Monitoring data trend analysis and decision suggestion generation method based on large language model agent

Abstract

The invention discloses a monitoring data trend analysis and decision suggestion generation method based on a large language model agent. The method comprises the following specific steps: step 1, constructing a domain knowledge enhancement module; 2, developing and registering a callable tool set, and compiling a series of performance functions by using Python; 3, constructing an agent arrangement platform and a core agent, and creating a master control agent in the agent arrangement platform as a scheduling center of an analysis task; step 4, executing an intelligent agent analysis process, and when an analysis request is received, enabling the intelligent agent to autonomously run according to the following process; 5, system integration and interface calling are carried out, the whole agent is packaged into RESTful API service and deployed in an enterprise server, by introducing a large language model agent architecture, an RAG enhanced reasoning mechanism and a callable tool set, the technical transition from passive data display to active insight generation is achieved, and the method has the advantages of being high in practicability and easy to popularize. And the analysis efficiency, the decision accuracy and the system intelligence level are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of large language model technology, specifically to a method for monitoring data trend analysis and decision suggestion generation based on a large language model intelligent agent. Background Technology

[0002] Currently, existing monitoring data analysis mainly relies on the following methods: threshold alarm mechanism: by setting fixed upper and lower thresholds, an alarm is triggered when the monitoring data exceeds the range.

[0003] Traditional statistical analysis and visualization tools, such as line charts, bar charts, and heatmaps, combined with statistical methods like mean, variance, and sliding window, are used by operations and maintenance personnel to manually judge data trends. These methods are highly dependent on human experience, have low analytical efficiency, are ill-suited for handling large-scale, high-frequency data streams, and cannot automatically generate analytical conclusions and decision-making recommendations in natural language.

[0004] Machine learning-based prediction models: Some systems have introduced time series prediction models (such as ARIMA and LSTM) for trend prediction or anomaly detection, which has improved prediction accuracy. However, the output results lack interpretability, making it difficult for non-professional users to understand the basis of their judgments. Furthermore, the model outputs are mostly numerical values ​​or labels (such as "anomaly probability 0.85"), which cannot be directly converted into actionable maintenance suggestions (such as "suggest checking component XX"), still requiring manual intervention for interpretation and decision-making.

[0005] Rule engine system: This system uses pre-defined expert rules (such as "If the temperature continues to rise and the vibration value exceeds the standard, then determine the bearing overheating") to make logical judgments. While simple to implement, this method lacks understanding of data trends and context, making it prone to false alarms or missed alarms. It cannot predict potential faults or provide in-depth analytical suggestions. Such systems are effective in specific scenarios, but rule writing is costly, maintenance is difficult, it struggles to cover complex and changing actual working conditions, and it lacks adaptability and generalization capabilities.

[0006] This application aims to address the technical challenges of traditional monitoring data analysis, such as reliance on human experience, low analysis efficiency, lack of contextual explanation for alarm information, and difficulty in generating actionable decision-making recommendations. In existing systems, users often need to manually judge trends and risks when faced with massive amounts of sensor data. This is prone to overlooking key information due to the large volume and multiple dimensions of data, leading to untimely and inaccurate assessments of equipment or structural health. This application proposes an automated analysis method based on a large language model intelligent agent. By constructing an AI intelligent agent with task planning, tool invocation, and natural language generation capabilities, it automatically completes the entire analysis process from data acquisition, trend identification, anomaly diagnosis to maintenance recommendation generation when users query sensor data, and outputs an easily understandable health status assessment report in natural language. This solution significantly improves the intelligence level of data analysis and user experience, realizing a shift from "passively viewing data" to "actively gaining insights." This application has been applied to our company's independently developed "Anxin Cloud" monitoring data cloud platform. In this platform, when users view historical or real-time data from sensors, the system can automatically trigger the intelligent agent described in this invention to perform analysis and generate specific suggestions such as "the vibration value of this measuring point has been showing a continuous upward trend in the past 7 days, and it is recommended to check the tightness of the equipment foundation." This provides services for a wide range of project scenarios, including infrastructure structural safety and bridge and tunnel monitoring, effectively reducing the operation and maintenance threshold and improving risk warning capabilities and decision-making efficiency. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for trend analysis and decision suggestion generation of monitoring data based on a large language model intelligent agent. The specific steps of this method for trend analysis and decision suggestion generation of monitoring data based on a large language model intelligent agent are as follows: Step 1: Construct Domain Knowledge Enhancement Module (RAG) and establish a vectorized knowledge base in the intelligent agent orchestration platform, including two sub-bases: an industry standard base and a historical case base; All documents are segmented and vectorized and then stored in a vector database (pgvector). When the agent makes a judgment, the system automatically retrieves the knowledge fragments most relevant to the current analysis context and uses them as context input to the large model to enhance the accuracy of reasoning. Step 2: Develop and register a set of callable tools. Write a series of functional functions in Python and register them as callable tools for the intelligent agent using the MCP (Model Context Protocol). Step 3: Construct an intelligent agent orchestration platform and core intelligent agents. In the intelligent agent orchestration platform, create a master intelligent agent as the scheduling center for analysis tasks. This intelligent agent is based on a large language model (such as DeepSeek, Qwen and other domestic high-performance LLMs), and is given role definitions, task objectives and workflow specifications through carefully designed system prompts. Step 4: Agent analysis process execution. Upon receiving an analysis request, the agent autonomously runs according to the following process: Analyze user questions or system commands to clarify the analysis objectives; Obtain the raw data; Choose whether to call utility functions for numerical analysis based on the data characteristics; By combining the industry standards database and historical case database retrieved by RAG, we can comprehensively determine the possible causes of data anomalies. Generate natural language reports that include trend descriptions, risk levels, attribution analysis, and operational recommendations; Step 5: System integration and API call. The entire intelligent agent is encapsulated as a RESTful API service and deployed on the enterprise server. This RESTful API receives JSON format requests from the "Anxin Cloud" platform, triggers the intelligent agent to execute the analysis process, and returns structured results (including text reports, key indicators, recommendation levels, etc.), which are finally displayed to the user in the form of a sidebar on the "Anxin Cloud" front-end interface.

[0008] Preferably, the industry standard library includes documents such as structural safety assessment standards and equipment operation and maintenance manuals.

[0009] Preferably, the historical case library stores typical past failure events and their handling solutions.

[0010] Preferably, in step 2, the tools specifically include: To query the basic information of the structure to which a certain sensor belongs, namely, the structure type and location; Acquire historical monitoring data from a specified sensor; Check the recent alarm records of this sensor; Acquire weather data for the corresponding time period at the sensor location, namely temperature, humidity, wind speed, and precipitation; Calculate the Pearson correlation coefficient between two sets of time series data (such as displacement and temperature); Calculate the statistical characteristics of the data, such as the maximum value, minimum value, average value, and rate of change; The tools mentioned above are all essential core components that ensure that the agent has the ability to acquire information from multiple sources and perform specialized analysis.

[0011] Preferably, in step 2, the JSON format request includes the sensor ID and time range parameters.

[0012] Preferably, the structured results include text reports, key metrics, and recommendation levels.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: by introducing a large language model intelligent agent architecture, a RAG enhanced inference mechanism, and a callable toolset, a technological leap from "passively displaying data" to "actively generating insights" is achieved, resulting in significant improvements in analysis efficiency, decision-making accuracy, and system intelligence. Specific effects are as follows: Significantly improves analysis efficiency and reduces reliance on manual intervention: Traditional methods require data analysts to manually examine multidimensional data, call analysis tools, and consult standard documents to make judgments, averaging over 30 minutes. This invention automates the entire process of data querying, statistical analysis, knowledge retrieval, and report generation through an intelligent agent, with a response time controlled within 2 minutes, significantly improving analysis efficiency. Its core mechanism lies in the intelligent agent's task planning capabilities, enabling it to call multiple tools in parallel (such as sensor data querying, weather data acquisition, and correlation calculation), avoiding manual operations item by item and significantly shortening the analysis chain.

[0014] Improving the accuracy of analysis and the professionalism of recommendations: Existing systems can only provide threshold alarms or simple trend charts, lacking contextual explanations and resulting in a high false alarm rate. This application embeds "industry standards" and "historical cases" into the reasoning process through the RAG mechanism, enabling the large model to refer to real engineering standards and past experience when generating recommendations. For example, when identifying displacement anomalies, the system can calculate correlation coefficients by combining temperature change data and referencing standard provisions to determine whether it falls within the normal range of thermal expansion and contraction, thereby reducing the false alarm rate and improving the feasibility of recommendations.

[0015] Achieving explainable intelligent decision support: Compared with the "black box" output of traditional machine learning models, the intelligent agent in this application will explicitly explain the basis for judgment when generating suggestions (such as "the settlement rate in the past 7 days reached 0.5 mm / d, exceeding the allowable value of 0.3 mm / d in standard GB50982-2014"), forming a traceable and verifiable analysis logic chain, which enhances the user's trust in the AI ​​output.

[0016] Simplify operation process and improve platform usability: This application encapsulates complex analysis capabilities into APIs. After being embedded into the "Anxin Cloud" platform, users only need to click the "AI Analysis" button to obtain professional-grade reports. No data analysis skills or industry standards are required, which achieves the goal of enabling non-professionals to quickly understand the structural health status and effectively reduces the threshold for system use.

[0017] In summary, this application solves the technical defects of traditional methods, such as low efficiency, poor accuracy, and reliance on manual labor, through the collaborative mechanism of "intelligent agent + tool invocation + knowledge enhancement". It realizes the automation, intelligence and interpretability of monitoring data analysis, and has outstanding substantive features and significant progress. Attached Figure Description

[0018] Figure 1This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the intelligent agent process of the present invention. Detailed Implementation

[0019] 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.

[0020] Agent architecture design based on large language models: The core innovation of this application lies in replacing traditional static models or rule engines with an LLM agent that possesses task planning, tool invocation, memory mechanisms, and multi-step reasoning capabilities. This enables the system to proactively decompose complex analysis tasks (such as trend identification → anomaly attribution → suggestion generation) and realize an autonomous decision-making process similar to that of a human analyst.

[0021] The ability of intelligent agents to dynamically invoke data analysis tools: The intelligent agent can autonomously call external preset data processing programs (such as moving average, FFT, clustering algorithms, etc.) according to the analysis needs, to achieve deep integration of LLM and professional algorithms, and break through the limitation that large models cannot directly process raw numerical data.

[0022] Combining contextual reasoning and suggestion generation with domain knowledge base: The analysis process incorporates structured knowledge bases from various industry sectors (such as industry standards and historical cases), and guides the agent to perform causal reasoning through RAG (Retrieval Enhanced Generation) and prompting engineering, significantly improving the professionalism and accuracy of the generated suggestions.

[0023] Automated triggering and natural language output mechanisms integrated with cloud platforms: The system is embedded in the "Anxin Cloud" platform, which can trigger intelligent agent analysis when users query sensor data and output reports containing trend judgments, risk warnings and actionable suggestions in natural language, realizing a closed loop of user experience of "querying is insight".

[0024] Example 1

[0025] I. System Deployment Environment This system is deployed in an enterprise private cloud environment. Its core components include: FastGPT intelligent agent orchestration platform, vector database, Python tool service module and "Anxin Cloud" front-end platform.

[0026] II. Agent Construction and Knowledge Enhancement Create a workflow application in the FastGPT platform with the role set as "Structural Safety Monitoring Expert". The system prompt clearly states its responsibilities as: "Based on user queries, call tools to obtain data, combine industry standards and historical cases, analyze sensor data trends, and generate natural language reports containing risk assessments and maintenance suggestions."

[0027] Simultaneously, the RAG (Retrieval Enhancement Generation) module is integrated into FastGPT to build a vector knowledge base, including industry standards such as GB 50982-2014 Technical Specification for Monitoring of Building and Bridge Structures and DLT 5416-2009 Technical Specification for Safety Monitoring of Hydraulic Structures under Strong Earthquakes. These documents are then sliced ​​and converted into vectors using the xiaobu-embedding-v2 model and stored in a vector database for semantic retrieval.

[0028] When the agent performs reasoning, the system automatically uses the current analysis context as the query keywords, retrieves document fragments with a relevance greater than 0.4 from the knowledge base, and concatenates them into the input context of the large model to assist it in making professional judgments.

[0029] III. Development and Integration of Callable Toolsets A set of analysis tools was written using Python 3.10. All tools are encapsulated according to the Model Context Protocol (MCP) specification to ensure compatibility with LLM. They are then registered through FastGPT's "MCP Toolset," specifically including: get_structure_info(structure_id: number) → dict Enter the structure ID to query information such as structure type, design parameters, and geographical location from the Anxin Cloud Platform backend interface.

[0030] get_sensor_data(sensor_id: str, start: str, end: str) → list Query monitoring data (such as displacement, tilt, and vibration values) within a specified time period from the Anxin Cloud Platform.

[0031] get_alert_history(sensor_id: str, start: str, end: str) → list Query the alarm data of the sensor within a specified time period to determine the frequency of historical anomalies.

[0032] get_weather_data(location: str, start: str, end: str) → dict Call the meteorological API to obtain environmental data such as temperature, humidity, wind speed, and precipitation corresponding to the geographical location.

[0033] compute_pearson_corr(series_a: list, series_b: list) → float The Pearson correlation coefficient of the two sets of time series data is calculated to determine the correlation between sensor data and weather factors.

[0034] compute_stats(data: list) → dict Calculate the statistical characteristics of the data, such as the maximum value, minimum value, average value, standard deviation, and rate of change.

[0035] The aforementioned tools are all essential technical features that enable intelligent agents to acquire information from multiple sources and perform specialized calculations.

[0036] IV. Intelligent Agent Workflow When a user queries certain sensor data on the "Anxin Cloud" platform and clicks the "AI Analysis" button, the intelligent agent executes the following process: Receive request parameters (sensor_id, time_range); Call get_sensor_data to retrieve the raw data; Use compute_stats to analyze data trends; Call get_weather_data to get the weather data for the same period; The `compute_pearson_corr` function is used to determine the correlation between the data and weather data such as temperature and wind speed. Use the RAG search function to check if there are any relevant historical cases or standards for reference; If the correlation is weak but the data is abnormal, the risk level is determined by combining get_alert_history and the knowledge base; Based on all the information, a natural language report is generated from the large model, such as: "The strain monitoring data of the XX Bridge shows that the two measuring points exhibit compressive strain (mean -61.33με) and tensile strain (mean 105.21με) respectively, which are strongly correlated with temperature (0.82-0.85). At the same time, the system has detected continuous data interruption alarms from 10 CTL series measuring points on the left side since XX, and there is a risk to the integrity of the monitoring data."

[0037] The strain data exhibited a clear temperature correlation. The strain range at the first measuring point was -158.46 to 14.73 με, and at the second measuring point it was -57.59 to 205.8 με. The fluctuations were highly synchronized with temperature changes. The correlation between humidity and precipitation was below 0.04, indicating that ambient temperature was the primary influencing factor. Data interruption alarms persisted for over XX days, affecting monitoring continuity. According to specifications, strain monitoring of beam bridges should focus on key sections of the main girder, and the analysis of monitoring data should include correlation analysis between structural response and environmental parameters. The bridge health monitoring system should monitor the bridge site environment and structural condition in real time, promptly identifying abnormal changes. The current data interruption problem violates the basic requirement of ensuring data continuity for the monitoring system. It is recommended to immediately repair the interrupted measuring points to restore monitoring data continuity; strengthen the periodic analysis of the impact of temperature on strain; improve the early warning mechanism to ensure timely identification of structural response anomalies; and implement continuous monitoring of strain at key sections in accordance with specifications. V. System Integration Method The agents configured in FastGPT are encapsulated as RESTful API services (using FastGPT's API publishing functionality) and deployed on an intranet server. The API interface uses the standard OPEN AI GPT official interface specification, defined as follows: URL: POST / api / v1 / chat / completions Input: { "token": "xxxxxxxxxxxxx", "structure_id": 4762, "begin": 1641016880831, "end": 1761112880831, "user_id": 205, "station_id": "66309,66310" } Output: { "id": "adsfasf", "model": "", "usage": { "prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 1 }, "choices": [ { "message": { "role": "assistant", "content": "The director of the film 'Suzuger's Journey' is Makoto Shinkai." }, "finish_reason": "stop", "index": 0 } ] } It supports streaming response output, meaning that instead of returning all results at once, it returns them gradually, reducing the initial response time and improving the user experience.

[0038] The “Anxin Cloud” platform calls this API from the backend, and displays the AI ​​analysis results in the form of a sidebar on the frontend, achieving seamless integration.

[0039] VI. Optional Extension Methods Image analysis tools can be added to support the use of CV models to identify crack images; A user feedback mechanism can be introduced to incorporate suggestions into the optimization of prompts. It supports multi-sensor linkage analysis to identify regional risks.

[0040] This implementation method has been deployed in a commercial environment and is running stably in the "Anxin Cloud" platform.

[0041] 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 monitoring data trend analysis and decision suggestion generation based on a large language model intelligent agent, characterized in that: The specific steps of the monitoring data trend analysis and decision suggestion generation method based on large language model intelligent agents are as follows: Step 1: Construct a domain knowledge enhancement module and establish a vectorized knowledge base in the intelligent agent orchestration platform, including two sub-bases: an industry standard base and a historical case base; All documents are segmented and vectorized and then stored in a vector database. When the agent makes a judgment, the system automatically retrieves the knowledge fragments most relevant to the current analysis context and uses them as context input to the large model to enhance the accuracy of reasoning. Step 2: Develop and register a set of callable tools. Write a series of function calls in Python and register them as callable tools for the intelligent agent using the MCP method. Step 3: Construct an intelligent agent orchestration platform and core intelligent agents. In the intelligent agent orchestration platform, create a master intelligent agent as the scheduling center for analysis tasks. This intelligent agent is based on a large language model and is given role definitions, task objectives, and workflow specifications through carefully designed system prompts. Step 4: Agent analysis process execution. Upon receiving an analysis request, the agent autonomously runs according to the following process: Analyze user questions or system commands to clarify the analysis objectives; Obtain the raw data; Choose whether to call utility functions for numerical analysis based on the data characteristics; By combining the industry standards database and historical case database retrieved by RAG, we can comprehensively determine the possible causes of data anomalies. Generate natural language reports that include trend descriptions, risk levels, attribution analysis, and operational recommendations; Step 5: System integration and API call. The entire intelligent agent is encapsulated as a RESTful API service and deployed on the enterprise server. This RESTful API receives JSON format requests from the "Anxin Cloud" platform, triggers the intelligent agent to execute the analysis process, and returns structured results, which are finally displayed to the user in the form of a sidebar on the "Anxin Cloud" front-end interface.

2. The method for monitoring data trend analysis and decision suggestion generation based on a large language model intelligent agent according to claim 1, characterized in that: The industry standard library includes structural safety assessment standards and equipment operation and maintenance manuals.

3. The method for monitoring data trend analysis and decision suggestion generation based on a large language model intelligent agent according to claim 1, characterized in that: The historical case library stores typical past failure events and their handling solutions.

4. The method for monitoring data trend analysis and decision suggestion generation based on a large language model intelligent agent according to claim 1, characterized in that: In step 2, the tools specifically include: To query the basic information of the structure to which a certain sensor belongs, namely, the structure type and location; Acquire historical monitoring data from a specified sensor; Check the recent alarm records of this sensor; Acquire weather data for the corresponding time period at the sensor location, namely temperature, humidity, wind speed, and precipitation; Calculate the Pearson correlation coefficient between two sets of time series data; Calculate the statistical characteristics of the data, such as the maximum value, minimum value, average value, and rate of change; The tools mentioned above are all essential core components that ensure that the agent has the ability to acquire information from multiple sources and perform specialized analysis.

5. The method for monitoring data trend analysis and decision suggestion generation based on a large language model intelligent agent according to claim 1, characterized in that: In step 2, the JSON format request includes the sensor ID and time range parameters.

6. The method for monitoring data trend analysis and decision suggestion generation based on a large language model intelligent agent according to claim 1, characterized in that: The structured results include text reports, key metrics, and recommendation levels.

Images