Intelligent monitoring method and system for device state based on AI large model and device

Abstract

The embodiment of the application relates to the technical field of energy internet of things, and discloses an equipment state intelligent monitoring method, system and equipment based on an AI large model, which comprises the following steps: an open source development platform receives an input natural language used for query, sends the natural language to an AI model to perform intention recognition and convert the natural language into a structured execution instruction, and sends the structured execution instruction to a tool server; the tool server queries a tool according to the structured execution instruction, queries a time series database, and obtains corresponding original running data; the original running data is subjected to standardization processing to obtain standardized data; an analysis tool is called again to perform analysis, standardized analysis results are obtained, and the standardized analysis results are transmitted to the AI model; the AI model performs intelligent analysis on the standardized analysis results, generates an analysis report in the form of natural language, and transmits the analysis report to the open source development platform to present the analysis report to a user. The embodiment of the application solves the problem of high data analysis threshold, and significantly improves the intelligent level and efficiency of hydrogen production equipment operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of energy Internet of Things (IoT) technology, specifically to a method, system, and device for intelligent monitoring of device status based on an AI large-scale model. Background Technology

[0002] With the large-scale deployment of hydrogen production projects through water electrolysis, the monitoring of hydrogen production equipment operation status faces new challenges. Traditional monitoring systems mostly adopt a passive mode of "humans searching for data and reading reports," typically storing massive amounts of time-series data in a database and then extracting and analyzing the data by having professionals write query scripts (such as SQL).

[0003] This results in on-site maintenance personnel having a deep understanding of equipment principles and processes, but lacking familiarity with database operations and programming skills. They face extremely high technical barriers when attempting to independently and flexibly explore the root causes of equipment anomalies and perform multi-parameter correlation analysis. This makes it difficult for their valuable field experience to effectively interact with the data system, leading to long problem response cycles. In emergency situations such as equipment malfunctions, the complex operating interface and fragmented information presentation force maintenance personnel to repeatedly switch between multiple screens to piece together complete information, severely impacting the speed of judgment and decision-making. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention provide a method, system and device for intelligent monitoring of equipment status based on an AI large model, which is used to solve the technical problems of low operation and maintenance efficiency and insufficient intelligence level caused by the passive interaction mode, high threshold for use and inflexible function expansion of existing hydrogen production equipment monitoring methods.

[0005] According to one aspect of the present invention, a method for intelligent monitoring of device status based on an AI large model is provided, the method comprising: The open-source development platform receives natural language input for querying and sends the natural language to the AI ​​model; the natural language includes factory identifiers, equipment identifiers, time ranges, and status query elements; The AI ​​model recognizes the received natural language execution intent, converts the natural language into structured execution instructions, and sends the structured execution instructions to the tool server. The structured execution instructions include: factory identifier, equipment identifier, time range parameters, query requirements, and a list of tools to be invoked. The tool server receives the structured execution instruction, calls the pre-encapsulated query tool in the tool server according to the tool list, queries the time series database, and obtains the raw operating data of the target device within a specified time range; the raw operating data is standardized to obtain standardized data; the raw operating data is a set of data points consisting of timestamps and corresponding values ​​collected by the device sensors and stored in the database; the tool server is a server program that follows the model context protocol, and based on the model context protocol, the internal functions of the server program are standardized and encapsulated to obtain the tool, which includes the query tool and the analysis tool; According to the tool list, the pre-packaged analysis tools in the tool server are called to analyze and process the standardized data, obtain standardized analysis results, and transmit them to the AI ​​model. The analysis tools include anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools. The AI ​​model performs intelligent analysis based on the received standardized analysis results, generates an analysis report in natural language, and transmits it to the open-source development platform for presentation to the user.

[0006] In one alternative approach, the AI ​​model deployment mode includes at least one of cloud API mode and local private deployment mode, and the AI ​​model deployment mode can be smoothly switched between the two modes.

[0007] In one optional approach, the tool server receives the structured execution instruction, invokes a pre-packaged query tool in the tool server according to the tool list, queries the time-series database, and obtains the raw operating data of the target device within a specified time range; the raw operating data is then standardized to obtain standardized data, including: The query tool is invoked, and its internal server-side parameter verification module, query optimization decision module, database execution module, and post-processing module are sequentially linked for processing, including: The parameter verification module is invoked to verify the factory identifier, the equipment identifier, and the time range parameter. The verification content includes the legality and completeness of the parameters. If the verification passes, the query optimization decision module is invoked to generate a query strategy based on the query requirements. The query strategy includes direct query, downsampling query, and query by invoking a pre-computed materialized view. The time-series database execution module is invoked to send a query request to the time-series database according to the query strategy; The post-processing module receives the returned raw running data, performs basic cleaning and format standardization on the raw running data to obtain standardized data, and sends the standardized data to the analysis tool for analysis and processing.

[0008] In one alternative approach, the step of calling a pre-packaged analysis tool from the tool server according to the tool list to analyze and process the standardized data, obtain standardized analysis results, and transmit them to the AI ​​model includes: The post-processing module transmits the standardized data back to the tool layer of the tool server. The standardized data is analyzed and processed by calling the corresponding analysis tools according to the tool list; the analysis tools include at least one of anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools. The outputs of each analysis tool are integrated into the standardized analysis result, and the standardized analysis result is sent to the AI ​​model.

[0009] In one optional approach, the time-series database uses an adaptive multi-level data compression algorithm to store the collected device data, including: A corresponding compression strategy is determined based on the data characteristics of the device data, and the device data is divided into steady-state data, dynamic data, and state data based on the data characteristics; the compression strategy includes: The steady-state data is compressed and stored using incremental encoding combined with run-length encoding; An improved floating-point data compression algorithm is used to compress and store the dynamic data; The discrete state data is compressed and stored using dictionary encoding combined with bitmap compression.

[0010] In one alternative approach, the time-series database employs a hierarchical time-series data model to organize the device's operational data. The hierarchical time-series data model includes a root node layer, a site node layer, a device node layer, a functional group node layer, and a measurement point node layer.

[0011] In one alternative approach, the time-series database employs a three-tiered storage strategy to manage data packets, including: Hot data whose acquisition time is within the first time threshold range from the current time will be stored in a high-speed solid-state drive; Warm data whose acquisition time is within the second time threshold range from the current time will be stored in a regular mechanical hard drive; Cold data whose collection time is within the third time threshold range from the current time will be stored in a distributed object storage system.

[0012] In an alternative approach, the method further includes context-aware multi-turn dialogue: Context-aware multi-turn dialogue stores the data packets corresponding to the previously mentioned structured execution instructions and analysis reports as historical session data; When the received current natural language contains references to historical sessions, the historical session data and the current natural language are combined for intent merging and reference resolution to generate a complete current query intent.

[0013] According to a second aspect of the present invention, an intelligent monitoring system for device status based on an AI large model is provided, comprising: An open-source development platform is used to receive natural language input for queries and send the natural language to an AI model; the natural language includes factory identifiers, equipment identifiers, time ranges, and status query elements; The AI ​​model is used to identify the received natural language execution intent, convert the natural language into structured execution instructions, and send the structured execution instructions to the tool server; the structured execution instructions include: factory identifier, equipment identifier, time range parameters, query requirements, and a list of tools to be invoked; The tool server receives the structured execution instructions, calls pre-packaged query tools in the tool server according to the tool list, queries the time-series database, and obtains the raw operating data of the target device within a specified time range; it then standardizes the raw operating data to obtain standardized data; the raw operating data is a set of data points consisting of timestamps and corresponding values ​​collected by the device's sensors and stored in the database; the tool server is a server program that follows the model context protocol, and based on the model context protocol, the internal functions of the server program are standardized and encapsulated to obtain tools, including the query tool and the analysis tool; it calls pre-packaged analysis tools in the tool server according to the tool list to analyze and process the standardized data, obtains standardized analysis results, and transmits them to the AI ​​model; the analysis tools include anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools; The AI ​​model is also used to perform intelligent analysis based on the received standardized analysis results, generate an analysis report in natural language, and transmit it to the open-source development platform for presentation to the user.

[0014] According to a third aspect of the present invention, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform operations of any of the AI-based big model-based intelligent monitoring methods provided in the first aspect above.

[0015] In this embodiment of the invention, users only need to input natural language for querying through an open-source development platform to initiate a query. Subsequent intent recognition, query planning, data acquisition, analysis and calculation, and report generation are all automatically completed by the AI ​​model and tool server. This achieves automatic conversion from natural language input to standardized tool calls, enabling front-line operation and maintenance personnel to directly obtain hydrogen production equipment status data and analysis results without learning SQL or understanding the underlying database structure. This achieves intelligent monitoring of equipment status, reduces the response time of data analysis requests, and improves work efficiency.

[0016] This invention employs a server-side program conforming to the Model Context Protocol (MCP) as an intermediary bridge between the AI ​​model and the time-series database. Based on the MCP protocol, the internal functions of the server-side program are standardized and encapsulated to obtain query and analysis tools, ensuring standardized interaction between the AI ​​model and the time-series database. Through the unified tool interface specification defined by the MCP protocol, the AI ​​model does not need to concern itself with the specific query language or data format of the time-series database; it only needs to pass parameters according to the standardized tool call format to complete data operations. This allows for modular development, independent iteration, and flexible combination of query and analysis tools, significantly reducing system integration complexity.

[0017] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0018] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating the intelligent monitoring method for device status based on an AI large model provided in Embodiment 1 of the present invention is shown. Figure 2 This diagram illustrates the structure of the intelligent equipment status monitoring method system based on an AI large model provided in Embodiment 2 of the present invention. Figure 3 A schematic diagram of the structure of the computer device provided in Embodiment 3 of the present invention is shown. Detailed Implementation

[0019] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Example 1

[0020] This embodiment provides a method for intelligent monitoring of device status based on a large AI model, executed by a server, workstation, or edge computing device with a deployed monitoring system. For example... Figure 1 As shown, the method includes: S1: The open-source development platform receives natural language input for querying and sends the natural language to the AI ​​model; the natural language includes factory identifier, equipment identifier, time range, and status query elements.

[0021] An open-source development platform is used to provide a user interface that receives natural language queries input by users and displays analysis results. The open-source development platform can build / generate at least one of the following types of user interfaces: a dialog interface running on a web browser, an application interface running on a smart mobile terminal, or an embedded dialog plugin integrated into an industrial monitoring system or manufacturing execution system. Preferably, this open-source development platform is the Dify open-source platform.

[0022] Maintenance personnel can initiate queries by inputting natural language information about the operating status of the target hydrogen production equipment through an open-source development platform. The natural language includes the plant identifier, equipment identifier, time range, and status query elements. These status query elements include at least one of the following: operating status, temperature change, hydrogen purity, and efficiency over a specific time period. For example, the natural language could be: "The operating status of electrolyzer No. 1 in Plant A yesterday afternoon."

[0023] S2: The AI ​​model, based on the received natural language execution intent recognition, converts the natural language into structured execution instructions and sends the structured execution instructions to the tool server; the structured execution instructions include: factory identifier, equipment identifier, time range parameters, query requirements, and a list of tools to be invoked.

[0024] The AI ​​model deployment modes include cloud API mode and local private deployment mode, supporting smooth switching between the two modes based on hardware resource conditions and network connectivity. The cloud API mode calls a large language model for inference services; the local private deployment mode provides local inference services based on an open-source large language model or a fine-tuned model.

[0025] During the query process, calling the large language model for inference services through standard API interfaces can significantly reduce hardware investment while ensuring the data compliance of the output analysis reports and the stability of the query service. As business grows and hardware conditions improve, AI model deployment can be smoothly migrated to a local deployment mode, achieving higher data security and response speed.

[0026] The AI ​​model receives natural language, performs deep semantic analysis, identifies the user's query intent, and extracts entities. For example, if the query intent is identified as "query device operating efficiency" and "detect device malfunctions," the entity includes the device identifier, time range, and query parameters. The extracted entities are then mapped, for example: "Electrolyzer No. 1" is mapped to a standard equipment identifier, such as plant_01.electrolyzer_01; "Yesterday afternoon" should be standardized into specific start and end timestamps, such as 2023-10-26-12:00:00 to 2023-10-26-18:00:00.

[0027] Based on the identified intent and entities, a corresponding list of encapsulation tools is generated. The AI ​​model then encapsulates this information into a structured execution instruction in JSON format and sends it to the tool server. This structured execution instruction includes the device identifier (e.g., “electrolyzer_01”), time range parameters (e.g., “2023-10-26-12:00:00 to 2023-10-26-18:00:00”), query requirements (e.g., “obtain temperature, pressure, and current data and assess health status”), and the required list of encapsulation tools (e.g., “query tool, anomaly detection tool, health status assessment tool”). When multiple plants exist, the corresponding plant identifiers (e.g., “hydrogen production plant”) must also be included.

[0028] S3: The tool server receives the structured execution instruction, calls the pre-packaged query tool in the tool server according to the tool list, queries the time series database, and obtains the raw operating data of the target device within a specified time range; the raw operating data is standardized to obtain standardized data; the raw operating data is a set of data points consisting of timestamps and corresponding values ​​collected by the device sensors and stored in the database; the tool server is a server program that follows the model context protocol, and based on the model context protocol, the internal functions of the server program are standardized and encapsulated to obtain the tool, which includes the query tool and the analysis tool.

[0029] The Model Context Protocol (MCP) is an open communication protocol standard used to standardize the request format, parameter passing, and result return between AI models and the tool layer in the tool server. The message format uses JSON encoding. The encapsulation tool enables AI models to call through a unified protocol interface without being aware of the underlying database implementation details, achieving zero-learning-cost data access based on natural language queries.

[0030] The tool server consists of a tool layer and an internal module layer. The tools in the tool layer include query tools, anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools.

[0031] The query tool is used to quickly retrieve data for specified devices, parameters, and time ranges based on structured execution instructions, automatically handling time parsing, device location, and data extraction. For example, if a user queries "temperature change of Electrolytic Cell No. 1 yesterday," the query tool automatically parses the time "yesterday" as "from 00:00 on March 26, 2026 to 23:59 on March 26, 2026," locates the device "electrolyzer_01," and extracts the "pile temperature" data under the "temperature group."

[0032] Furthermore, S3 includes: The query tool is invoked, and its internal modules—parameter validation module, query optimization decision module, database execution module, and post-processing module—are sequentially linked for processing, including: The parameter verification module is invoked to verify the factory identifier, the equipment identifier, and the time range parameters. The verification includes the validity and completeness of the parameters. For example, it verifies whether "electrolyzer_01" exists in the "hydrogen production plant" equipment list, whether the time range is within the valid historical data interval, and whether the user has permission to query the device when the user query has permission restrictions.

[0033] If the verification passes, the query optimization decision module is invoked to generate a query strategy based on the query requirements. The query strategy includes direct query, downsampling query, and query by invoking a pre-computed materialized view.

[0034] The query optimization decision module automatically selects a downsampling strategy based on the target data frequency range of 1Hz to 1000Hz. High-frequency data (such as vibration sensor data at 1000Hz) uses aggregated downsampling, while low-frequency data (such as temperature sensor data at 1Hz) maintains its original resolution. Simultaneously, it determines whether to perform a direct query, a downsampling query, or use a pre-calculated materialized view based on the size of the query data.

[0035] The time series database execution module is invoked to send a query request to the time series database according to the query strategy.

[0036] The database execution module establishes a connection with the time-series database and initiates a query call. The time-series database is the Apache IoTDB time-series database, which organizes data using a hierarchical time-series data model. The hierarchical time-series data model includes a root node layer, a factory instance layer, a device instance layer, a measurement point grouping layer, and a physical measurement point layer. The path addressing format is root node identifier.factory identifier.device identifier.measurement point group identifier.physical measurement point identifier.

[0037] For example, "root.hydrogen production plant.electrolyzer No. 1.temperature group.pile temperature" uniquely identifies this measuring point. The database execution module generates IoTDB query statements, such as: SELECT AVG(temperature) FROM root.hydrogenplant.electrolyzer1.temperaturegroup.piletemperature WHERE time>'2026-03-27T00:00:00' GROUP BY([2026-03-27T00:00:00, 2026-03-27T23:59:59),1m), and send this IoTDB query statement to the time series database.

[0038] The post-processing module receives the returned raw running data, performs basic cleaning and format standardization on the raw running data to obtain standardized data, and sends the standardized data to the analysis tool for analysis and processing.

[0039] For example, when a user queries "temperature change of Electrolyzer No. 1 yesterday", the query tool automatically parses the time "yesterday" as "2026-03-26T00:00 to 2026-03-26T23:59", locates the device "electrolyzer_01", and extracts the "pile temperature" data under "temperature group". The query optimization decision module automatically selects a downsampling strategy based on the target data frequency range of 1 Hz to 1000 Hz, using aggregation downsampling for high-frequency data and maintaining the original resolution for low-frequency data. The database execution module generates an IoTDB query statement such as "SELECT AVG(temperature) FROM root.hydrogen plant.1 electrolyzer.temperature group.pile temperature WHERE time>'2026-03-27T00:00:00' GROUP BY". ([2026-03-27T00:00:00,2026-03-27T23:59:59),1m)” and send it to the time series database; the result post-processing module receives the returned raw running data, performs data format standardization and basic cleaning, generates a standardized time series dataset, and obtains standardized data.

[0040] S4: According to the tool list, call the pre-packaged analysis tools in the tool server to analyze and process the standardized data, obtain standardized analysis results, and transmit them to the AI ​​model. The analysis tools include anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools.

[0041] Specifically, the post-processing module receives the returned raw runtime data, processes it to obtain standardized data, which contains standardized and clean time-series data. The standardized data is then returned to the tool layer for further analysis and processing by analysis tools. The specific analysis tools invoked are determined based on the tool list in the structured execution instructions.

[0042] S4 includes: The post-processing module transmits the standardized data back to the tool layer of the tool server. The standardized data is analyzed and processed by calling the corresponding analysis tools according to the tool list; the analysis tools include at least one of anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools. The outputs of each analysis tool are integrated into the standardized analysis result, and the standardized analysis result is sent to the AI ​​model.

[0043] Each analysis tool receives a standardized time-series dataset generated by the query tool as input, and sequentially calls the internal processing modules to perform the following: calling the parameter verification module to verify the integrity of the input data; calling the query optimization decision module to determine the analysis strategy; calling the database execution module to obtain supplementary data; and calling the result post-processing module to perform algorithm calculations and result formatting, generating standardized analysis results corresponding to each tool.

[0044] The anomaly detection tool integrates multiple detection algorithms. It automatically selects the optimal algorithm based on the characteristics of the input standardized time-series dataset, outputs the anomaly detection results, and returns the anomaly detection results to the tool layer.

[0045] Specifically, the anomaly detection tool integrates multiple detection algorithms. It automatically selects the optimal algorithm based on the characteristics of the input standardized time-series dataset, performs anomaly labeling, and supports sensitivity adjustment to balance false positives and false negatives. These detection algorithms include statistical methods (such as the 3-sigma rule, Cumulative Sum Control Chart (CUSUM)), the Isolation Forest algorithm, Long Short-Term Memory (LSTM) autoencoders, and the Prophet algorithm. After each detector independently calculates anomaly scores, the system dynamically adjusts the weights of each algorithm based on historical accuracy, obtaining the final anomaly score through a weighted average. Anomaly determination is then performed using a dynamic threshold (the 99th percentile of historical scores). If all algorithm conclusions are consistent, the confidence level is high; if most are consistent, the confidence level is medium; otherwise, the confidence level is low and requires manual review. The anomaly detection tool outputs standard time-series data with anomaly labels.

[0046] The health assessment tool performs multi-dimensional calculations based on standardized data and / or the anomaly detection results to obtain a comprehensive health score, generates maintenance recommendations, and returns the comprehensive health score and the maintenance recommendations to the tool layer.

[0047] Specifically, the health assessment tool calculates a comprehensive score from multiple dimensions, including stability, degradation level, anomaly frequency, and energy efficiency, and generates specific maintenance recommendations. The operational stability score is calculated based on the coefficient of variation of each dimension's parameters, with a weighting of 40%; the performance degradation level score compares current efficiency with a historical baseline, with a weighting of 30%; the anomaly frequency score counts the number of anomaly detections within the assessment period and weights them according to severity, with a weighting of 20%; and the energy efficiency level score calculates the ratio of actual energy consumption to theoretical energy consumption, with a weighting of 10%. The comprehensive health index is calculated using a weighted formula and divided into four levels: Excellent (≥90 points), Good (≥75 points), Average (≥60 points), and requires maintenance (<60 points). Maintenance recommendations are automatically generated based on the health index and trends.

[0048] The trend prediction tool predicts the operating trend for future periods based on standardized data, outputs the predicted value and confidence interval, and returns the predicted value and the confidence interval to the tool layer.

[0049] Specifically, the trend prediction tool adopts a hybrid architecture of long short-term memory network, attention mechanism, and temporal convolutional network. It captures long-term temporal dependencies through bidirectional long short-term memory network layers, focuses on key time step features through multi-head attention layers, extracts local temporal patterns through temporal convolutional network layers, and alleviates gradient vanishing through residual connections. The training uses a composite loss function of mean squared error and trend loss, uses random inactivation as a Bayesian approximation to quantify uncertainty, estimates the prediction distribution through multiple inferences, and calculates the prediction confidence interval.

[0050] The visualization generation tool automatically selects the chart format based on the data type and analysis purpose, transforming the numerical data obtained from various tools into an intuitive visual presentation. Trend analysis uses time-series line charts with confidence intervals and outlier annotations; multi-device comparisons use multi-series line charts or heatmaps; data distribution analysis uses histograms or box plots; and parameter correlation analysis uses scatter plots and correlation coefficient heatmaps.

[0051] For example, the anomaly detection tool is invoked to perform statistical threshold detection and the Isolation Forest algorithm, marking abnormal time points and anomaly types, and outputting "standard time series data with anomaly labels" and the anomaly type; the health assessment tool is invoked to calculate multi-dimensional scores for stability, degradation, anomaly frequency, and energy efficiency, outputting a comprehensive health score and maintenance recommendations, such as "It is recommended to pay attention to temperature fluctuations at 3 pm"; the trend prediction tool is invoked to predict the temperature trend for the next 72 hours, outputting the predicted value and a 95% confidence interval. The post-processing modules of each analysis tool return the corresponding output to the tool layer.

[0052] S5: The AI ​​model performs intelligent analysis based on the received standardized analysis results, generates an analysis report in natural language, and transmits it to the open-source development platform for presentation to the user.

[0053] The AI ​​model performs intelligent analysis based on the received standardized analysis results and generates an analysis report. This intelligent analysis includes data interpretation, health assessment, trend prediction, and maintenance suggestion generation. For example, the data interpretation might state "Stable operation today, good health," the health assessment might be "87 points, good level," the trend prediction might be "Temperature will rise steadily over the next 72 hours," and the maintenance suggestion might be "Pay attention to temperature fluctuations at 3 PM, and check the cooling system if necessary."

[0054] For example, interpreting the data as "operating smoothly today, with good health", assessing the health as "87 points, good level", predicting the trend as "temperature will rise steadily in the next 72 hours", and generating maintenance suggestions as "it is recommended to pay attention to temperature fluctuations at 3 pm and check the cooling system if necessary".

[0055] This implementation method achieves standardized encapsulation and docking between large AI models and industrial time-series databases through the Model Context Protocol. Based on the MCP protocol, it standardizes the execution protocol encapsulation of the raw data access interface, anomaly detection algorithm interface, and health assessment calculation interface in the server program for accessing the time-series database. This allows AI models to call through a unified protocol interface without needing to be aware of the underlying database implementation details, achieving zero-learning-cost data access based on natural language. This significantly reduces the technical threshold for the operation and maintenance of hydrogen production equipment and improves the level of intelligent equipment management.

[0056] In one implementation, the intelligent monitoring method for device status based on an AI large model further includes: Context-aware multi-turn dialogue stores the data packets corresponding to the previously mentioned structured execution instructions and analysis reports as historical session data; When the received current natural language contains references to historical sessions, the historical session data and the current natural language are combined for intent merging and reference resolution to generate a complete current query intent.

[0057] For example, if a user first asks about the temperature of electrolytic cell 1, and after receiving the analysis report, the user then asks about the pressure, the AI ​​model can understand that "pressure" here still refers to the pressure data of electrolytic cell 1. In this way, users can ask follow-up questions without repeating previously mentioned equipment information, greatly improving the naturalness and convenience of the interaction.

[0058] Furthermore, the time-series database uses a hierarchical time-series data model to organize the operating data of the device. The hierarchical time-series data model includes a root node layer, a site node layer, a device node layer, a functional group node layer, and a measurement point node layer.

[0059] Specifically, the hierarchical time-series data model is a five-level tree-like hierarchical structure: the first level is the root node (e.g., "root"), which serves as the global starting node for data organization; the second level is the site nodes (e.g., "factory_A", "factory_B"), which are attached to the root node and used to identify factories or sites; the third level is the equipment nodes (e.g., "electrolyzer_01", "compressor_01"), which are attached to the site nodes and used to identify individual equipment; the fourth level is the functional grouping nodes (e.g., "temperature", "pressure", "flow", "electrical"), which are attached to the equipment nodes and used to group measurement points according to functional categories; and the fifth level is the measurement point nodes (e.g., "hd_pressure", "flow"), which are attached to the functional grouping nodes and used to store the time-series data of specific data acquisition points.

[0060] The path addressing format is: root node identifier.site identifier.device identifier.functional group identifier.measurement point identifier, for example, "root.factory_Beijing.electrolyzer_01.pressure.hd_pressure".

[0061] The five-level tree structure makes the data organization not only in line with the thinking habits of engineers (organized in the hierarchy of factory-equipment-function-measurement point), but also facilitates efficient system querying (quickly locate data through path prefixes), and has good scalability (new equipment or new measurement points can be dynamically added).

[0062] This invention employs a five-level tree-structured hierarchical organization of time-series data. Through a layered design of root nodes, site nodes, device nodes, functional group nodes, and measurement point nodes, it achieves semantic and structured data organization. On one hand, each level of nodes is named using identifiers with business semantics, making the data path intuitively reflect the physical location and functional attributes of the equipment, conforming to the thinking habits of industrial operations and maintenance personnel, and reducing the cognitive burden of data understanding and use. On the other hand, the tree-structured path addressing mechanism allows data queries to be quickly located through path prefixes, avoiding full table scans and significantly improving query efficiency. Practical tests show that after adopting this hierarchical model, the response time for complex queries is shortened compared to the traditional flat model, and the integration time for adding new equipment during system expansion is reduced from several days to several hours.

[0063] Furthermore, the time-series database uses an adaptive multi-level data compression algorithm to store the collected device data, including: A corresponding compression strategy is determined based on the data characteristics of the device data, and the device data is divided into steady-state data, dynamic data, and state data based on the data characteristics; the compression strategy includes: The steady-state data is compressed and stored using incremental encoding combined with run-length encoding; An improved floating-point data compression algorithm is used to compress and store the dynamic data; The discrete state data is compressed and stored using dictionary encoding combined with bitmap compression.

[0064] Specifically, the time-series database automatically analyzes data characteristics when writing data: for slow-changing steady-state data (such as equipment temperature, ambient humidity, etc.), the differences between adjacent data points are small. Incremental encoding (DeltaEncoding) is used to record the amount of data change, and run-length encoding (Run-Length Encoding) is used to compress repeated incremental values. The compression ratio between the collected original running data and the compressed original running data can reach 50-100:1.

[0065] For dynamic data with large fluctuations (such as current, power, vibration amplitude, etc.), an improved Gorilla floating-point data compression algorithm is adopted, which records the differences between adjacent data points through XOR operation. The compression ratio between the collected original running data and the compressed original running data can reach 10-30:1.

[0066] For discrete status data (such as device mode, alarm status, switch signals, etc.), dictionary encoding is used to map status values ​​to short codes, which are then stored in conjunction with bitmap compression. The compression ratio between the collected raw operating data and the compressed raw operating data can reach 20-50:1. The entire compression process is completely transparent to the user and requires no manual intervention.

[0067] This invention adaptively selects the optimal compression strategy based on data characteristics, achieving a balance between compression efficiency and accuracy. For hydrogen production equipment monitoring scenarios, the data characteristics of different parameters vary significantly—steady-state data such as temperature change slowly, suitable for high compression ratio algorithms; dynamic data such as current fluctuate greatly, requiring compression algorithms that maintain accuracy; and state data is discrete and has a limited value range, suitable for dictionary-based compression. Through adaptive multi-level compression, the overall system compression ratio reaches over 16:1, significantly reducing storage costs while ensuring the stored data accuracy fully meets analytical requirements.

[0068] Furthermore, the time-series database employs a three-tiered storage strategy to manage data packets, including: Hot data whose acquisition time is within the first time threshold range from the current time will be stored in a high-speed solid-state drive; Warm data whose acquisition time is within the second time threshold range from the current time will be stored in a regular mechanical hard drive; Cold data whose collection time is within the third time threshold range from the current time will be stored in a distributed object storage system.

[0069] Specifically, based on the timeliness and access frequency of the data, it is divided into three levels: Hot data from the most recent week (first time threshold) is stored in high-speed solid-state drives (SSDs), with query response times in the millisecond range, meeting the high-frequency access needs of real-time monitoring and recent trend analysis; Warm data from 8-90 days ago (second time threshold) is automatically migrated to ordinary hard disk drives (HDDs), with query times in the seconds range, used for periodic queries such as monthly analysis and weekly report generation; Cold data older than 90 days (third time threshold) is archived in a lower-cost distributed object storage system (such as MinIO, Ceph, etc.) for long-term storage, annual comparative analysis, and compliance auditing. The time-series database automatically performs data migration tasks every morning at midnight, migrating cold data from SSDs to HDDs according to preset strategies, and then migrating warm data from HDDs to object storage, all without manual intervention.

[0070] This invention employs a three-tiered storage strategy, dynamically adjusting the storage medium based on data timeliness and access frequency to achieve an optimal balance between query performance and storage cost. Hot data is stored on SSDs, ensuring millisecond-level response times for real-time queries; warm data is stored on HDDs, meeting periodic query needs at a lower cost; and cold data is archived to object storage, significantly reducing long-term storage costs. This tiered storage strategy reduces system storage costs by approximately 70% compared to an all-SSD solution, while maintaining unaffected hot data query performance. In practical applications, terabytes of historical data can be stored long-term at low cost, fully meeting the hydrogen production industry's regulatory requirements for data retention of 5-10 years. Furthermore, archived cold data can still be quickly loaded and analyzed when needed, enabling in-depth data value mining. Example 2

[0071] like Figure 2 As shown, this embodiment 3 presents a schematic diagram of the system structure of the intelligent equipment status monitoring method based on an AI large-scale model, which is based on the intelligent equipment status monitoring method based on an AI large-scale model provided in Embodiment 1. Figure 3 As shown, the system includes: an open-source development platform 100, an AI model 200, and a tool server 300. The open-source development platform 100 is used to receive natural language input for querying and send the natural language to the AI ​​model; the natural language includes factory identifier, equipment identifier, time range, and status query elements; AI model 200 is used to recognize the received natural language execution intent, convert the natural language into structured execution instructions, and send the structured execution instructions to the tool server; the structured execution instructions include: factory identifier, equipment identifier, time range parameters, query requirements, and a list of tools to be invoked; The tool server 300 is used to receive the structured execution instructions, call the pre-packaged query tools in the tool server according to the tool list, query the time series database, and obtain the raw operating data of the target device within a specified time range; perform standardization processing on the raw operating data to obtain standardized data; the raw operating data is a set of data points consisting of timestamps and corresponding values ​​collected by the device sensors and stored in the database; the tool server is a server program that follows the model context protocol, and based on the model context protocol, the internal functions of the server program are encapsulated and standardized to obtain tools, including the query tool and the analysis tool; according to the tool list, the pre-packaged analysis tools in the tool server are called to analyze and process the standardized data to obtain standardized analysis results, and transmit them to the AI ​​model; the analysis tools include anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools; The AI ​​model is also used to perform intelligent analysis based on the received standardized analysis results, generate an analysis report in natural language, and transmit it to the open-source development platform for presentation to the user.

[0072] Furthermore, the tool server 300 is also used to call the query tool, sequentially linking the parameter verification module, query optimization decision module, database execution module, and post-processing module within the tool server for processing, including: The parameter verification module is invoked to verify the factory identifier, the equipment identifier, and the time range parameter. The verification content includes the legality and completeness of the parameters. If the verification passes, the query optimization decision module is invoked to generate a query strategy based on the query requirements. The query strategy includes direct query, downsampling query, and query by invoking a pre-computed materialized view. The time-series database execution module is invoked to send a query request to the time-series database according to the query strategy; The post-processing module receives the returned raw running data, performs basic cleaning and format standardization on the raw running data to obtain standardized data, and sends the standardized data to the analysis tool for analysis and processing.

[0073] Furthermore, the tool server 300 is also used for the post-processing module to transmit the standardized data back to the tool layer of the tool server; to call the corresponding analysis tools according to the tool list to analyze and process the standardized data; the analysis tools include at least one of anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools; to integrate the output results of each analysis tool into the standardized analysis result, and to send the standardized analysis result to the AI ​​model.

[0074] In one implementation, the AI ​​model 200 is also used for context-aware multi-turn dialogue, storing the data packets corresponding to the previous structured execution instructions and analysis reports as historical session data; when the received current natural language contains references to the historical session, the historical session data and the current natural language are merged and the references are resolved to generate a complete current query intent.

[0075] In one implementation, the system further includes a time-series database, which is used to provide raw runtime data to the tool server 300.

[0076] The time-series database uses a hierarchical time-series data model to organize the operating data of the equipment. The hierarchical time-series data model includes a root node layer, a site node layer, an equipment node layer, a functional group node layer, and a measurement point node layer.

[0077] Furthermore, the time-series database uses an adaptive multi-level data compression algorithm to store the collected device data, including: A corresponding compression strategy is determined based on the data characteristics of the device data, and the device data is divided into steady-state data, dynamic data, and state data based on the data characteristics; the compression strategy includes: The steady-state data is compressed and stored using incremental encoding combined with run-length encoding; An improved floating-point data compression algorithm is used to compress and store the dynamic data; The discrete state data is compressed and stored using dictionary encoding combined with bitmap compression.

[0078] Furthermore, the time-series database employs a three-tiered storage strategy to manage data packets, including: Hot data whose acquisition time is within the first time threshold range from the current time will be stored in a high-speed solid-state drive; Warm data whose acquisition time is within the second time threshold range from the current time will be stored in a regular mechanical hard drive; Cold data whose collection time is within the third time threshold range from the current time will be stored in a distributed object storage system.

[0079] The AI-based intelligent equipment status monitoring system provided in this embodiment of the invention allows users to initiate queries by inputting natural language through an open-source development platform. Subsequent intent recognition, query planning, data acquisition, analysis and calculation, and report generation are all automatically completed by the AI ​​model and tool server. This achieves automatic conversion from natural language input to standardized tool calls, enabling frontline maintenance personnel to directly obtain hydrogen production equipment status data and analysis results without needing to learn SQL or understand the underlying database structure. This achieves intelligent monitoring of equipment status, reduces response time for data analysis requests, and improves work efficiency.

[0080] By employing a server-side program that conforms to the Model Context Protocol (MCP) as an intermediary bridge between the AI ​​model and the time-series database, and by standardizing and encapsulating the internal functions of the server-side program based on the MCP protocol to obtain query and analysis tools, standardized interaction between the AI ​​model and the time-series database is ensured. Through the unified tool interface specification defined by the MCP protocol, the AI ​​model does not need to know the specific query language or data format of the time-series database; it only needs to pass parameters according to the standardized tool calling format to complete data operations. This allows query and analysis tools to be developed modularly, iterated independently, and flexibly combined, significantly reducing system integration complexity. Example 3

[0081] Figure 3 The diagram shows a structural schematic of a computer device provided in Embodiment 3 of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0082] like Figure 3 As shown, the computer device may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0083] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408. Communication interface 404 is used for communication with other network elements such as clients or other servers. The processor 402 executes program 410, specifically performing the relevant steps described above in any of the AI-based large-scale model-based intelligent device status monitoring method embodiments provided in Example 1.

[0084] Specifically, program 410 may include program code, which includes computer-executable instructions.

[0085] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The device may include one or more processors of the same type, such as one or more CPUs; or it may include processors of different types, such as one or more CPUs and one or more ASICs.

[0086] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0087] Specifically, program 410 can be called by processor 402 to cause the computer device to perform the following operations: This invention provides a computer-readable storage medium storing at least one executable instruction that, when executed on a computer device, causes the computer device to perform the AI-based large model-based intelligent device status monitoring method described in any of the above method embodiments.

[0088] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of the present invention are not directed to any particular programming language. It should be understood that the content of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0089] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0090] Similarly, it should be understood that, in order to streamline the invention and aid in understanding one or more of the various aspects of the invention, features of the embodiments of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the above description of exemplary embodiments of the invention. However, this disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim.

[0091] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0092] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. A method for intelligent monitoring of equipment status based on an AI large-scale model, characterized in that, The method includes: The open-source development platform receives natural language input for querying and sends the natural language to the AI ​​model; the natural language includes factory identifiers, equipment identifiers, time ranges, and status query elements; The AI ​​model recognizes the received natural language execution intent, converts the natural language into structured execution instructions, and sends the structured execution instructions to the tool server. The structured execution instructions include: factory identifier, equipment identifier, time range parameters, query requirements, and a list of tools to be invoked. The tool server receives the structured execution instruction, calls the pre-packaged query tool in the tool server according to the tool list, queries the time series database, and obtains the raw operating data of the target device within a specified time range; the raw operating data is standardized to obtain standardized data; the raw operating data is a set of data points consisting of timestamps and corresponding values ​​collected by the device sensors and stored in the database; the tool server is a server program that follows the model context protocol, and the internal functions of the server program are standardized and encapsulated based on the model context protocol to obtain the tool, which includes the query tool and the analysis tool; According to the tool list, the pre-packaged analysis tools in the tool server are called to analyze and process the standardized data, obtain standardized analysis results, and transmit them to the AI ​​model. The analysis tools include anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools. The AI ​​model performs intelligent analysis based on the received standardized analysis results, generates an analysis report in natural language, and transmits it to the open-source development platform for presentation to the user.

2. The intelligent equipment status monitoring method based on an AI large model according to claim 1, characterized in that, The deployment mode of the AI ​​model includes at least one of cloud API mode and local private deployment mode, and the deployment mode of the AI ​​model can be smoothly switched between the two modes.

3. The intelligent equipment status monitoring method based on an AI large model according to claim 1, characterized in that, The tool server receives the structured execution instruction, calls a pre-packaged query tool in the tool server according to the tool list, queries the time-series database, and obtains the raw operating data of the target device within a specified time range; the raw operating data is then standardized to obtain standardized data, including: The query tool is invoked, and its internal server-side parameter verification module, query optimization decision module, database execution module, and post-processing module are sequentially linked for processing, including: The parameter verification module is invoked to verify the factory identifier, the equipment identifier, and the time range parameter. The verification content includes the legality and completeness of the parameters. If the verification passes, the query optimization decision module is invoked to generate a query strategy based on the query requirements. The query strategy includes direct query, downsampling query, and query by invoking a pre-computed materialized view. The time-series database execution module is invoked to send a query request to the time-series database according to the query strategy; The post-processing module receives the returned raw running data, performs basic cleaning and format standardization on the raw running data to obtain standardized data, and sends the standardized data to the analysis tool for analysis and processing.

4. The intelligent equipment status monitoring method based on an AI large model according to claim 3, characterized in that, The step of calling a pre-packaged analysis tool in the tool server according to the tool list to analyze and process the standardized data, obtain standardized analysis results, and transmit them to the AI ​​model includes: The post-processing module transmits the standardized data back to the tool layer of the tool server. The standardized data is analyzed and processed by calling the corresponding analysis tools according to the tool list; the analysis tools include at least one of anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools. The outputs of each analysis tool are integrated into the standardized analysis result, and the standardized analysis result is sent to the AI ​​model.

5. The intelligent monitoring method for equipment status based on an AI large model according to claim 1, characterized in that, The time-series database uses an adaptive multi-level data compression algorithm to store the collected device data, including: A corresponding compression strategy is determined based on the data characteristics of the device data, and the device data is divided into steady-state data, dynamic data, and state data based on the data characteristics; the compression strategy includes: The steady-state data is compressed and stored using incremental encoding combined with run-length encoding; An improved floating-point data compression algorithm is used to compress and store the dynamic data; The discrete state data is compressed and stored using dictionary encoding combined with bitmap compression.

6. The intelligent equipment status monitoring method based on an AI large model according to claim 1, characterized in that, The time-series database uses a hierarchical time-series data model to organize the operating data of the equipment. The hierarchical time-series data model includes a root node layer, a site node layer, an equipment node layer, a functional group node layer, and a measurement point node layer.

7. The intelligent equipment status monitoring method based on an AI large model according to claim 1, characterized in that, The time-series database employs a three-tiered storage strategy to manage data packets, including: Hot data whose acquisition time is within the first time threshold range from the current time will be stored in a high-speed solid-state drive; Warm data whose acquisition time is within the second time threshold range from the current time will be stored in a regular mechanical hard drive; Cold data whose collection time is within the third time threshold range from the current time will be stored in a distributed object storage system.

8. The intelligent monitoring method for equipment status based on an AI large model according to claim 1, characterized in that, The method also includes context-aware multi-turn dialogue: Context-aware multi-turn dialogue stores the data packets corresponding to the previously mentioned structured execution instructions and analysis reports as historical session data; When the received current natural language contains references to historical sessions, the historical session data and the current natural language are combined for intent merging and reference resolution to generate a complete current query intent.

9. A device status intelligent monitoring system based on an AI large-scale model, characterized in that, The system includes: An open-source development platform is used to receive natural language input for queries and send the natural language to an AI model; the natural language includes factory identifiers, equipment identifiers, time ranges, and status query elements; The AI ​​model is used to identify the received natural language execution intent, convert the natural language into structured execution instructions, and send the structured execution instructions to the tool server; the structured execution instructions include: factory identifier, equipment identifier, time range parameters, query requirements, and a list of tools to be invoked; The tool server receives the structured execution instructions, calls pre-packaged query tools in the tool server according to the tool list, queries the time-series database, and obtains the raw operating data of the target device within a specified time range; it then standardizes the raw operating data to obtain standardized data; the raw operating data is a set of data points consisting of timestamps and corresponding values ​​collected by the device's sensors and stored in the database; the tool server is a server program that follows the model context protocol, and based on the model context protocol, the internal functions of the server program are standardized and encapsulated to obtain tools, including the query tool and the analysis tool; it calls pre-packaged analysis tools in the tool server according to the tool list to analyze and process the standardized data, obtains standardized analysis results, and transmits them to the AI ​​model; the analysis tools include anomaly detection tools, health assessment tools, trend prediction tools, and visualization generation tools; The AI ​​model is also used to perform intelligent analysis based on the received standardized analysis results, generate an analysis report in natural language, and transmit it to the open-source development platform for presentation to the user.

10. A computer device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation of the AI-based large model-based intelligent monitoring method for device status as described in any one of claims 1-8.

Images