Method for automatically generating structure monitoring data analysis report based on workflow and industry knowledge base

Abstract

The invention discloses a workflow and industry knowledge base-based structure monitoring data analysis report automatic generation method, which is based on a FastGPT workflow task arrangement mechanism, splits the whole report generation process into data acquisition, processing, analysis, visualization and generation modules, and automatically controls the execution sequence through a configurable workflow. And automatic operation of low codes or zero codes is realized. And generating and controlling industry knowledge enhancement, and embedding LLM into the domain knowledge base. A natural language generation engine driven by a large language model converts a structured analysis result into text content with semantic coherence and professional and credible by utilizing LLM (such as depth-render), and the generated content has logicality, context consistency and industry term adaptability. Through an extensible module plug-in and scene multiplexing mechanism, a user can add or combine an analysis module and a language generation module according to service requirements, cross-project / industry multiplexing is supported, and the system flexibility and the life cycle value are improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically to a method for automatically generating structure monitoring data analysis reports based on workflow and industry knowledge base. Background Technology

[0002] Currently, the industry mainly uses the following methods to automatically generate data analysis reports: (1) Template-based report generation tools: Some business intelligence (BI) platforms (such as Tableau, Power BI, FineBI, etc.) provide chart generation and fixed template report export functions. Users can preset templates and then export analysis documents containing charts and fixed text by dragging and dropping charts and selecting data fields. This approach focuses on visualization rather than semantic analysis and automatic text writing.

[0003] (2) Rule engine-based statement concatenation method: Some industries (such as finance and manufacturing) use predefined rules (such as "generate a warning statement when sales drop by more than 10%) to map data to fixed statement fragments and then concatenate them into a report. This type of method is usually driven by code rules or business engines and does not have language flexibility and deep understanding capabilities.

[0004] (3) Preliminary automatic summary generation scheme using NLP technology: Some scientific research or report applications introduce traditional natural language processing (NLP) technologies, such as TextRank, LDA, and bag-of-words model, to generate simple summaries of the analysis results. However, these technologies have difficulty understanding context, domain terminology, or performing structured expressions, and their applicability is limited.

[0005] (4) Some AI products integrate LLM but do not have workflow capabilities: Currently, some AI document products (such as Notion AI, ChatGPT plugin, Tencent Hunyuan, etc.) have tried to use large language models for text generation, but these tools are usually geared towards general scenarios and lack a complete task process "from raw data to written report", and cannot realize the arrangement and implementation of processes such as structured data access, analysis module call, and semantic style control.

[0006] While existing solutions have some practicality in specific dimensions, they still face the following major problems when dealing with data reporting needs that require high automation, strong analytical capabilities, and flexible language generation: (1) Lack of full-process automation: Manual data processing is time-consuming and inefficient, making it difficult to meet high-frequency analysis needs.

[0007] (2) The generated content lacks intelligent understanding and flexibility: the template method is fixed and the rule splicing method is rigid, which cannot adapt to dynamic needs; the generated content lacks semantic understanding ability, and the result appears rigid.

[0008] (3) Limited generation quality: It relies on manual rules, resulting in high maintenance costs; traditional NLP methods cannot generate natural language content with logical flow, supporting conclusions, and coherent structure; and it cannot be optimized by combining industry knowledge.

[0009] (4) Lack of open workflow capabilities: Current large language model applications on the market lack programmability oriented towards actual business processes. Users cannot define the "process control capabilities" of key steps such as data source access logic, analysis logic selection, and report structure. Summary of the Invention

[0010] This invention provides a method for generating structural monitoring data analysis reports based on a combination of workflow and industry knowledge base. By defining the entire process of input-processing-output through workflow, the method achieves automatic execution of analysis tasks. By combining the industry knowledge base, the method calls the analysis module to complete the generation and execution of statistical analysis code for the data and obtains the calculation results. The method uses a large language model to convert the calculation results into natural language, forming paragraph-based and professional analysis conclusions.

[0011] To achieve the above objectives, the present invention adopts the following technical solution: A method for automatically generating structure monitoring data analysis reports based on workflow and industry knowledge base. The specific steps of this method are as follows: Step 1: User defines the analysis range and triggers task startup. The user inputs the range of measurement point data to be analyzed through the system front-end interface, including the structure name, monitoring factors, and query time period information. The system automatically starts the preset workflow based on this. Step 2: Extract query parameters and complete preprocessing. The workflow first extracts the query parameters of the measurement point data from the user input, including: structure name, monitoring factor name (such as "foundation pit settlement" and "retaining displacement"), and start and end time. The system performs preliminary analysis on the above parameters and standardizes the parameters into a data query format. Step 3: Parameter integrity check and fuzzy matching. The system performs integrity checks on the above parameters to ensure that there are no missing items or semantic ambiguities. For the names of monitoring factors, the system compares them with predefined monitoring factors through a fuzzy matching mechanism to improve fault tolerance and user experience. Step 4: Request measurement point data and generate trend chart display. The system calls the HTTP request node to connect to the backend monitoring database interface to complete data acquisition. After acquiring the data, the system immediately draws a trend chart to visualize the evolution of the data and for subsequent analysis. Step 5: Extract data features and generate analysis code. The system calls the code execution node to perform feature extraction operations on the measurement point data, including but not limited to: structure type, monitoring item name and unit, and measurement point data. The system combines the preset industry knowledge base to automatically generate Python analysis code for the data. The code content includes analysis algorithms and explanatory comments. Step 6: Execute the analysis code and verify the results. The system submits the automatically generated Python code to the backend code execution service through another HTTP request node. After execution, the system verifies the executability of the code. If an exception is found, the system automatically triggers the code regeneration logic. Step 7: Call the large language model to generate an analysis report. Based on the validity of the analysis results, the system calls the large language model (Alibaba Deepseek-r1 model) to complete the semantic parsing and text generation process.

[0012] Preferably, in step S1, the monitored factors include settlement, displacement, and tilt.

[0013] Preferably, in step 4, the system instantly generates a trend chart as a time series curve.

[0014] Preferably, in step 5, the preset industry knowledge base includes structural engineering specifications and expert cases.

[0015] Preferably, in step 7, the text generation logic includes the following sub-steps: 1) Use the analysis results as input for user questions; 2) Relevant content in the joint knowledge base, namely industry standards, project plans, technical documents, and expert lectures; 3) Generate a data analysis report containing a summary, analysis process, conclusions and recommendations according to the preset structure.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. Improve data analysis and report generation efficiency, saving manpower. Through a pre-set workflow, the entire process from data reading, processing, and analysis to report generation is automatically executed. Users only need to input a few parameters such as structures and monitoring factors to achieve one-click automatic report generation. Report generation time is reduced from hours / days to minutes, achieving 100% unattended operation, greatly saving labor costs, and supporting batch automated processing of multiple monitoring points.

[0017] 2. Improve the accuracy and standardization of data analysis results, avoiding human error. The system automatically matches analysis rules and terminology using a built-in industry knowledge base (such as industry standards and expert lecture materials), and interprets and expresses the semantics of the analysis through a large language model, avoiding interference from subjective judgment. The report content features standardized terminology and format, strong logical consistency in analysis, a low error rate, and automatically references industry knowledge to ensure the authority and reliability of technical conclusions.

[0018] 3. Possesses error detection and self-recovery capabilities, improving system stability. A built-in code execution verification mechanism automatically checks the output validity after the analysis script is executed. If the check fails, it can be automatically regenerated and retried, forming a closed-loop fault tolerance mechanism. The stability of analysis tasks is significantly improved, the system has self-healing capabilities, reducing the need for manual troubleshooting and ensuring a controllable success rate in large-scale tasks. 4. Enhance the semantic coherence and stylistic consistency of report expression. Based on a large language model (such as Alibaba Deepseek-r1), the system automatically generates fluent and logically rigorous text by combining data context and analysis conclusions. The reports are naturally fluent, logically sound, and adaptable to the needs of different user groups and scenarios, significantly improving the readability and professionalism of the reports.

[0019] 5. Enhance the system's generalizability and cross-industry adaptability by adopting a modular design concept, supporting rapid adaptation to different project scenarios by changing data source interfaces, analysis templates, language prompt templates, and knowledge base content. The same system architecture can serve multiple engineering monitoring or data analysis scenarios, reducing redundant development investment, increasing system reusability, and improving project delivery speed and maintainability. Attached Figure Description

[0020] Figure 1 This is a flowchart of the system processing of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0022] Example 1

[0023] Figure 1 The system processing flowchart for this patent can be roughly divided into three stages: preprocessing, data analysis, and report generation. Each stage contains multiple logical steps and includes anomaly detection and fault tolerance mechanisms, as detailed below: First, in the user-defined analysis scope step, the user inputs the data range to be analyzed through the front-end interface, including the structure name, monitoring factors (such as settlement, displacement, tilt, etc.), and the corresponding query time period. This operation serves as the trigger condition for the process, starting the entire automated task.

[0024] After entering the preprocessing stage, the system extracts the query parameters input by the user and parses and standardizes their format. Then, the system performs a parameter integrity check to determine whether the provided parameters are missing or contain semantic errors. If the parameters are complete, the process proceeds to the next stage; if missing or unrecognizable parameters are found, an error message is returned directly, and the process terminates.

[0025] During the data analysis phase, the system first requests measurement point data by initiating a request to the backend monitoring database via an HTTP interface. Upon successful data acquisition, the system automatically generates a trend chart to visualize the data changes. Subsequently, the system extracts features from the raw data (such as structure type, monitoring item name and unit, key measurement point data, etc.) and, combined with a pre-set industry knowledge base, automatically generates Python analysis code containing analytical logic and annotations. The generated code is then sent to the code execution module for execution.

[0026] After the code is executed, the result verification and judgment stage begins. If the execution is successful, the process continues to the report generation stage; if the execution fails, the system checks whether more than 4 retries have been made. If the limit has not been exceeded, the Python code will be regenerated and executed; if the retries have been exceeded, an error message will be returned and the process will terminate.

[0027] If the analysis code executes successfully, the system enters the report generation stage, calling a large language model (such as Alibaba Deepseek-r1) to generate a natural language analysis report. This model combines the analysis results with relevant content from the knowledge base (such as industry standards, project plans, expert lectures, etc.) to output a complete report containing a summary, analysis process, conclusions, and recommendations, according to a preset structure. Finally, the system outputs the analysis report in Markdown format for users to view and archive.

[0028] This flowchart, through its clearly structured modular distribution and error handling mechanism, achieves fully automated processing from data input to report output, greatly improving report generation efficiency and intelligence. It can be widely used in various industry scenarios such as structural monitoring and environmental monitoring.

[0029] 1. Preprocessing stage (1) Users log in to the system and fill in the analysis request through the graphical interface. The input content includes the name of the structure, the monitoring factors, and the query time range (such as "the tilt data of the rotary tiller factory at No. 235 Wuyi Road in the past two weeks").

[0030] (2) The system starts the workflow task and first enters the parameter extraction module to identify the structure ("No. 235 Wuyi Road, Building 3 Rotary Tiller Factory"), monitoring factors ("inclination") and time period ("the past two weeks") fields, and standardizes the input text (e.g., unifying the time format to "2025-06-11" and "2025-06-25").

[0031] (3) The parameter checking module verifies the integrity of the extracted parameters and calls the fuzzy matching algorithm to compare the monitored factors. For example, "tilt" can be automatically matched as "tilt rate".

[0032] (4) If the parameters are missing or incorrect, the system will terminate the process and prompt the front-end interface with the message "To obtain the measurement point data, you need to provide: the name of the structure and the name of the monitoring factor".

[0033] 2. Data Analysis Phase (1) After confirming that the parameters are correct, the system accesses the backend structure monitoring database (accessing the ClickHouse interface) via an HTTP request node and automatically queries the corresponding measurement point data of the selected structure within the specified time period. This data includes fields such as measurement point number and name, acquisition time, and measurement value.

[0034] (2) After the data is acquired, the system automatically draws a trend chart (time series line chart) to show the change of the slope over time and saves the image as an intermediate result.

[0035] (3) Subsequently, the system enters the data feature extraction module, which automatically analyzes the fluctuation range, trend of change, peak interval, etc. of the data. For example, it detects the maximum and minimum slope, average value, rate of change, outlier location, etc., and records these features as a structured feature set.

[0036] (4) The system calls industry knowledge bases (such as municipal engineering management standards, building foundation pit specifications, expert cases, etc.), searches for corresponding analysis strategies in the template library based on the structure type and monitoring factors, and automatically generates Python analysis code in combination with data characteristics.

[0037] (5) The generated code will be submitted to the backend code execution service (such as a Flask container) via an HTTP request. The system will execute the code and return the results (such as the mean, maximum and minimum values, and rate of change of each measurement point).

[0038] (6) The system verifies whether the code execution was successful. If the execution fails (e.g., syntax error, data out of bounds), the code regeneration mechanism is triggered, and it can be retried up to 4 times. If it still fails after 4 attempts, the system will return an error message and stop the process.

[0039] 3. Report Generation Stage (1) Based on the validity of the analysis results, the system calls the pre-configured large language model interface (such as deepseek-r1), takes the analysis results as user questions, and generates structured analysis report content in conjunction with knowledge base content (such as industry standards, project plans, etc.).

[0040] (2) The report structure includes: summary, data trend analysis, analysis and calculation process, basis for anomaly judgment, and final recommendations. An example excerpt is as follows: 1. Data Overview: Based on user-provided data, the tilt rate monitoring results for the building "No. 235, Wuyi Road, Building 3 (Rotary Tiller Factory)" are as follows: Monitoring point CX-1: X-direction tilt rate: average value 0.353%, rate of change 0.000272, range 0.321%~0.403%.

[0041] Y-direction tilt rate: average value 0.035%, rate of change 0.00000567, range 0.022%~0.055%.

[0042] Monitoring point CX-2: X-direction tilt rate: average value 0.350%, rate of change 0.000246, range 0.285%~0.416%.

[0043] Y-direction tilt rate: average value 0.000454%, rate of change -0.0000125, range -0.0382% to 0.0436%.

[0044] The tilt rate is defined as the ratio of the horizontal displacement of the top of the monitored object relative to its bottom to the height difference, and is used to calculate the tilt, direction, and rate.

[0045] 2. Trend Analysis: Overall trend: The rate of change of tilt at all monitoring points is less than 0.0003 (absolute value), indicating that the data fluctuation is minimal and the tilt state is stable. The average tilt rate in the X direction is around 0.35%, while in the Y direction it is 0.035% at CX-1 and close to 0% at CX-2, showing that the main tilt direction of the building is the X-axis.

[0046] CX-1 analysis: The rate of change in the X direction is 0.000272, indicating a slight positive increase that can be ignored; the rate of change in the Y direction is 0.00000567, indicating essentially no change.

[0047] CX-2 analysis: The rate of change in the X direction is 0.000246, consistent with CX-1; the rate of change in the Y direction is -0.0000125, showing a slight negative fluctuation, but the absolute value is extremely small (<0.0001), which does not affect the overall stability.

[0048] The formula for calculating the tilt rate is: Tilt Rate = (Top Horizontal Displacement - Bottom Horizontal Displacement) / Building Height. Current data shows that the tilt angle is extremely small and within the safe range.

[0049] 3. Possible influencing factors, foundation and load factors: Building tilt is usually affected by uneven foundation settlement or external loads (such as equipment vibration and surcharge). The higher tilt rate in the X direction may be related to the load offset caused by the long-term operation of rotary tillers inside the plant, but the current data does not show significant deterioration.

[0050] Environmental factors: Temperature changes or humidity fluctuations can cause material expansion and contraction, indirectly affecting the tilt rate. Small fluctuations in the Y direction (such as negative values ​​of CX-2) may originate from short-term environmental changes, but the rate of change is extremely low and the impact is limited.

[0051] Monitoring method factors: If non-automated instruments (such as theodolites or laser plumb bobs) are used, human error or instrument calibration deviation may cause slight fluctuations in the data. For example, the minimum value of -0.0382% in the Y direction of CX-2 may be a measurement error and needs to be verified in conjunction with instrument records.

[0052] 4. Future Development Forecast, Short-Term Forecast (1-3 Months): Based on the current rate of change (all <0.0003), the tilt rate is expected to remain stable. The X-direction may slowly increase to within 0.36%, but will not exceed the safe threshold of 0.5% (a common upper limit in the industry).

[0053] Long-term forecast (6 months or more): If there are no significant changes in load or environment (such as the addition of heavy equipment or extreme weather), the tilt rate will remain unchanged. However, the cumulative effect of the rate of change should be noted; if the rate of change continues to increase positively, the tilt rate in the X direction may reach 0.4%, requiring enhanced monitoring.

[0054] 5. Abnormal fluctuation assessment: No significant abnormal fluctuations were observed. All rates of change were below 0.0003, and the differences between the maximum and minimum values ​​were small (maximum difference in the X direction was approximately 0.12%, and the maximum difference in the Y direction was approximately 0.08%), which is within the normal range. The negative value in the Y direction of CX-2 (-0.0382%) is a minor fluctuation and does not constitute a risk.

[0055] Possible causes: Minor fluctuations may stem from instrument accuracy errors (such as the lack of a forced alignment device) or short-term environmental disturbances (such as wind loads). Inclination calculations must ensure stable station layout and avoid the influence of external factors.

[0056] 6. Optimization Recommendations: For improved monitoring methods, it is recommended to use an automated inclinometer or total station system to reduce human error. A dual-axis inclinometer can simultaneously capture X / Y direction data, improving accuracy.

[0057] Enhanced data processing: Regularly re-measure outliers (such as the Y-direction of CX-2) and submit tilt curves to analyze trends. The formula tilt rate = Δdisplacement / height needs to be verified in conjunction with building height data.

[0058] Preventive measures: Check the uniformity of the foundation and the load distribution of the equipment to avoid localized settlement. Measurement stations should be set up in stable locations at a height of 1.5 to 2 times the target to ensure data reliability.

[0059] Reporting standards: Improve daily reports, including monitoring time, instrument calibration records, and cumulative increments, to facilitate traceability.

[0060] Finally, the system will format the analysis report in a preset Markdown format for users to view or archive.

[0061] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for automatically generating structural monitoring data analysis reports based on workflow and industry knowledge base, characterized in that: The specific steps of this method for automatically generating structure monitoring data analysis reports based on workflow and industry knowledge base are as follows: Step 1: User defines the analysis range and triggers task startup. The user inputs the range of measurement point data to be analyzed through the system front-end interface, including the structure name, monitoring factors, and query time period information. The system automatically starts the preset workflow based on this. Step 2: Extract query parameters and complete preprocessing. The workflow first extracts the query parameters of the measurement point data from the user input, including: structure name, monitoring factor name, start and end time. The system performs preliminary analysis on the above parameters and standardizes the parameters into a data query format. Step 3: Parameter integrity check and fuzzy matching. The system performs integrity checks on the above parameters to ensure that there are no missing items or semantic ambiguities. For the names of monitoring factors, the system compares them with predefined monitoring factors through a fuzzy matching mechanism to improve fault tolerance and user experience. Step 4: Request measurement point data and generate trend chart display. The system calls the HTTP request node to connect to the backend monitoring database interface to complete data acquisition. After acquiring the data, the system immediately draws a trend chart to visualize the evolution of the data and for subsequent analysis. Step 5: Extract data features and generate analysis code. The system calls the code execution node to perform feature extraction operations on the measurement point data, including but not limited to: structure type, monitoring item name and unit, and measurement point data. The system combines the preset industry knowledge base to automatically generate Python analysis code for the data. The code content includes analysis algorithms and explanatory comments. Step 6: Execute the analysis code and verify the results. The system submits the automatically generated Python code to the backend code execution service through another HTTP request node. After execution, the system verifies the executability of the code. If an exception is found, the system automatically triggers the code regeneration logic. Step 7: Call the large language model to generate an analysis report. Based on the validity of the analysis results, the system calls the large language model to complete the semantic parsing and text generation process.

2. The method for automatically generating structural monitoring data analysis reports based on workflow and industry knowledge base according to claim 1, characterized in that: In step S1, the monitored factors include settlement, displacement, and tilt.

3. The method for automatically generating structural monitoring data analysis reports based on workflow and industry knowledge base according to claim 1, characterized in that: In step 4, the system instantly generates a trend chart as a time series curve.

4. The method for automatically generating structural monitoring data analysis reports based on workflow and industry knowledge base according to claim 1, characterized in that: In step 5, the pre-set industry knowledge base includes structural engineering specifications and expert cases.

5. The method for automatically generating structural monitoring data analysis reports based on workflow and industry knowledge base according to claim 1, characterized in that: In step 7, the text generation logic includes the following sub-steps: 1) Use the analysis results as input for user questions; 2) Relevant content in the joint knowledge base, namely industry standards, project plans, technical documents, and expert lectures; 3) Generate a data analysis report containing a summary, analysis process, conclusions and recommendations according to the preset structure.

Images