What the data for this category looks like
Revenue and market data for oil and gas extraction comes primarily from publicly available domestic oil and gas industry statistical datasets, oil and gas block production briefings, and spot and futures market quotes. Data updates occur daily at midnight, with full calculated data for the prior day released. Documents store data in structured JSON or CSV format. Each entry includes operating block number, production cycle, extraction cost, crude oil selling price, per-well output volume, and comprehensive revenue calculation fields. Extraction cost is measured in yuan per ton of crude oil. Crude oil selling price is measured in US dollars per barrel. Per-well output volume is measured in cubic meters per day. The comprehensive revenue calculation field is a relative value without a percentage unit.
What constraints these characteristics impose on model integration and configuration
Data sources include pricing fields across multiple currencies. Unified unit conversion rules must be configured to prevent the model from reading parameters with mixed units. Data updates occur once daily. The scheduled trigger cycle for model integration must be matched to a fixed daily time window to ensure the latest prior-day data is pulled. Structured data includes multiple related fields. Field mapping rules must be configured to ensure the model correctly identifies core revenue calculation fields, and to avoid splitting related data across fields. Single data entries have a moderate size but tightly linked fields. Context window length must be adjusted to fully contain complete block revenue calculation information, and to avoid truncating critical parameters.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
Scheduled Task Cycle | Trigger once daily at 02:00 | Matches the T+1 update cadence of oil and gas extraction daily reports, and avoids pulling data during peak business hours |
PARSE_FILE_TIMEOUT_SECONDS | 900 seconds | Each structured data entry includes multiple fields, and the parsing process handles unit conversion and field relationships, so sufficient timeout time must be reserved |
Number of Recalled Entries | Top 3 entries | Oil and gas extraction block data is dispersed, and only a small number of precise recalls of core block revenue data can cover most analysis needs |
Similarity Threshold | 0.75–0.85 | Distinguishes revenue characteristics of different blocks, and avoids recalling historical data from unrelated blocks |
maxContext | 4000 characters | Each daily report entry includes multiple fields, so sufficient context must be reserved to contain complete block revenue calculation information |
Field Mapping Rules | Map lifting_cost and selling_price to input fields recognizable by the model | Adapts to differences in original field naming for structured data, and ensures the model correctly reads core calculation parameters |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require targeted analysis. It is recommended to test on your own samples before finalizing settings.
Three common configuration mistakes
- Symptom: After upgrading to version 4.8.20, the knowledge base can recall data normally, but the large language model returns no content. Cause: The offline reranking model adaptation switch is not enabled, so recalled data cannot be correctly passed to the large language model pipeline.
- Symptom: When configuring an offline reranking model, a call failure prompt appears. Cause: The deployment path of the reranking model is not added to the system environment variables, so FastGPT cannot load the offline model files.
- Symptom: After parsing oil and gas extraction daily report data, the model returns mixed units for revenue parameters. Cause: No exchange rate conversion rules are configured, and no unified conversion is performed between US dollar-priced selling prices and renminbi-priced costs, so the model reads data with mixed units.
How to confirm successful configuration
- Manually trigger the scheduled task once, check whether the previous day's oil and gas extraction data is successfully pulled to the knowledge base, and verify that the data fields match the configured mapping rules.
- Submit a test query, enter a revenue query requirement for an oil and gas extraction block, and check that the large language model returns content that includes the configured core calculation fields.
- View the system logs, confirm that the model call status code is 200, and that there are no timeout or parameter error prompts.
- Compare the manually calculated block revenue with the model's returned results, and adjust relevant configuration parameters to match the expected matching accuracy.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.