single-unit data. Long-form documents and frequently updated data require retrieval to adapt to large chunk recall logic. It must also support fast synchronization of incremental updates. Large fluctuations in raw material properties require retrieval to prioritize matching data associated with current crude oil evaluation and unit load. This filters historical content that does not match current operating conditions. Multi-department collaboration scenarios require retrieval to filter corresponding documents by post permissions. It must also ensure information consistency during cross-department collaboration.
Configuration Recommendations
This configuration is adapted for FastGPT v4.17.0. Below are knowledge base retrieval and recall configuration recommendations for petroleum refining scenarios:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for refining process parameters, filters low-relevance general process descriptions, and retains core interlock and load control content |
limit | Top 8 entries | Adapts to multi-unit coupled business scenarios, covers cross-unit associated data, and avoids one-sided single results |
searchMode | mixedRecall | Combines keyword and semantic retrieval, adapts to structured process parameter reports and unstructured operation procedure documents |
usingReRank | Enabled | Reranks chunks of long documents, corrects local deviations from semantic retrieval, and improves ranking accuracy for interlock threshold and material balance content |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands associated queries for process questions. For example, expands "atmospheric tower load" to "atmospheric tower heater temperature, tower top pressure, crude oil feed rate", adapting to multi-parameter correlation logic in refining processes |
collectionIds | Collection IDs divided by business department | Divides knowledge base collections by production operation, process, scheduling and other departments, implements post permission filtering, and complies with refinery enterprise department data isolation requirements |
Three Common Misconfigurations
- Phenomenon: Process flow diagram images in uploaded PDF documents within the knowledge base fail to load properly in retrieval results. Cause: Knowledge base image indexing configuration is not enabled, or PDF vector image formats are not correctly parsed during upload. This results in only text content being extracted during retrieval.
- Phenomenon: The number of returned results when calling the RAG knowledge base in a workflow does not match the set
limitparameter. Cause: The specifiedcollectionIdsare not correctly bound. This leads the retrieval scope to cover irrelevant enterprise documents, exceeding the expected result count. - Phenomenon: Expired historical data is returned when searching for crude oil evaluation related questions. Cause: Incremental update trigger rules are not configured. The knowledge base does not synchronize real-time crude oil property data from DCS systems, resulting in recalled content that does not match current production operating conditions.
How to Verify Correct Configuration
- Upload a long PDF document containing a process flow diagram, initiate a corresponding process retrieval, and confirm that the image displays normally in the retrieval results.
- Initiate a query for cross-unit load coordination, and confirm that the returned results include associated process parameters from at least 2 units.
- Adjust the
similarityparameter, verify that low-similarity irrelevant content is filtered out, and highly relevant interlock ledger content is displayed first. - Check the knowledge base update log, and confirm that the data synchronization frequency matches the business requirements of real-time, per shift, or daily rhythm.
The values provided are common starting points and should be measured against the reader's own samples.