What Data for This Category Looks Like
Data sources for petroleum refining scenarios cover DCS/APC control systems, LIMS laboratory information management systems, tank farm metering systems, crude oil evaluation databases, equipment management systems, and energy management systems. Daily documents include operating procedures, production plans, material balance sheets, distillation outlet quality reports, interlock ledgers, shift handover records, change management documents, and more. Distillation outlet quality reports contain quantifiable fields with units such as density, sulfur content, and distillation range temperature. Interlock ledgers record interlock tag numbers, trigger thresholds, and action logic. Data update frequencies fall into three categories: real-time (DCS process parameters), shift-based (shift handover records), and daily (material balance and quality reports). Data is maintained collaboratively by production operations, process, scheduling, and other departments. Data structures cover both structured tables and unstructured text.
Constraints Imposed by These Characteristics on Tool Calling and Plugins
Multi-source heterogeneous data sources and differentiated update frequencies require tool calling to adapt to different pull intervals and permission scopes. Real-time process data requires short-cycle calling. Shift reports and daily documents support batch pulling. Diverse document types require plugins to support both structured field extraction and unstructured text parsing. This prevents missing tag information from interlock ledgers or key exception descriptions in shift handover records. Multi-unit coupled business logic requires tool calling to associate parameters from multiple systems. For example, crude oil property data must link load configuration parameters from multiple units. It also requires unifying unit formats across different systems to prevent calculation errors caused by unit mismatches. Additionally, the rigor of inspection and process record audit requires tool calling to include data traceability information. This ensures the audit link can trace back to the original document source.
Recommended Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | Set to auto | Adapts to multi-type tool calling needs in refinery scenarios, covering laboratory data query, process parameter pulling, document parsing, and other business scenarios |
| Single tool call timeout | 300–600 seconds | Matches the time characteristics of DCS real-time data pulling and crude oil evaluation report parsing, avoids call interruptions due to large data volume |
| Tool return result limit | Top 8 entries | Conforms to the core business data volume characteristics of refinery scenarios, avoids context overload affecting the model's judgment of key parameters |
| Workflow/API function-call configuration | Enable forced mode (when processing structured documents) | Ensures the accuracy of field extraction for structured documents such as interlock ledgers and material balance sheets, avoids the model generating unexpected results |
| Workflow/API function-call configuration | Set to none (when processing unstructured documents) | Adapts to natural language parsing needs for unstructured text such as shift handover records and operating procedures |
| Plugin permission scope | Only open tool interfaces related to laboratory and production scheduling | Matches the data permission division of refinery enterprise departments, prevents unauthorized access to sensitive systems such as energy and equipment management |
| FastGPT version | FastGPT v4.17.0 | This version fixes compatibility issues with parallel plugin calls, adapts to multi-tool linkage needs in refinery scenarios |
Three Common Misconfigurations
- Returned results after calling tools do not include the referenced file name. Cause: The result traceability parameter was not enabled in the plugin configuration, and the storage path and file name information of document metadata were not associated.
- After enabling
functionCallforced mode, field missing occurs in the parsing results of unstructured documents. Cause: ThefunctionCallforced mode was not turned off for unstructured scenarios, and the model was restricted to only extracting structured fields. - Using an application key to call the knowledge base query interface returns a permission error. Cause: The application key only supports in-application API calls, cross-interface call permissions were not configured, and the request was not initiated through a legal authentication process.
How to Verify Proper Configuration
- Initiate a tool call request for distillation outlet quality reports, check if the returned results include the document file name and corresponding quantifiable field information.
- Trigger tool calls for structured and unstructured documents separately, confirm that the parsing results in both scenarios conform to the expected format.
- Trigger a test call for a custom plugin, confirm that the interface returned data is consistent with the actual data of the target LIMS system or tank farm metering system.
- Use the application key to initiate an out-of-application API request, verify that a normal response other than a permission error is returned.
Note: The configuration values listed above are common starting points. They should be evaluated against the reader's own samples.