What the data for this category looks like
Data related to shipping port yield rates mainly comes from port authority public operational reports, international shipping trading platforms, and customs declaration data. Most data updates occur daily. Real-time berth operation data for some trunk ports updates every 1 hour. The data is presented in structured tables. Core fields include port UNLOCODE, berth number, daily container throughput, route freight rate points, vessel berth on-time rate, and daily operation duration. Some extended fields include yard storage rate and fuel surcharge values.
Constraints Imposed by These Characteristics on Model Access and Configuration
Multiple data sources require configuring multi-data-source access authentication and format adaptation rules. This prevents model parsing errors caused by differing field names returned by different platforms. Fixed daily update rhythms require configuring scheduled trigger task parameters. This ensures the model uses the latest daily operation and freight rate data during calls. The diversity of structured fields requires pre-defining Function Call parameter schemas. Match the units and value ranges of core fields such as port codes, throughput, and freight rates. This prevents unit confusion or missing fields during model calls. For berth data scenarios with high real-time requirements, adjust the data pull timeout threshold. This ensures timely data acquisition.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
functionCallSchema | Include four required fields: portUnlocode, containerThruput, freightRate, updateTime. Field types are string, integer, float, and ISO time string respectively | Matches the core field definitions of shipping port yield rate data, ensuring large models can correctly parse structured return results |
dataSyncInterval | 86400 seconds (once daily). Adjust to 3600 seconds for real-time scenarios | Matches the regular update rhythm of port operation data, balancing data timeliness and interface call costs |
fieldMappingRule | Uniformly map throughput fields from different data sources to containerThruput, retain TEU as the default unit | Resolves naming differences for throughput fields across data sources, preventing missing fields during model parsing |
maxContextWindow | 8000–12000 characters | Adapts to the total length of spliced multi-segment port operation data, ensuring large models can fully process batch data |
apiRequestTimeout | 600 seconds | Requires sufficient waiting time when pulling multiple data sources, preventing timeout failures due to large data volumes |
ragRetrieveTopK | Top 3 entries | Shipping port yield rate related data entries are limited. Excessive recall increases model processing load |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material form, data volume and business rules. Specific issues require case-by-case analysis, and it is recommended to test on your own samples before finalizing.
Three Common Configuration Mistakes
- Phenomenon: The model returns only structured statistical values and does not include original database snippets. Cause: The original data retention switch for Function Call is not configured, or the prompt does not explicitly require referencing original database content.
- Phenomenon: The model selection drop-down box keeps jumping and cannot fix the target model. Cause: The fixed priority parameter for the model is not configured, or there are abnormal entries in the interface returned model list causing front-end rendering exceptions.
- Phenomenon: Context is lost during multi-turn conversations, and the previous port yield rate query logic cannot be continued. Cause: The multi-turn context storage switch for the application is not enabled, or the configured context window length is insufficient to cover the content of multi-turn interactions.
How to Confirm Successful Configuration
- Execute a manually triggered data synchronization task. Verify that the returned structured data fields match the content defined in
functionCallSchema, and that units match the preset mapping rules. - Initiate a single port yield rate query. Confirm that the model return results include original database snippets, and that no missing fields or unit confusion issues occur.
- Initiate two consecutive related queries. Confirm that the context is correctly stored and the model can continue the answering logic based on the previous round's query information.
- View the model configuration page. Confirm that the selected target model does not automatically switch, verifying that the fixed priority configuration takes effect.
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.