Data for this category
The data sources for this category include DCS/APC control systems, LIMS laboratory information management systems, tank farm metering systems, crude oil evaluation databases, equipment management systems, and energy management systems. Data update frequencies fall into three categories: real-time, shift-based, and daily. Daily documents include operating procedures, production plans, material balance reports, distillate export quality reports, interlock logs, shift handover records, and change management documents. Each document corresponds to fields such as unit parameters, material components, quality indicators, and equipment status. Units cover common industrial measurement standards including temperature, pressure, distillation range, and flow rate.
Constraints for model integration and configuration
The multi-source heterogeneous data with real-time, shift-based, and daily update rhythms requires model integration to support mixed scheduling logic for incremental synchronization and batch pulling. Diverse specialized documents and industrial fields require configuration to adapt to vector recall rules for refinery-specific terminology, to avoid semantic matching deviations. The logistics coupling characteristics of multiple continuous units require the context window to cover cross-unit associated data, ensuring coherent analysis logic. The high timeliness requirement of real-time process data requires limiting the maximum waiting time for model responses, while distinguishing call priorities for different data sources.
Recommended configuration settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Covers cross-unit material balance, shift handover records and other multi-document associated context, adapts to long-logic analysis for refinery business |
maxTokens | 4000 token | Matches the single-batch output length of distillate export quality reports and crude oil evaluation data, avoids truncating professional analysis content |
temperature | 0.3–0.6 | Balances the rigor and flexibility of process analysis, avoids generating abnormal conclusions that do not comply with refinery operating procedures |
AIPROXY_API_ENDPOINT | Intranet API address of the deployment node | Adapts to the internal network isolated deployment requirements of FastGPT v4.17.0, ensures secure data transmission |
| Workflow/API tool selection | auto | Automatically matches the tool call logic required for supplier audits and quality anomaly disposal, adapts to multi-link business scenarios |
quoteMaxToken | 1500 token | Limits the length of referenced documents, ensures recalled content focuses on the current audit unit and material scope |
Three common misconfiguration issues
- Symptom: No streaming response in chat prompts, but curl calls to the model return content normally. Cause: The streaming response switch for
AIPROXY_API_ENDPOINTis not configured correctly, or the FastGPT version does not support the large model's SSE protocol. - Symptom: Model response is empty, logs show field exceptions. Cause: The connected third-party model does not correctly map the professional fields required for refinery business, or the
maxContextconfiguration exceeds the model's supported upper limit. - Symptom: Model calls fail after a version upgrade. Cause: The model integration proxy configuration was not updated synchronously, or context parameters from the old version are incompatible with the new version's rules.
How to verify proper configuration
- Run a curl command to call the configured model interface, check if the returned content includes professional terms and data formats related to refinery processes.
- Import a real distillate export quality report document, test whether the vector recall results match the quality indicator fields in the document.
- Enable the streaming response switch for model chats, observe whether the interface returns analysis content incrementally to confirm proper SSE protocol adaptation.
- Adjust the value range of the
temperatureparameter, verify whether analysis results under different values comply with the rigor requirements of refinery business.
The values provided are common starting points and should be measured against the reader's own samples.