What the data for this category looks like
Data sources include operation logs of grid-connected photovoltaic projects, monthly monitoring reports from industry associations, and real-time collected data from third-party operation and maintenance platforms. There are two update frequency categories: real-time data for individual projects is updated every 15 minutes, and industry average yield data is updated monthly. The primary presentation format is structured tables, including fields such as project ID, installed capacity, annual irradiation duration, on-grid electricity price, unit operation and maintenance cost, and cumulative power generation. The units of these fields are kilowatt, hour, yuan per kilowatt-hour, yuan per kilowatt, and kilowatt-hour respectively.
What constraints these characteristics impose on multi-turn dialogue and prompt engineering
First, the 15-minute update frequency of real-time data requires that each call in a multi-turn dialogue must explicitly specify the data time window to avoid retrieving expired information. Second, the differentiated update rhythms of industry average and individual project data require adding restrictions on data source and type in the prompt to prevent the model from confusing the two types of data. Third, the unit differences across multiple fields require the prompt to explicitly require attaching the standard unit of each field when outputting, to avoid unit conversion errors. Fourth, the binding requirement for individual project data requires that the multi-turn dialogue context must be associated with the project ID to prevent cross-project data misuse.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 characters | Photovoltaic yield-related data includes multi-field historical records; longer context can retain project ID and time range restrictions |
recall_top_k | Top 6–8 entries | Photovoltaic data includes multiple related fields such as installed capacity, irradiation duration, and electricity price; sufficient related information must be recalled while avoiding redundant interference |
PROMPT_TEMPLATE | Fixed inclusion of "Only use the corresponding data of the specified photovoltaic project, and attach the standard unit of each field in the output" | Clear restrictions on data scope and format are required to prevent the model from confusing different projects or making unit errors |
global_variable_scope | Bound within the session | Photovoltaic project data must be bound to the project ID to prevent cross-session or cross-project data misuse |
maxToolCallTimes | 3–5 times | Photovoltaic yield calculation requires calling multiple tool nodes such as irradiation amount, on-grid electricity price, and operation and maintenance cost in sequence; infinite loops must be avoided |
UPLOAD_FILE_MAX_SIZE | 50 MB | Documents such as operation manuals and grid connection agreements for photovoltaic projects usually do not exceed this size, meeting common upload requirements |
The parameter values provided on this page are conventional recommendations used as a starting point 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 mistakes
- Symptom: After switching sessions, the global variable corresponding to the previously bound photovoltaic project ID cannot be retained. Cause: The
global_variable_scopeconfiguration was not set to bound within the session, causing the global variable scope to exceed the current session range. - Symptom: After uploading the grid connection agreement document for a photovoltaic project, the system prompts parsing failure or returns empty fields. Cause: The
UPLOAD_FILE_MAX_SIZEwas not adjusted to match the actual size of the document, or the prompt did not specify the target fields for parsing photovoltaic-related documents. - Symptom: After calling the yield calculation tool, the result is automatically inserted at the end of the conversation history. Cause: The "Auto-append to context" switch for the AI conversation node was not turned off, causing the tool return result to be mixed into the conversation flow.
How to confirm the configuration is correct
- Initiate a test conversation that includes the project ID, check whether the context retains the project ID, and confirm that the context length and global variable scope configurations are effective.
- Upload a photovoltaic document of the expected size, check whether the parsing result includes the preset target fields, and confirm that the uploaded file size and prompt template configurations are correct.
- Call the yield calculation tool, check whether the returned result only includes the calculated value and field units, with no additional conversation content, and confirm that the auto-append switch for the AI conversation node has been turned off.
- Switch to another test project session, verify that the global variable automatically updates to the new project ID, and confirm that the scope configuration meets expectations.
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.