Equipment state management method and device, electronic equipment, medium and program product

Abstract

The invention provides an equipment state management method and device, electronic equipment, a medium and a program product, relates to application of a large model in an equipment management scene, and can be applied to the technical field of artificial intelligence and the technical field of cloud computing. The method comprises the steps of obtaining multi-source operation data of target equipment, and processing the multi-source operation data to obtain multi-source processing data; inputting the multi-source processing data into a first prediction model and a second prediction model in parallel for state prediction to obtain a first prediction result and a second prediction result; performing integrated analysis on the first prediction result and the second prediction result based on a fusion model to generate a prediction conclusion; and performing equipment state management on the target equipment based on the prediction conclusion.

Description

Technical Field

[0001] This application relates to the application of large models in equipment management scenarios, which can be applied to the fields of artificial intelligence, cloud computing, and financial technology. More specifically, it relates to a method, device, equipment, medium, and program product for equipment status management. Background Technology

[0002] With the rapid development of cloud computing and big data, data center scale is constantly expanding. The operational stability and energy efficiency management of infrastructure, including air conditioning, power supply and distribution, and server racks, are crucial to overall reliability and operating costs. However, existing technologies largely rely on threshold alarms and manual inspections, which suffer from drawbacks such as strong lag and insufficient predictive capabilities, making it difficult to detect potential risks in a timely manner. Meanwhile, the large amount of multi-source heterogeneous data generated by data center operations has characteristics such as high dimensionality, nonlinearity, and temporal sequence. Traditional methods cannot fully explore its deep patterns, leading to inaccurate equipment status predictions, insufficient energy efficiency optimization, and maintenance relying on manual experience, resulting in low efficiency and failing to meet the demands of modern data centers for high reliability, low energy consumption, and intelligent operation and maintenance. Summary of the Invention

[0003] In view of the above problems, this application provides equipment status management methods, apparatus, equipment, media and program products.

[0004] According to a first aspect of this application, a device status management method is provided, the method comprising: acquiring multi-source operational data of a target device; processing the multi-source operational data to obtain multi-source processed data; inputting the multi-source processed data in parallel into a first prediction model and a second prediction model for status prediction to obtain a first prediction result and a second prediction result, wherein the first prediction model is used to model the time dependency and global features of the multi-source processed data, and the second prediction model is used to perform inference analysis in combination with the multi-source processed data and device operation and maintenance context information; integrating and analyzing the first prediction result and the second prediction result based on a fusion model to generate a prediction conclusion; and performing device status management on the target device based on the prediction conclusion.

[0005] According to an embodiment of this application, the step of modeling the temporal dependency and global features of the multi-source processed data using the first prediction model includes: performing temporal dependency modeling on the time series indicators in the multi-source processed data and outputting a first feature representation; performing global context modeling on the multi-source processed data using an attention mechanism and outputting a second feature representation; and inputting the first feature representation and the second feature representation into a fusion layer to output the first prediction result.

[0006] According to an embodiment of this application, the step of using the second prediction model to perform reasoning analysis in conjunction with the multi-source processing data and equipment operation and maintenance context information includes: concatenating the multi-source processing data and the equipment operation and maintenance context information to construct a structured text representation; and guiding the second prediction model to perform nonlinear relationship reasoning and causal analysis based on the structured text representation, and outputting the second prediction result.

[0007] According to an embodiment of this application, the training process of the first prediction model includes a process of hyperparameter optimization using a random forest model, specifically including: defining a hyperparameter search space; generating multiple candidate hyperparameter configurations based on the hyperparameter search space using random sampling, inputting the candidate hyperparameter configurations into the first prediction model for training and validation to obtain corresponding performance indicators and form an observation dataset; establishing a mapping relationship between the candidate hyperparameter configurations and the performance indicators using a random forest model, calculating the expected improvement value based on the mapping relationship, and determining the updated candidate hyperparameter configuration based on the expected improvement value; inputting the updated candidate hyperparameter configuration into the first prediction model for training and validation to update the observation dataset; and repeating the process of establishing the mapping relationship, determining the updated candidate hyperparameter configuration, and updating the observation dataset until a preset stopping condition is met, and selecting a hyperparameter configuration whose performance meets the set conditions from the observation dataset as the target parameter configuration.

[0008] According to an embodiment of this application, the second prediction result includes: a quantitative prediction value, used to represent the trend of changes in the operating parameters of the target device within a future preset time period; and qualitative analysis information, used to represent the risk factors that cause the changes in the operating parameters and the corresponding inference links.

[0009] According to an embodiment of this application, acquiring multi-source operating data of the target device includes: collecting a first indicator of the multi-source operating data based on a first preset frequency, the first indicator including at least one of temperature, humidity, power, current, and central processing unit; collecting a second indicator of the multi-source operating data based on a second preset frequency, the second indicator including at least one of device logs, network traffic, and access control data, wherein the first preset frequency is higher than the second preset frequency; and collecting a third indicator of the multi-source operating data in real time, the third indicator including at least one of alarm signals and access control switch events.

[0010] According to an embodiment of this application, device status management of the target device based on the prediction conclusion includes: comparing the prediction conclusion with an absolute threshold and a dynamic threshold respectively to generate a first comparison result and a second comparison result; generating a status determination signal based on the first comparison result, the second comparison result and the qualitative analysis information; and performing hierarchical management operations according to the status determination signal.

[0011] According to an embodiment of this application, the step of integrating and analyzing the first prediction result and the second prediction result based on the fusion model to generate a prediction conclusion includes: when the first prediction result and the second prediction result indicate that the target device has the same risk, outputting a consistency conclusion and generating corresponding alarm information; when the first prediction result and the second prediction result differ, outputting a divergence conclusion and generating a prompt message; when both the first prediction result and the second prediction result indicate that the target device is in a low-load state, outputting a resource optimization conclusion and generating a resource allocation adjustment suggestion.

[0012] According to an embodiment of this application, the step of integrating and analyzing the first prediction result and the second prediction result based on the fusion model to generate a prediction conclusion includes: obtaining confidence information corresponding to the first prediction result and the second prediction result; calculating a prediction uncertainty index based on the confidence information; performing weighted fusion of the prediction uncertainty index and the equipment operation and maintenance context information; and matching a predefined prediction conclusion based on the weighted fusion result.

[0013] According to embodiments of this application, the device operation and maintenance context information includes at least one of historical time-series data, device metadata, work order status, and historical alarm records.

[0014] A second aspect of this application provides an equipment status management device, the device comprising: a data acquisition module, configured to: acquire multi-source operating data of a target equipment, process the multi-source operating data, and obtain multi-source processed data; a status prediction module, configured to: input the multi-source processed data in parallel into a first prediction model and a second prediction model to perform status prediction, and obtain a first prediction result and a second prediction result, wherein the first prediction model is used to model the time dependency and global features of the multi-source processed data, and the second prediction model is used to perform inference analysis in combination with the multi-source processed data and equipment operation and maintenance context information; an integration analysis module, configured to: integrate and analyze the first prediction result and the second prediction result based on a fusion model to generate a prediction conclusion; and an equipment status management module, configured to: perform equipment status management of the target equipment based on the prediction conclusion.

[0015] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0016] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0017] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0018] According to embodiments of this application, by introducing a collaborative mechanism of dual prediction models, both the time-dependent characteristics and global correlation characteristics of equipment operation data can be considered simultaneously, and causal reasoning can be performed by combining operation and maintenance context information, thereby improving the accuracy and interpretability of prediction results. Furthermore, by integrating and analyzing different prediction results through a fusion model, the risk of misjudgment caused by the bias of a single model can be effectively reduced, improving the reliability of prediction conclusions. Attached Figure Description

[0019] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustrations depict application scenarios of the device status management method, apparatus, device, medium, and program product according to embodiments of this application.

[0021] Figure 2 A flowchart illustrating a device status management method according to an embodiment of this application is shown schematically.

[0022] Figure 3 A flowchart illustrating a method for hyperparameter optimization according to some exemplary embodiments of this application is shown schematically.

[0023] Figure 4 The flowchart illustrates a method for inference analysis using a second prediction model according to some exemplary embodiments of this application.

[0024] Figure 5 A schematic diagram illustrating the structure of a device status management apparatus according to an embodiment of this application is shown; and

[0025] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a device state management method according to an embodiment of this application. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] The following will explain and clarify the terms used in this article.

[0031] Long Short-Term Memory (LSTM) networks are recurrent neural networks specifically designed for processing and predicting time series data. LSTMs effectively capture long-term dependencies in time series data by introducing memory units and gating mechanisms.

[0032] Transformer: A deep learning model based on self-attention mechanism. The core of the Transformer model is the self-attention mechanism, which can dynamically assign weights to each element in the sequence, thereby capturing the complex relationships between different elements in the sequence.

[0033] Random forest: an ensemble learning method that improves the accuracy and stability of predictions by building multiple decision trees and voting on them.

[0034] With the rapid development of cloud computing, big data, and artificial intelligence technologies, data center facilities have become a crucial infrastructure for the digital economy and information society. Currently, an increasing number of enterprises and institutions are deploying their core businesses in the cloud, leading to a continuous surge in demand for computing, storage, and networking. This has directly driven the continuous expansion of data center scale and the diversification of equipment types, evolving from traditional single server rooms into massive clusters integrating thousands of servers, hundreds of racks, and complex power supply and cooling systems. In this context, ensuring the stable operation of data center infrastructure and efficient energy management has become a key factor affecting the overall performance and operating costs of data centers.

[0035] Data center infrastructure typically includes air conditioning systems, power supply and distribution systems, server racks, network switching equipment, fire protection and security systems, and environmental monitoring equipment. Among these, the air conditioning system controls temperature and humidity and ensures efficient heat dissipation; power supply equipment, including uninterruptible power supplies (UPS), distribution cabinets, and generators, determines power reliability; and server racks and computing devices handle core business processing tasks, requiring a strict operating environment. As the number and complexity of facilities increase, the coupling between devices becomes increasingly tight. An anomaly in any link can trigger a chain reaction, leading to business interruptions or even severe economic losses.

[0036] In existing technologies, traditional monitoring methods mainly rely on threshold alarms and manual inspections. For example, sensors are deployed on air conditioning systems or power supply equipment, triggering alarm signals when monitored parameters exceed preset thresholds, which are then inspected and addressed by maintenance personnel. Simultaneously, data centers often schedule regular manual inspections to identify potential problems. This approach is effective when the data volume is small, but its limitations become increasingly apparent as the complexity of modern data centers increases. First, threshold alarm mechanisms are lagging; when certain equipment parameters approach a fault state, an alarm may not be triggered until the threshold is reached, often resulting in missed intervention opportunities. Second, the regular inspection model relies on human experience, resulting in slow response times and failing to meet the closed-loop requirements of modern data centers for "real-time detection—predictive early warning—automatic handling."

[0037] Furthermore, data center operations generate diverse and heterogeneous environmental and equipment data. This includes multi-dimensional indicators such as temperature, humidity, airflow velocity, voltage, current, power factor, server utilization, and coolant flow rate. These indicators are not only numerous but also exhibit significant nonlinear relationships and complex time-series dependencies. For instance, the cooling efficiency of an air conditioning system may be closely related to load fluctuations of servers within the rack and power variations of power supply equipment. Simply relying on linear analysis or rule-based judgments is insufficient to reveal these deep-seated coupling patterns. Traditional methods often only allow for threshold setting based on single-dimensional or limited features, failing to fully exploit potential patterns in high-dimensional time-series data, leading to insufficient prediction accuracy.

[0038] Furthermore, modern data centers place even higher demands on energy efficiency management. Data center energy consumption accounts for an increasingly larger share of global energy consumption, a significant portion of which comes from air conditioning cooling and power supply losses. Without in-depth analysis and intelligent optimization of multi-source operational data, data center equipment often operates inefficiently. For example, localized overheating in some server racks may cause air conditioners to operate at high loads for extended periods without achieving global optimized airflow scheduling; similarly, insufficient or excessive redundancy in uninterruptible power supply (UPS) equipment will lead to unreasonable energy utilization and increased operating costs.

[0039] Therefore, the main shortcomings of existing technologies can be summarized as follows: threshold alarm modes can only respond to anomalies that have occurred or are close to the critical point, lacking the ability to predict potential risks in their early stages; traditional methods struggle to build effective models and fail to fully uncover deep-seated patterns when faced with high-dimensional, nonlinear, and strongly time-series data; reliance on manual inspections and experience-based judgments is labor-intensive, and the response speed cannot meet the needs of large-scale real-time operation and maintenance; and the lack of intelligent analysis of the overall correlation between power supply, cooling, and load makes it difficult to achieve refined energy efficiency management.

[0040] Based on this, embodiments of this application provide a device status management method, the method comprising: acquiring multi-source operating data of a target device; processing the multi-source operating data to obtain multi-source processed data; inputting the multi-source processed data in parallel into a first prediction model and a second prediction model for status prediction to obtain a first prediction result and a second prediction result, wherein the first prediction model is used to model the time dependency and global features of the multi-source processed data, and the second prediction model is used to perform inference analysis in combination with the multi-source processed data and device operation and maintenance context information; integrating and analyzing the first prediction result and the second prediction result based on a fusion model to generate a prediction conclusion; and performing device status management on the target device based on the prediction conclusion. According to embodiments of this application, by introducing a collaborative mechanism of dual prediction models, the time dependency and global correlation features of device operating data can be taken into account simultaneously, and causal inference can be performed in combination with operation and maintenance context information, thereby improving the accuracy and interpretability of the prediction results. At the same time, by integrating and analyzing different prediction results through a fusion model, the risk of misjudgment caused by the bias of a single model can be effectively reduced, and the reliability of the prediction conclusion can be improved.

[0041] It should be noted that the device status management method, apparatus, equipment, medium, and program product defined in this application involve the application of large-scale models in device management scenarios and can be used in the fields of artificial intelligence, cloud computing, and fintech, as well as in various other fields besides artificial intelligence, cloud computing, and fintech. The application fields of the device status management method, apparatus, equipment, medium, and program product provided in the embodiments of this application are not limited.

[0042] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0043] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0044] Figure 1 The illustration shows an application scenario diagram of the device status management method, apparatus, device, medium, and program product according to embodiments of this application.

[0045] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0046] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0047] In the embodiments of this application, the first terminal device 101 can be an example of the first device, and the second terminal device 102 and / or the third terminal device 103 can be an example of at least one second device. The first device and the second device can communicate collaboratively through an internal client mechanism to implement the data distribution and rendering logic described in the device state management method.

[0048] In some embodiments, the first device and at least one second device may be different display modules, windows or screens on the same computing terminal (such as a host), or multiple physical devices that work together through a network, such as different client instances deployed on a desktop computer, tablet terminal or mobile device respectively.

[0049] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smart mobile terminals, tablet computers, laptop computers, and desktop computers.

[0050] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0051] It should be noted that the device status management method provided in this application embodiment can generally be executed by server 105. Correspondingly, the device status management device provided in this application embodiment can generally be located in server 105. The device status management method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the device status management device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0052] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0053] The following will be based on Figure 1 The described scene, through Figures 2-4 The device status management method of the disclosed embodiments will be described in detail.

[0054] Figure 2 A flowchart illustrating a device status management method according to an embodiment of this application is shown schematically.

[0055] like Figure 2 As shown, the device status management method 200 of this embodiment includes operations S210 to S240.

[0056] In operation S210, multi-source operating data of the target device is acquired, and the multi-source operating data is processed to obtain multi-source processed data.

[0057] In embodiments of this application, multi-source operating data may include raw data directly acquired from the sensor ports of the device, such as signals generated by temperature sensors, current sensors, vibration sensors, and speed sensors, and the signals are preprocessed to form multi-source processed data that can be used for subsequent analysis.

[0058] In some embodiments, multi-source operational data may also include equipment control command logs, operator input records, historical maintenance records of equipment operation, and scheduling information from the production management system. For example, in large-scale production equipment scenarios, scheduling information can reflect equipment load and production rhythm at different time periods, while maintenance records can include replacement parts, fault records, etc., thereby forming multi-dimensional, multi-source processed data covering the physical layer, operation and maintenance layer, and management layer.

[0059] In some embodiments, multi-source operational data can also be extended to environmental data and external contextual information. For example, in the scenario of predicting the condition of wind power generation equipment, meteorological data (such as wind speed, humidity, and air pressure) and equipment operation data can be collected together and processed uniformly.

[0060] In some embodiments, the acquired multi-source operational data can be pre-processed on a local terminal using edge computing to reduce latency caused by data transmission. For example, in applications involving industrial robots, the robot controller can perform local preprocessing on sensor data to form compressed feature data before transmitting it to a centralized analysis server for subsequent prediction, thereby improving the system's real-time performance and fault tolerance.

[0061] In some embodiments, the acquisition of multi-source operational data can also be achieved by combining it with an IoT platform and a blockchain-based evidence storage system. For example, during operation, the device uploads key data to the blockchain in real time, forming an immutable data record. During the data acquisition process, the system can directly obtain the target device's operational data from the blockchain and cross-validate it with locally collected real-time data to improve data credibility and security.

[0062] In the embodiments of this application, after acquiring multi-source operational data, the multi-source operational data can be cleaned and organized to ensure the validity and consistency of the multi-source processed data. Specifically, redundant information in the multi-source operational data can be identified and deleted first. For example, when duplicate records are generated in the device sensor data due to differences in acquisition frequency, the system can automatically determine the duplication based on the timestamp and device number, and perform a deduplication operation to avoid duplicate samples causing deviations in model training.

[0063] In some embodiments, different processing strategies can be selected based on the proportion of missing data to address the missing value problem. For example, when the proportion of missing values ​​is extremely low, imputation can be performed using the mean or median of neighboring data to maintain data continuity; when the proportion of missing values ​​is high and the missing pattern is random, the sample can be deleted or imputed using interpolation-based methods, such as time series interpolation or inference interpolation based on similar device operating patterns, thereby preserving the overall characteristics of the data as much as possible.

[0064] In some embodiments, anomalies in multi-source operational data can be identified based on the physical constraints of equipment operation. For example, for fan speed, when the value exceeds the design limit or is negative, it can be automatically identified as invalid data and discarded; for parameters such as current and voltage, a reasonable range can be set in combination with the equipment operating status and load conditions to identify abnormal records caused by sensor failure or communication errors.

[0065] In some embodiments, the cleaned and valid data can be standardized to eliminate differences between different physical dimensions and orders of magnitude. Specifically, a standardized score method can be used to transform various feature values ​​into a standard normal distribution with a mean of 0 and a standard deviation of 1. For example, temperature data (range 0–300℃) and power data (range 0–500kW) can be uniformly mapped to a standardized space, enabling the model to converge faster during training and avoiding gradient imbalance caused by differences in different feature dimensions.

[0066] In some embodiments, the normalization method can also be replaced by an interval scaling approach, scaling all feature values ​​to the [0,1] interval to adapt to normalization-sensitive prediction models. Furthermore, for time series data, a sliding window normalization method can be used to dynamically capture short-term fluctuation characteristics.

[0067] In operation S220, the multi-source processing data is input in parallel into the first prediction model and the second prediction model to perform state prediction and obtain the first prediction result and the second prediction result. The first prediction model is used to model the time dependency and global features of the multi-source processing data, and the second prediction model is used to perform inference analysis in combination with the multi-source processing data and equipment operation and maintenance context information.

[0068] In the embodiments of this application, the equipment operation and maintenance context information may include at least one of historical time-series data, equipment metadata, work order status, and historical alarm records. By inputting this information into the prediction model along with multi-source operational data, the comprehensiveness and accuracy of inference analysis can be significantly improved, thereby providing more reliable data support for equipment status prediction and management. The equipment operation and maintenance context information may include historical time-series data, such as continuous monitoring records of temperature, power, pressure, and vibration during past operation, used to reveal long-term operating trends and periodic patterns. Furthermore, the equipment operation and maintenance context information may also include equipment metadata, such as equipment model, rated power, manufacturing date, and service life, used to reflect the basic attributes and design constraints of the equipment. The equipment operation and maintenance context information may also include the status information of maintenance work orders, such as whether there are currently unfinished maintenance tasks, the content and frequency of historical maintenance, and replacement records of key components. By combining this work order information with real-time operational data, the system can more accurately determine the potential causes behind abnormal states. For example, when the prediction model detects a continuous rise in temperature, if the context information indicates "there is an unclosed work order for the cooling system compressor," it is easier to infer that the fault originates from insufficient cooling. Equipment operation and maintenance context information can also include historical alarm records, such as high temperature alarms, overload alarms, or network communication anomaly alarms triggered by the equipment at different stages of operation. By cross-referencing with real-time monitoring data, the system can identify recurring hazard patterns and improve the reliability and foresight of prediction conclusions.

[0069] In the embodiments of this application, the first prediction model can be a deep learning model based on recurrent neural networks, long short-term memory networks, or gated recurrent units, used to capture dependencies and trend changes in time series data, thereby better characterizing the dynamic evolution of equipment operating parameters. Simultaneously, the second prediction model can use a large language model as its core reasoning framework, synchronously receiving semantically transformed equipment operation and maintenance context information from multi-source processed data.

[0070] In some embodiments, the first prediction model may also employ a time-series modeling method based on a self-attention mechanism, such as a Transformer structure, to improve its ability to process data over long time spans. For example, for wind turbine state prediction, the first prediction model based on a self-attention mechanism can extract potential degradation trends from long-term wind speed variations and historical vibration patterns. Simultaneously, the second prediction model can incorporate inference frameworks based on logistic regression, decision trees, or Bayesian networks to analyze equipment operation and maintenance context information and output state judgments that corroborate the first prediction results, thereby avoiding the risks associated with distortion from a single model.

[0071] In some embodiments, the first prediction model can jointly model global features from multiple sources, such as performing multivariate time-series analysis on multidimensional data like temperature, pressure, current, and voltage, to discover potential coupling relationships between different sensors. The second prediction model can incorporate expert rules and domain knowledge to simulate the reasoning methods of human maintenance engineers analyzing equipment anomalies. For example, when equipment power fluctuates within the normal range, but the cooling system maintenance time has far exceeded the recommended cycle, the second prediction model can determine a potential overheating hazard based on contextual information and output this as a prediction result.

[0072] In some embodiments, the first and second prediction models can also feature heterogeneous computing architectures to improve overall system efficiency. For example, the first prediction model is deployed at the edge to capture the time-series characteristics of equipment operation in real time; the second prediction model is deployed in the cloud to perform deep inference by combining historical operational big data with external environmental parameters. Through parallel prediction via edge-cloud collaboration, a state prediction capability that balances high real-time performance and high accuracy can be achieved, thereby ensuring the stability and safety of equipment operation in complex industrial scenarios.

[0073] In some embodiments, the second prediction model can employ a large language model as its core reasoning framework. Specifically, equipment operation and maintenance context information can be input into the large language model in natural language form, enabling it to output analysis results of the equipment's operating status based on its powerful semantic understanding and knowledge association capabilities. For example, when sensor data indicates an abnormally high vibration amplitude, the large language model can combine historical maintenance texts to identify that the anomaly may be related to bearing wear or insufficient lubrication, thereby generating a corresponding status prediction conclusion.

[0074] Furthermore, the second prediction model can be combined with a vector database to achieve rapid retrieval and inference of massive amounts of operation and maintenance documents and experience knowledge. Specifically, the system first converts key indicators from multi-source processed data into query vectors, which are then input into the vector database for similar scenario retrieval. Subsequently, the retrieval results and real-time data are input into the large language model, which performs contextual association analysis. For example, when the input indicators are highly similar to previous failure cases of a specific model, the large language model can refer to the repair conclusions of that case to make inferences about the current device status.

[0075] It should be noted that, in this embodiment, "parallel input" refers to the system simultaneously transmitting multi-source processed data to both the first and second prediction models within the same time window, rather than executing them sequentially. Specifically, after unified preprocessing and format standardization, the multi-source processed data is divided into two input paths: one path enters the first prediction model to model the time dependencies and global trend characteristics of the data; the other path enters the second prediction model to perform semantic reasoning or knowledge reasoning in conjunction with operational context information. Parallel input avoids the delays caused by sequential execution, improving the system's real-time performance; simultaneously, it ensures that the data versions received by the two models are consistent, avoiding prediction result deviations caused by the order of processing.

[0076] In operation S230, the first prediction result and the second prediction result are integrated and analyzed based on the fusion model to generate a prediction conclusion.

[0077] In some embodiments, the fusion model can employ a weighted fusion approach to integrate and analyze the first and second prediction results. Specifically, the system pre-assigns weight factors to the first and second prediction models, for example, dynamically adjusting the weights based on the models' accuracy or recall in historical prediction tasks. Subsequently, the fusion model outputs the final prediction conclusion using a weighted average, thereby ensuring the accuracy of time series modeling while also taking into account the semantic supplementation provided by contextual reasoning, making the prediction results more robust.

[0078] In some embodiments, the fusion model can employ an adaptive fusion approach based on an attention mechanism. This approach dynamically allocates the importance of the first and second prediction results during the integration analysis process. For example, when significant fluctuations occur in real-time sensor data, the weight of the first prediction model is increased; conversely, when operational context information indicates potential risks, the weight of the second prediction model is increased. Through this dynamic attention mechanism, the system can adaptively favor more reliable information sources under different operating scenarios, thereby making the final prediction conclusion more consistent with the actual operating state of the equipment.

[0079] In some embodiments, the fusion model can employ a stacked ensemble learning approach. Specifically, the first and second prediction results can be used as features input to a high-level classification or regression model, which can be logistic regression, gradient boosting trees, or deep neural networks. Through training, this high-level model can learn how to optimally output the predicted state of the device under different combinations of inputs. For example, in complex operating conditions, there may be discrepancies between the trend analysis of the first prediction model and the contextual inference of the second prediction model. The high-level fusion model can automatically learn the nonlinear combination relationship between the two, thereby improving the accuracy and robustness of the prediction.

[0080] In some embodiments, the fusion model can also support the integration of results across time dimensions. Specifically, the system not only integrates the first and second prediction results within the current period, but also combines historical prediction conclusions for trend comparison. For example, when the second prediction model indicates the presence of potential faults in multiple consecutive periods, while the time-series prediction results of the first prediction model show little fluctuation, the fusion model can generate a more cautious prediction conclusion based on the cumulative judgment over time, such as indicating the presence of potential early signs of faults.

[0081] In some embodiments, the fusion model may also employ an interpretative fusion framework enhanced with a large language model or knowledge graph. Specifically, the system can use the first and second prediction results as input prompts, combine historical operational knowledge and experience cases, and generate prediction conclusions with reasoning processes.

[0082] In operation S240, the target device is managed in terms of device status based on the prediction conclusion.

[0083] In some embodiments, the system can perform status monitoring and alarm management on the target device based on the predicted conclusions. When the predicted conclusions output by the fusion model indicate that the device is in a potential abnormal trend, the system can automatically trigger an alarm mechanism, such as by popping up a prompt on the control center interface, sending a mobile notification to maintenance personnel, or generating a voice alarm message, thereby achieving an immediate response to the abnormal state. This status management method can provide early warnings before device failures occur, reducing downtime and economic losses.

[0084] In some embodiments, operation S240 can be used for dynamic control of equipment operating parameters. For example, when a prediction indicates that the equipment is at risk of overload, the system can automatically issue adjustment commands to reduce the equipment's operating power or speed, thereby keeping the equipment within a safe operating range. Conversely, when a prediction indicates that the equipment is stable and has performance redundancy, the system can also appropriately increase the load, thereby improving resource utilization and overall production efficiency. This automated control function makes equipment operation more intelligent and flexible.

[0085] Furthermore, condition management is not limited to real-time control but can be extended to the optimization of maintenance plans. When predictions indicate that a component is showing signs of degradation, the system can automatically generate a maintenance plan based on the component's risk level, such as scheduling advance repairs, replacing parts, or adjusting inspection cycles. Compared to traditional periodic maintenance methods, this prediction-based condition management solution can significantly reduce maintenance costs and extend equipment lifespan.

[0086] In some embodiments, when the prediction results indicate that the operating status of a single device is abnormal, the system can further determine whether the abnormality will have a cascading effect on similar or cooperating devices. For example, in a scenario with multiple parallel generator sets, if one unit malfunctions, the system can automatically adjust the output power of the remaining units based on the prediction results to ensure the stability and safety of the overall system. This group-level state management helps to achieve a higher level of scheduling optimization and risk isolation.

[0087] In other embodiments, the predictions can also drive equipment lifecycle management. The system can classify the health status of equipment based on the prediction results, for example, into levels such as normal, sub-healthy, risky, or severely faulty. Corresponding to different levels, the system can automatically take differentiated management measures, such as only logging, increasing monitoring frequency, reducing usage intensity, or directly shutting down for maintenance. Through this hierarchical management strategy, refined management from operational monitoring to the entire lifecycle can be achieved.

[0088] According to embodiments of this application, by acquiring and processing multi-source operational data, multi-dimensional information such as sensor data, logs, environmental data, and operational context can be fused to improve the consistency and completeness of input data, thereby avoiding information loss due to insufficient data from a single source. A parallel input approach is adopted to simultaneously drive a first prediction model and a second prediction model, enabling the former to fully capture time dependencies and global features, while the latter combines contextual information to complete deep inference. The two complement each other, improving the accuracy and robustness of the prediction results. Furthermore, by integrating and analyzing different prediction results through a fusion model, not only are multiple fusion methods such as weighted, attention, or ensemble learning supported, but it can also dynamically adapt to different operating conditions, achieving adaptive optimization of prediction conclusions. Finally, equipment status management based on prediction conclusions can not only achieve real-time alarms and parameter control, but also extend to preventative maintenance, collaborative management of group equipment, and lifecycle management, significantly improving the safety, stability, and economy of equipment operation.

[0089] The device status management method of this application will be specifically described below by way of preferred embodiments.

[0090] In the embodiments of this application, the first prediction model can adopt a deep learning-based multi-branch structure to better model the temporal dependencies and global features of multi-source processed data. Specifically, the first prediction model includes an LSTM branch and a Transformer branch. The LSTM branch utilizes the memory and forgetting mechanism of gated recurrent units to effectively capture short-term and long-term dependencies in equipment operation data, thereby extracting the temporal evolution patterns and periodic features of multi-source processed data. For example, in data center infrastructure monitoring and prediction, indicators such as temperature, humidity, and power often have strong temporal correlations. The LSTM branch can filter out redundant information and generate corresponding first feature representations while ensuring long-term dependencies.

[0091] Meanwhile, the Transformer branch is used for global context modeling of multi-source processed data. Through the self-attention mechanism, the Transformer can establish dependencies between different time points over a longer time span, effectively avoiding the information decay problem that may occur in long-term modeling of LSTM. For example, when predicting the operating status of data center equipment, the Transformer branch can not only capture the operating trends of the past few weeks or even months, but also focus on short-term events such as power outages and sudden abnormal fluctuations, and then extract the second feature representation.

[0092] For example, the LSTM branch can be configured with a hidden layer dimension of 128 and an input sequence length of 60 time steps to model the running features over a recent period; while the Transformer branch can employ a multi-head attention mechanism, whose parameter configuration can include 8 attention heads, 4 encoder layers, a hidden layer dimension of 256, and a feedforward network dimension of 512 to ensure sufficient expressive power in global feature modeling.

[0093] Subsequently, the output features of the LSTM and Transformer branches can be integrated through a fusion layer. The fusion layer can employ attention weighting or residual concatenation to generate the final first prediction result while maintaining temporal dependencies and global feature complementarity. To avoid overfitting, the first prediction model can also introduce a random deactivation mechanism (e.g., set to 0.2) and train the overall model using an adaptive moment estimation optimization algorithm (e.g., with a learning rate set to 0.001). The loss function can be the mean squared error to ensure the stability and accuracy of the prediction output. Through this multi-branch fusion structure, the first prediction model can simultaneously consider the refinement of temporal modeling and the sensitivity to global context, thereby improving the accuracy and robustness of device state prediction.

[0094] Figure 3 The flowchart illustrating a method for hyperparameter optimization according to some exemplary embodiments of this application is shown schematically.

[0095] like Figure 3 As shown, the method for hyperparameter optimization includes operations S310 to S350.

[0096] In operation S310, the hyperparameter search space is defined.

[0097] For example, as shown in the table below, the hyperparameter search space can encompass learning rate, dropout rate, training batch size, hidden layer dimension, number of encoder layers, and number of attention heads. Furthermore, it can also include multiple dimensions such as time step, feedforward network size, and regularization coefficient to ensure the sufficiency and diversity of the optimization scope.

[0098] Table 1 Examples of Hyperparameter Search Spaces

[0099]

[0100] In operation S320, based on the hyperparameter search space, multiple candidate hyperparameter configurations are generated using random sampling. The candidate hyperparameter configurations are then input into the first prediction model for training and validation to obtain the corresponding performance indicators and form an observation dataset.

[0101] In operation S330, a mapping relationship between the candidate hyperparameter configuration and the performance index is established using a random forest model. The expected improvement value is calculated based on the mapping relationship, and the updated candidate hyperparameter configuration is determined based on the expected improvement value.

[0102] Random forest models demonstrate excellent fitting capabilities for nonlinear relationships and high-dimensional features, thus more accurately characterizing the dependency between hyperparameter configuration and model performance. Based on this, the system calculates the expected performance improvement value according to the mapping relationship, which measures the potential of different candidate configurations to improve performance, and determines new updated candidate hyperparameter configurations based on the expected performance improvement value.

[0103] In operation S340, the updated candidate hyperparameter configuration is input into the first prediction model for training and validation to update the observation dataset. Through this process, the system can continuously expand and optimize the observation dataset, enabling it to cover more hyperparameter regions that may be close to the optimal solution.

[0104] In operation S350, the process of establishing mapping relationships, determining and updating candidate hyperparameter configurations, and updating the observation dataset is repeated until the preset stopping condition is met. Then, the hyperparameter configuration whose performance meets the set conditions is selected from the observation dataset as the target parameter configuration.

[0105] The stopping conditions may include reaching the upper limit of the number of iterations, the performance index no longer showing significant improvement in several consecutive rounds of optimization, or reaching a set performance threshold. When the stopping conditions are met, the system selects hyperparameter configurations with good performance or that meet the set conditions from the observation dataset and applies them as target parameter configurations to the first prediction model.

[0106] By introducing surrogate modeling based on random forest and an evaluation mechanism for expected improvement during hyperparameter optimization, this method can quickly approach the globally optimal configuration with limited training and validation overhead, significantly reducing the computational cost required for hyperparameter tuning. Furthermore, due to the adoption of iterative updates and adaptive sampling mechanisms, this method can dynamically focus on the hyperparameter regions most likely to improve performance, thereby increasing optimization efficiency and convergence speed, ultimately resulting in a model configuration that outperforms in accuracy, stability, and generalization performance.

[0107] Figure 4 The flowchart illustrates a method for inference analysis using a second prediction model according to some exemplary embodiments of this application.

[0108] like Figure 4 As shown, the method for reasoning analysis using the second prediction model includes operations S410 to S420.

[0109] In operation S410, the multi-source processing data and the equipment operation and maintenance context information are concatenated to construct a structured text representation.

[0110] Specifically, by embedding and vectorizing numerical indicators, semantically encoding textual context, and then concatenating them according to a unified template, a structured text representation (Prompt) that can simultaneously reflect both numerical dynamic features and semantic background information is formed.

[0111] For example: "You are an equipment status prediction expert. Please analyze the following time series data: [15:00, Temperature = 26.3°C, Power Consumption = 4.3kW, Load = 33%]; [15:01, 27.9°C, 4.7kW, 37%]; … [60 points in total]. The target equipment is cabin-01, which has been in operation for 5 years. There is an unclosed work order: 'Abnormal noise from the air conditioner compressor.' Please predict the temperature change trend and potential risks for the next hour and determine whether an early warning is needed."

[0112] In operation S420, the second prediction model is guided to perform nonlinear relational reasoning and causal analysis based on the structured text representation, and the second prediction result is output.

[0113] Specifically, the second prediction model can be a large language model with language understanding capabilities, capable of semantic fusion and correlation modeling of data from different sources based on natural language prompts. On the one hand, the model can identify long-term trends and cyclical patterns in time-series data, determining whether the current state is in an abnormal fluctuation phase; on the other hand, the model can perform causal attribution based on semantic information in the context, thereby inferring the root causes of risk.

[0114] By leveraging structured text representation, numerical time-series parameters are concatenated with semantic information such as maintenance work orders and alarm records in a unified format, avoiding the information fragmentation problem caused by isolated processing of different modalities. Simultaneously, the large language model can perform nonlinear relationship modeling and causal inference on structured text, accurately identifying long-term trends and cyclical patterns.

[0115] In some embodiments, the second prediction result may include two parts: a quantitative prediction value and qualitative analysis information. Specifically, the quantitative prediction value is used to visually represent the trend of changes in the operating parameters of the target device over a preset time period in the future, such as the numerical prediction results of temperature, current, power, or vibration amplitude over the next hour, day, or week.

[0116] Building upon this, the second prediction result can also include qualitative analysis information to reveal the risk factors causing changes in the operating parameters and their corresponding reasoning pathways. For example, when quantitative prediction indicates that the temperature will continue to rise in the future, qualitative analysis information can further point out potential causes such as "decreased cooling system efficiency," "excessive power consumption," or "abnormal ambient temperature and humidity," and provide a causal chain description, such as "the continuous rise in the power consumption curve leads to increased heat dissipation load, resulting in insufficient cooling efficiency, and consequently, increased temperature." In this way, the system can not only output trend results but also provide more interpretable and traceable qualitative descriptions, enabling maintenance personnel to take more targeted scheduling or maintenance measures.

[0117] Compared with traditional methods that only output a single numerical prediction, the embodiments of this application provide both quantitative prediction and qualitative reasoning information in the second prediction result. This not only improves the accuracy and understandability of the prediction results, but also significantly enhances the decision support capabilities of operation and maintenance personnel, thereby achieving an organic combination of forward-looking early warning and interpretable reasoning in equipment management.

[0118] In embodiments of this application, acquiring multi-source operational data of the target device may include employing differentiated acquisition strategies for different types of indicators to balance real-time performance, completeness, and resource utilization efficiency. Specifically, a first indicator may be acquired based on a first preset frequency. The first indicator may be a numerical operational parameter, such as temperature, humidity, power, current, and the load of the central processing unit (CPU). Since these parameters change frequently over time, the acquisition system may be set to a high sampling frequency, such as acquiring data at the second or minute level, to ensure continuous monitoring and dynamic tracking of the device's operating status.

[0119] Based on this, the second indicator can also be collected at a second preset frequency. The second indicator includes equipment logs, network traffic data, and access control data, etc., which have a relatively low frequency of change and a large data volume. Therefore, a sampling strategy with a lower frequency than the first preset frequency can be adopted, such as hourly or daily periodic collection, to reduce storage and processing pressure, while retaining key operating environment and security management information.

[0120] Furthermore, the third indicator can be collected in real time to ensure immediate awareness of emergencies. This third indicator includes alarm signals and access control events, which are highly correlated with security incidents, operational anomalies, or unauthorized operations. Therefore, real-time acquisition needs to be achieved at the millisecond level or through event-driven triggering mechanisms. Once such an event is captured, the system can immediately combine it with existing time-series operational data and contextual information, thereby providing rapid response capabilities for subsequent prediction and management.

[0121] By combining multi-frequency and real-time data acquisition methods as described above, the embodiments of this application can effectively integrate low-frequency environmental and log information while ensuring continuous monitoring of high-frequency indicators, and promptly capture sudden events, thereby forming a more comprehensive, hierarchical, and highly reliable multi-source operational data input.

[0122] In some embodiments, managing the state of a target device based on prediction results may include the following process: First, the prediction results are compared with an absolute threshold and a dynamic threshold, respectively, to generate a first comparison result and a second comparison result. The absolute threshold can be a pre-set fixed critical value used to determine whether the device's operating parameters exceed safety limits; the dynamic threshold can be calculated in real time based on historical operating data and environmental conditions to reflect the reasonable fluctuation range of the device under different operating conditions. By comparing the two thresholds, both the safety baseline and the adaptability to the operating environment can be considered simultaneously, thereby obtaining a more accurate basis for risk assessment.

[0123] Based on this, the system can further combine the first comparison result, the second comparison result, and qualitative analysis information to generate a state determination signal. Specifically, when the predicted conclusion exceeds the absolute threshold, the state determination signal can directly characterize a high-risk state; when the predicted conclusion does not exceed the absolute threshold but deviates from the dynamic threshold, the state determination signal can combine qualitative analysis information to output potential anomalies or early risk warnings. For example, when the qualitative analysis information indicates "cooling system efficiency declines" and the predicted trend is close to the upper limit of the dynamic threshold, the system can generate a warning-level state determination signal. By integrating quantitative comparison and qualitative reasoning, the state determination signal has stronger interpretability and reliability.

[0124] Furthermore, the system can perform hierarchical management operations based on the status determination signals. Specifically, when the status determination signal indicates normal operation, the equipment maintains normal operation; when the status determination signal indicates a potential anomaly, the system can trigger lightweight intervention measures, such as increasing the sampling frequency, strengthening monitoring, or notifying maintenance personnel; when the status determination signal indicates a serious risk, measures such as shutdown protection, switching to backup equipment, or forced scheduling can be directly taken. Through this hierarchical management, differentiated strategies can be adopted under different risk levels, thereby achieving refined and intelligent equipment status management and avoiding losses caused by overreaction or underreaction.

[0125] In some embodiments, the first and second prediction results are integrated and analyzed based on a fusion model, which can derive different conclusions and output strategies based on preset rules. Specifically, when both the first and second prediction results indicate that the target device has the same risk, the fusion model can directly output a consistent conclusion and simultaneously generate corresponding alarm information. For example, when both prediction results indicate that the temperature is continuously rising and may exceed the safety threshold, the system will output a consistent risk conclusion and push a high-priority alarm to the maintenance personnel to prompt them to take timely maintenance or load reduction measures.

[0126] When there is a discrepancy between the first and second prediction results, the fusion model can output a divergent conclusion and generate a prompt message. For example, if the first prediction model infers that the equipment is operating normally based on time-series data, while the second prediction model infers a potential hidden danger in the cooling system based on context, the fusion model will output a divergent conclusion and generate a prompt message to remind maintenance personnel to pay attention to the divergence. If necessary, it can trigger manual review or further data collection to avoid the risks caused by misjudgment from a single model.

[0127] When both the first and second prediction results indicate that the target device is under low load, the fusion model can output resource optimization conclusions and generate resource allocation adjustment suggestions. For example, in a data center scenario, when the prediction results indicate that the server group is continuously under low load, the system can output resource optimization conclusions and suggest shutting down some redundant servers, reducing cooling power, or adjusting task scheduling strategies to achieve energy saving, consumption reduction, and efficient resource utilization.

[0128] In this way, the fusion model can not only enhance the credibility of conclusions in consistent scenarios, but also provide suggestive information to support human intervention in divergent scenarios, and play a role in resource optimization in low-load scenarios, thus demonstrating intelligence and practicality in both risk management and resource scheduling.

[0129] In other embodiments, the integrated analysis of the first and second prediction results based on the fusion model may further include the following process: First, obtaining the confidence information corresponding to the first and second prediction results. The confidence information can be the probability distribution value, confidence interval, or uncertainty estimation result calculated by each prediction model when outputting the result, used to characterize the credibility of the prediction conclusion. By introducing confidence information, the system can rely not only on the prediction result itself in subsequent analysis, but also comprehensively consider the reliability of the model output.

[0130] Based on this, a prediction uncertainty index can be further calculated using the aforementioned confidence information. Specifically, the difference in confidence levels between the first and second prediction results can be transformed into a quantitative uncertainty index using methods such as variance, entropy, or Bayesian uncertainty estimation, thereby reflecting the stability and consistency of the prediction results. For example, when the values ​​of the two prediction results are close but the confidence levels differ significantly, the prediction uncertainty index will be high, indicating that the conclusion may be biased.

[0131] Subsequently, the predicted uncertainty index can be weighted and fused with the equipment operation and maintenance context information. Through this weighting mechanism, the system can rely more on contextual information for auxiliary correction in high-uncertainty scenarios, while prioritizing the predicted results in low-uncertainty scenarios. The weighted fusion can be achieved by statically setting weights or by using a dynamic adaptive algorithm to automatically adjust the weight ratios according to the scenario, thus enabling more flexible integration of predicted results.

[0132] Finally, based on the weighted fusion result, predefined prediction conclusions can be matched. These predefined prediction conclusions can include categories such as normal state, potentially abnormal state, severely risky state, or resource-optimized state. Through the combined effect of uncertainty indicators and contextual information, the system can output more robust and interpretable prediction conclusions, providing a clear basis for subsequent equipment status management.

[0133] Corresponding to the above-described device status management method, embodiments of this application also provide a device status management apparatus.

[0134] Figure 5 A schematic block diagram of a device status management apparatus according to an embodiment of this application is shown.

[0135] like Figure 5 As shown, the device status management device 500 in this embodiment includes a data acquisition module 510, a status prediction module 520, an integration and analysis module 530, and a device status management module 540.

[0136] The data acquisition module 510 can be used to acquire multi-source operating data of the target device, process the multi-source operating data, and obtain multi-source processed data. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0137] The state prediction module 520 can be used to input the multi-source processed data in parallel into a first prediction model and a second prediction model to perform state prediction, obtaining a first prediction result and a second prediction result. The first prediction model is used to model the time dependency and global features of the multi-source processed data, while the second prediction model is used to perform inference analysis in conjunction with the multi-source processed data and equipment operation and maintenance context information. In one embodiment, the state prediction module 520 can be used to execute the operation S220 described above, which will not be repeated here.

[0138] The integration analysis module 530 can be used to integrate and analyze the first prediction result and the second prediction result based on the fusion model to generate a prediction conclusion. In one embodiment, the integration analysis module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0139] The device status management module 540 can be used to manage the device status of the target device based on the prediction conclusion. In one embodiment, the device status management module 540 can be used to perform the operation S240 described above, which will not be repeated here.

[0140] According to an embodiment of this application, the state prediction module 520 can also be used to perform time-series dependency modeling on the time-series indicators in the multi-source processed data and output a first feature representation; to perform global context modeling on the multi-source processed data using an attention mechanism and output a second feature representation; and to input the first feature representation and the second feature representation into a fusion layer and output the first prediction result.

[0141] According to an embodiment of this application, the state prediction module 520 can also be used to concatenate the multi-source processed data and the equipment operation and maintenance context information to construct a structured text representation; and guide the second prediction model to perform nonlinear relationship reasoning and causal analysis based on the structured text representation, and output the second prediction result. The second prediction result includes: a quantitative prediction value, used to represent the trend of the operating parameters of the target equipment in a future preset time period; and qualitative analysis information, used to represent the risk factors that cause the changes in the operating parameters and the corresponding reasoning links.

[0142] According to an embodiment of this application, the data acquisition module 510 can also be used to collect a first indicator of the multi-source operating data based on a first preset frequency, the first indicator including at least one of temperature, humidity, power, current and central processing unit; collect a second indicator of the multi-source operating data based on a second preset frequency, the second indicator including at least one of device logs, network traffic and access control data, the first preset frequency being higher than the second preset frequency; and collect a third indicator of the multi-source operating data in real time, the third indicator including at least one of alarm signals and access control switch events.

[0143] According to an embodiment of this application, the device status management device 500 may further include a training module. The training module can be used to: define a hyperparameter search space; based on the hyperparameter search space, generate multiple candidate hyperparameter configurations using random sampling, input the candidate hyperparameter configurations into the first prediction model for training and validation, obtain corresponding performance indicators, and form an observation dataset; establish a mapping relationship between the candidate hyperparameter configurations and the performance indicators using a random forest model, calculate the expected improvement value based on the mapping relationship, and determine the updated candidate hyperparameter configuration based on the expected improvement value; input the updated candidate hyperparameter configuration into the first prediction model for training and validation to update the observation dataset; and repeat the process of establishing the mapping relationship, determining the updated candidate hyperparameter configuration, and updating the observation dataset until a preset stopping condition is met, and select a hyperparameter configuration whose performance meets the set conditions from the observation dataset as the target parameter configuration.

[0144] According to an embodiment of this application, the integration analysis module 530 can also be used to output a consistency conclusion and generate corresponding alarm information when the first prediction result and the second prediction result indicate that the target device has the same risk; output a divergence conclusion and generate a prompt information when the first prediction result and the second prediction result differ; and output a resource optimization conclusion and generate a resource allocation adjustment suggestion when both the first prediction result and the second prediction result indicate that the target device is in a low load state.

[0145] According to an embodiment of this application, the integration analysis module 530 can also be used to obtain confidence information corresponding to the first prediction result and the second prediction result, calculate a prediction uncertainty index based on the confidence information, perform weighted fusion of the prediction uncertainty index and the equipment operation and maintenance context information, and match a predefined prediction conclusion based on the weighted fusion result. The equipment operation and maintenance context information includes at least one of historical time-series data, equipment metadata, work order status, and historical alarm records.

[0146] According to an embodiment of this application, the device status management module 540 can also be used to compare the prediction conclusion with an absolute threshold and a dynamic threshold respectively to generate a first comparison result and a second comparison result; generate a status determination signal based on the first comparison result, the second comparison result and the qualitative analysis information; and perform hierarchical management operations according to the status determination signal.

[0147] According to embodiments of this application, any multiple modules among the data acquisition module 510, state prediction module 520, integration analysis module 530, and device state management module 540 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 510, state prediction module 520, integration analysis module 530, and device state management module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the data acquisition module 510, the status prediction module 520, the integration analysis module 530, and the equipment status management module 540 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0148] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a device state management method according to an embodiment of this application.

[0149] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0150] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0151] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0152] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0153] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0154] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the device state management method provided in the embodiments of this application.

[0155] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0156] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0157] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0158] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0159] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0160] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0161] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A method for managing equipment status, characterized in that, The method includes: Acquire multi-source operational data of the target device, process the multi-source operational data, and obtain multi-source processed data; The multi-source processed data is input in parallel into the first prediction model and the second prediction model to perform state prediction, and the first prediction result and the second prediction result are obtained. The first prediction model is used to model the time dependency and global features of the multi-source processed data, and the second prediction model is used to perform inference analysis in combination with the multi-source processed data and equipment operation and maintenance context information. The first and second prediction results are integrated and analyzed based on a fusion model to generate prediction conclusions; and Based on the prediction results, the target device is managed in terms of device status.

2. The method according to claim 1, characterized in that, The step of modeling the time dependencies and global features of the multi-source processed data using the first prediction model includes: Perform time-series dependency modeling on the time-series indicators in the multi-source processed data, and output the first feature representation; The multi-source processed data is modeled globally using an attention mechanism to output a second feature representation; and The first feature representation and the second feature representation are input into the fusion layer, and the first prediction result is output.

3. The method according to claim 1, characterized in that, The step of using the second prediction model in conjunction with the multi-source processed data and equipment operation and maintenance context information for inference analysis includes: The multi-source processed data and the equipment operation and maintenance context information are concatenated to construct a structured text representation; and Based on the structured text representation, the second prediction model is guided to perform nonlinear relational reasoning and causal analysis, and outputs the second prediction result.

4. The method according to claim 1 or 2, characterized in that, The training process of the first prediction model includes a hyperparameter optimization process using a random forest model, specifically including: Define the hyperparameter search space; Based on the hyperparameter search space, multiple candidate hyperparameter configurations are generated by random sampling. The candidate hyperparameter configurations are then input into the first prediction model for training and validation to obtain the corresponding performance indicators and form an observation dataset. A mapping relationship between the candidate hyperparameter configurations and the performance metrics is established using a random forest model. The expected improvement value is calculated based on the mapping relationship, and the updated candidate hyperparameter configurations are determined based on the expected improvement value. The updated candidate hyperparameter configurations are input into the first prediction model for training and validation to update the observation dataset; and The process of establishing mapping relationships, determining and updating candidate hyperparameter configurations, and updating the observation dataset is repeated until a preset stopping condition is met. Then, a hyperparameter configuration whose performance meets the set conditions is selected from the observation dataset as the target parameter configuration.

5. The method according to claim 1 or 3, characterized in that, The second prediction result includes: Quantitative predicted values ​​are used to represent the trend of changes in the operating parameters of the target device over a preset time period in the future; and Qualitative analysis information is used to represent the risk factors that cause changes in the operating parameters and the corresponding reasoning links.

6. The method according to claim 1, characterized in that, The acquisition of multi-source operational data of the target device includes: A first indicator is collected based on a first preset frequency of the multi-source operating data, the first indicator including at least one of temperature, humidity, power, current and central processing unit; A second indicator is collected based on a second preset frequency of the multi-source operational data, the second indicator including at least one of device logs, network traffic, and access control data, wherein the first preset frequency is higher than the second preset frequency; and A third indicator for real-time collection of the multi-source operational data, the third indicator including at least one of alarm signals and access control switch events.

7. The method according to claim 5, characterized in that, Based on the prediction results, the target device is managed in terms of device status, including: The prediction conclusion is compared with the absolute threshold and the dynamic threshold respectively to generate a first comparison result and a second comparison result. Based on the first comparison result, the second comparison result, and the qualitative analysis information, a state determination signal is generated; and Hierarchical management operations are performed based on the status determination signal.

8. The method according to claim 1, characterized in that, The step of integrating and analyzing the first prediction result and the second prediction result based on the fusion model to generate prediction conclusions includes: When the first prediction result and the second prediction result indicate that the target device has the same risk, a consistent conclusion is output and a corresponding alarm message is generated. When there is a difference between the first prediction result and the second prediction result, a divergence conclusion is output and a prompt message is generated. When both the first prediction result and the second prediction result indicate that the target device is in a low-load state, the resource optimization conclusion is output and the resource allocation adjustment suggestion is generated.

9. The method according to claim 1, characterized in that, The step of integrating and analyzing the first prediction result and the second prediction result based on the fusion model to generate prediction conclusions includes: Obtain the confidence information corresponding to the first prediction result and the second prediction result, and calculate the prediction uncertainty index based on the confidence information; The prediction uncertainty index and the equipment operation and maintenance context information are weighted and fused; and The results of weighted fusion are matched with predefined prediction conclusions.

10. The method according to claim 1 or 9, characterized in that, The device operation and maintenance context information includes at least one of the following: historical time-series data, device metadata, work order status, and historical alarm records.

11. A device for managing equipment status, characterized in that, The device includes: The data acquisition module is used to: acquire multi-source operating data of the target device, process the multi-source operating data, and obtain multi-source processed data; The state prediction module is used to: input the multi-source processed data into a first prediction model and a second prediction model in parallel to perform state prediction, and obtain a first prediction result and a second prediction result. The first prediction model is used to model the time dependency and global features of the multi-source processed data, and the second prediction model is used to perform inference analysis in combination with the multi-source processed data and equipment operation and maintenance context information. The integration analysis module is used to: integrate and analyze the first prediction result and the second prediction result based on the fusion model to generate prediction conclusions; and The equipment status management module is used to: manage the equipment status of the target equipment based on the prediction conclusion.

12. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 10.

14. A computer program product comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 10.

Images