Cloud machine room intelligent predictive monitoring system and method based on digital twinborn and multi-mode perception

Abstract

The invention relates to the technical field of cloud machine rooms, and discloses a cloud machine room intelligent predictive monitoring method based on digital twinborn and multi-modal perception, which comprises the following steps: S1, multi-modal data acquisition and fusion, S2, construction and synchronization of a digital twinborn model, S3, analysis, judgment and prediction, S4, early warning and decision support, and S5, feedback and model self-optimization. The invention further discloses a cloud machine room intelligent predictive monitoring system based on digital twinning and multi-mode sensing. The system comprises a multi-mode sensing module, a data fusion and processing module, a digital twinning module, an analysis and prediction module and a decision support module which are connected in sequence. Through multi-modal data fusion and a digital twinborn model, rich contexts and association relationships are provided for data analysis, during problem diagnosis, single indexes are not analyzed in an isolated manner, and root cause analysis is quickly performed in combination with equipment topology and a dynamic environment, so that crossing from single-point monitoring to global cognition is realized, and fault diagnosis is more accurate and intelligent.

Description

Technical Field

[0001] This invention relates to the field of cloud data center technology, and more particularly to the field of predictive monitoring of cloud data centers, specifically to an intelligent predictive monitoring system and method for cloud data centers based on digital twins and multimodal perception. Background Technology

[0002] A cloud data center, also known as a cloud computing data center, is a physical facility specifically designed, built, and operated to support cloud computing services. It encompasses a complex set of physical and logical infrastructure, including building, power, cooling, network, computing, storage, and security. Its core purpose is to host and deliver cloud services in an efficient, reliable, and scalable manner.

[0003] Predictive monitoring for cloud data centers is an advanced operational paradigm based on data and artificial intelligence. It refers to the continuous collection and analysis of multi-dimensional operational data of cloud data center infrastructure (including IT equipment, power, environment, etc.), and the use of machine learning and other algorithm models to proactively identify abnormal patterns, diagnose potential root causes of failures, and predict future performance degradation or hardware failures.

[0004] Traditional cloud data center monitoring systems have the following problems:

[0005] 1. Data from power, environment, and IT equipment are independent of each other, forming data silos. When a fault occurs, maintenance personnel need to manually correlate and analyze multiple isolated systems, which is inefficient and prone to misjudgment.

[0006] 2. Alarms based on fixed thresholds cannot adapt to dynamic changes in business, leading to false alarms or missed alarms. Furthermore, alarms can only be issued after a fault occurs or an abnormal indicator appears, making it impossible to predict future risks.

[0007] Therefore, we need to propose an intelligent predictive monitoring system and method for cloud data centers based on digital twins and multimodal perception. Summary of the Invention

[0008] The purpose of this invention is to provide an intelligent predictive monitoring system and method for cloud data centers based on digital twins and multimodal perception. Through multimodal data fusion and digital twin models, it provides rich context and correlations for data analysis. When diagnosing problems, it does not analyze individual indicators in isolation, but combines equipment topology and power environment to quickly perform root cause analysis, realizing a leap from single-point monitoring to full-domain cognition, making fault diagnosis more accurate and intelligent, thereby solving the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a cloud data center intelligent predictive monitoring method based on digital twin and multimodal perception, comprising the following steps:

[0010] S1. Multimodal data acquisition and fusion: Deploy various sensors, cameras, and power and environmental monitoring equipment in the cloud data center to collect data in real time, and process and fuse the collected data to form a dataset with contextual information.

[0011] S2. Construct and synchronize a digital twin model. Based on the 3D model of the cloud data center, equipment asset information, and logical relationships, construct a digital twin and drive it with the collected real-time data to achieve real-time synchronization mapping with the physical data center.

[0012] S3, Analysis, Judgment and Prediction: Utilizes fused data and digital twin models to monitor multiple indicators in real time, perform trend analysis and correlation analysis; and quickly conduct root cause analysis based on anomalies, and predict the future health status of devices based on historical data and machine learning models.

[0013] S4. Early warning and decision support: Based on the analysis and prediction results, generate early warning information of different levels and notify the operation and maintenance personnel.

[0014] S5. Feedback and Model Self-Optimization: The actions taken by operations and maintenance personnel based on the early warning information, and the effects of those actions, serve as new feedback data to evaluate and optimize the accuracy of predictions.

[0015] Preferably, in step S1, the various sensors include temperature sensors, humidity sensors, vibration sensors, and noise sensors. The cloud data center also deploys management interfaces for data access, including IPMI, SNMP, MQTT, CoAP, Modbus, BACnet, Redfish, Kubernetes API, and cloud platform API.

[0016] Data processing includes:

[0017] Data cleaning, handling missing data and outliers;

[0018] Data denoising involves using filtering algorithms to smooth the data while preserving the true trend.

[0019] Data standardization, data with the same units and dimensions;

[0020] Data fusion includes any one or more of the following: data-level fusion, feature-level fusion, and decision-level fusion.

[0021] Preferably, in step S2, the 3D model includes the 3D shape and layout of the device. The device asset information is understood by embedding the physical rules of the device. The physical rules of the device include the power-heat conversion model of the server, the cooling capacity model of the air conditioner, and the airflow dynamics model. The digital twin defines the behavioral logic of the device in a specific state and accurately describes the dependencies between devices.

[0022] Preferably, in step S3, when analyzing the device status, multiple indicators for each device automatically learn normal behavior patterns to form a dynamic baseline that changes over time and date, and do not view individual indicators of the device in isolation, but automatically analyze the correlation between different indicators of the device.

[0023] When determining the cause of equipment malfunctions, the methods used include root cause localization based on logical and physical topology and causal inference engines;

[0024] When predicting the future state of equipment, the remaining service life, performance, and capacity of the equipment are predicted based on the equipment's historical data.

[0025] Preferably, in step S4, the methods for notifying maintenance personnel of the warning information include large screens, SMS, email, voice, and APP push; based on the warning information level and combined with the business impact of the root cause of the fault, the warning information is automatically sent to the relevant department heads. At the same time, IT service management tools are integrated in the cloud data center to automatically create and assign maintenance work orders, and track the overall processing lifecycle of maintenance work orders.

[0026] Preferably, in step S5, the feedback data includes the direct operation and marking of the early warning information by the operation and maintenance personnel, the accuracy of the prediction of the future health status of the equipment, and the digital twin model continuously calculates the prediction accuracy, recall rate, and false alarm rate.

[0027] The present invention also provides an intelligent predictive monitoring system for cloud data centers based on digital twins and multimodal perception, used to implement the intelligent predictive monitoring method for cloud data centers based on digital twins and multimodal perception described above. The system includes a multimodal perception module, a data fusion and processing module, a digital twin module, an analysis and prediction module, and a decision support module connected in sequence. The decision support module is electrically connected to the digital twin module.

[0028] Preferably, the multimodal sensing module is responsible for the access, configuration and management of sensors, cameras and power environment monitoring equipment, as well as the collection and transmission of raw data;

[0029] The data fusion and processing module receives raw data from the multimodal perception module and performs cleaning, noise reduction, alignment, and format standardization on the raw data. It then performs data fusion based on time rules, spatial rules, and device association rules.

[0030] Preferably, the digital twin module is responsible for maintaining the 3D geometric model, equipment asset information model, physical rule model and business logic model of the cloud data center, receiving data from the data fusion and processing module, and updating the status of the virtual entities in the digital twin in real time;

[0031] The analysis and prediction module is responsible for performing real-time streaming computation on the data transmitted from the digital twin, detecting anomalies, periodically mining historical data, training and optimizing prediction models, and performing diagnostic inference.

[0032] Preferably, the decision support module is used to intuitively display the global status, early warning information, and analysis results of the cloud data center, manage alarm rules and send alarm notifications, and automatically generate operation and maintenance work orders based on the prediction results.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. This invention provides rich context and correlation for data analysis through multimodal data fusion and digital twin model. When diagnosing problems, it does not analyze a single indicator in isolation, but combines equipment topology and power environment to quickly perform root cause analysis, realizing the leap from single-point monitoring to full-domain cognition, making fault diagnosis more accurate and intelligent.

[0035] 2. This invention analyzes historical data through machine learning models to predict the remaining service life and performance capacity bottlenecks of equipment, enabling maintenance personnel to shift from intervention after a failure occurs to early warning before a failure occurs. It generates actionable decision suggestions with specific time windows in advance, making maintenance more planned and controllable, and minimizing unplanned downtime.

[0036] 3. This invention features data feedback and model self-optimization, and uses the intervention effect of operation and maintenance personnel as feedback data to continuously evaluate and optimize the prediction model, enabling the system to adapt to equipment aging and environmental changes, and to continuously learn from operation and maintenance experience, making it more accurate in future fault analysis and prediction. Attached Figure Description

[0037] Figure 1 This is a flowchart of the present invention;

[0038] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Please see Figure 1 This invention provides a technical solution: a cloud data center intelligent predictive monitoring method based on digital twins and multimodal perception, comprising the following steps:

[0041] S1. Multimodal data acquisition and fusion: Deploy various sensors, cameras, and power and environmental monitoring equipment in the cloud data center to collect data in real time, and process and fuse the collected data to form a dataset with contextual information.

[0042] It overcomes the problem of data silos in traditional monitoring systems, providing a more three-dimensional and complete perception of the cloud data center status. Multi-source data cross-verification can effectively identify false alarms or data distortions from a single sensor, improve data credibility, and enrich the context. For example, it integrates the power consumption data of the cabinet with the CPU load data of the server inside the cabinet and the air outlet temperature data of the air conditioner in front of the cabinet, providing a richer context for analysis.

[0043] Multiple sensors are used, including temperature sensors, humidity sensors, vibration sensors, and noise sensors. The cloud data center also has management interfaces for data access, including IPMI, SNMP, MQTT, CoAP, Modbus, BACnet, Redfish, Kubernetes API, and cloud platform API.

[0044] Among them, the MQTT and CoAP interfaces are used for low-power transmission of massive sensor data, while the Modbus and BACnet interfaces are used to interface with power and environmental equipment such as precision air conditioners and UPS.

[0045] Data processing includes:

[0046] Data cleaning, handling missing data and outliers (such as instantaneous spikes and glitches in sensors).

[0047] Data denoising involves using filtering algorithms (such as Kalman filtering) to smooth the data and preserve the true trend.

[0048] Data standardization, unifying data in different units and dimensions (e.g., temperature is standardized to degrees Celsius, power consumption is standardized to kilowatts).

[0049] Data fusion includes any one or more of the following: data-level fusion, feature-level fusion, and decision-level fusion.

[0050] Data-level fusion directly integrates the raw data. For example, it averages the readings of multiple adjacent temperature sensors to obtain a more stable regional temperature value.

[0051] Feature-level fusion first extracts representative features from various data sources, and then fuses these features. For example, it extracts "intrusion" events from camera videos and "spectral features" from vibration data, and then feeds these two features, along with temperature values, into the analysis model.

[0052] In decision-level fusion, each data source first makes a preliminary decision or judgment using its own data, and then the system integrates these decisions. For example, a vibration sensor judges "the fan may be malfunctioning" based on pattern recognition, a noise sensor judges "there is an abnormal high-frequency sound", and a temperature sensor judges "the temperature is slowly rising". Finally, by integrating these "judgments", a high-confidence conclusion of "fan bearing wear" is reached.

[0053] S2. Construct and synchronize a digital twin model. Based on the 3D model of the cloud data center, equipment asset information, and logical relationships, construct a digital twin and drive it with the collected real-time data to achieve real-time synchronization mapping with the physical data center.

[0054] The complex physical data center and its operating status are presented in an intuitive 3D visualization, enabling a "one-map overview" and greatly improving operation and maintenance efficiency.

[0055] Digital twins are not merely passive mappings, but powerful simulation sandboxes. They allow for "what-if" analyses in virtual space, such as simulating temperature field changes after an air conditioner is shut down, or the impact of adding a new server on airflow and power consumption, without the need to risk exposure in a real-world environment.

[0056] Furthermore, discrete multimodal data are linked and presented in a unified three-dimensional space, so that the data is no longer isolated numbers, but "information entities" with spatial location and correlation.

[0057] The 3D model includes the 3D shape and layout of the equipment. By embedding the physical rules of the equipment, we can understand the equipment asset information. The physical rules of the equipment include the power-heat conversion model of the server, the cooling capacity model of the air conditioner, and the airflow dynamics model. The digital twin defines the behavioral logic of the equipment in a specific state, such as the alarm chain triggered after the server crashes and the process of the UPS switching to the backup power supply. It also accurately describes the dependencies between the equipment, such as which server groups will be affected by the failure of a network switch or which cabinets a precision air conditioner is responsible for cooling.

[0058] The real-time, fused data provided in step S1 is the "blood" of the twin. Through data-driven technology, the twin is no longer a static "display model," but a dynamic system that can reflect or even "preview" the state of the physical entity in real time.

[0059] S3, Analysis, Judgment and Prediction: Utilizes fused data and digital twin models to monitor multiple indicators in real time, perform trend analysis and correlation analysis; and quickly conduct root cause analysis based on anomalies, and predict the future health status of devices based on historical data and machine learning models.

[0060] By shifting from post-processing to pre-event warning, predictive maintenance is achieved, unplanned downtime is avoided, and complex root cause analysis is automatically completed. This frees up maintenance personnel from tedious log and indicator troubleshooting, allowing them to quickly locate the essence of the problem and discover subtle signs and complex patterns that are difficult for the human eye to detect, making prediction and diagnosis more accurate.

[0061] When analyzing device status, multiple indicators for each device automatically learn normal behavior patterns to form a dynamic baseline that changes over time and date. Furthermore, it does not view individual device indicators in isolation but automatically analyzes the correlation between different device indicators.

[0062] When determining the cause of equipment malfunctions, the methods used include root cause localization based on logical and physical topology and causal inference engines;

[0063] When predicting the future state of equipment, the remaining service life, performance, and capacity of the equipment are predicted based on the equipment's historical data.

[0064] Step S3 obtains the multimodal data stream from step S1, and obtains it from step S2:

[0065] Data Relationship Map: Digital twins clearly define the spatial and logical relationships between devices. When analyzing abnormal server temperatures, it knows which air conditioner or power distribution cabinet's related data to query, instead of blindly searching through the entire dataset.

[0066] Physical model constraints: For example, when diagnosing insufficient cooling, calculations are performed by combining the cooling capacity model of the air conditioner and the heat load model of the computer room to determine whether it is "decrease in equipment efficiency" or "excessive heat load", so that the diagnostic conclusion conforms to physical laws.

[0067] S4. Early Warning and Decision Support: Based on analysis and forecasting results, it generates early warning information of different levels and notifies operations and maintenance personnel; it provides specific and actionable decision-making suggestions, shortening the decision-making path for operations and maintenance personnel and improving emergency response capabilities. Clear early warnings and maintenance recommendations enable operations and maintenance work to be carried out in a planned manner, minimizing the risk of business interruption.

[0068] For example, the forecast results are as follows: The cooling efficiency of the A03 precision air conditioner is expected to drop below the threshold in 15 days, and filter cleaning is recommended during the next maintenance window (early Saturday morning). Or, the hard drive failure probability of server SVR-47 has reached 85%, and hot-migrating and hard drive replacement are recommended within 48 hours.

[0069] The methods for notifying operations and maintenance personnel of early warning information include large screens, SMS, email, voice, and APP push notifications. Based on the level of early warning information and the business impact of the root cause of the fault, the early warning information is automatically sent to the relevant department heads. At the same time, IT service management tools are integrated into the cloud data center to automatically create and assign operation and maintenance work orders, and track the overall processing lifecycle of operation and maintenance work orders.

[0070] S5. Feedback and Model Self-Optimization: The actions taken by operations personnel based on early warning information, and the effects of those actions, serve as new feedback data to evaluate and optimize the accuracy of predictions. This forms a closed-loop learning process, making the system increasingly intelligent and accurate over time. It continuously learns from practical operational experience, adapts to changes in the data center environment and equipment aging, and continuously improves prediction accuracy, avoiding the problem of AI models "failing" due to environmental changes and ensuring the long-term effectiveness of the system.

[0071] Feedback data includes direct operations and marking of early warning information by maintenance personnel (such as whether the early warning situation is true and whether it has been handled), the accuracy of predictions of the future health status of equipment (such as whether a hard drive predicted to fail in 7 days actually failed in about 7 days), and the digital twin model continuously calculates the accuracy, recall, and false alarm rate of predictions.

[0072] Accuracy, the percentage of correctly predicted faults out of the total number of predictions;

[0073] Recall rate is the percentage of actual failures that were successfully predicted.

[0074] False alarm rate is the percentage of cases where a warning is issued but the malfunction does not occur.

[0075] Please see Figure 2 The present invention also provides an intelligent predictive monitoring system for cloud data centers based on digital twins and multimodal perception, used to implement the intelligent predictive monitoring method for cloud data centers based on digital twins and multimodal perception described above. The system includes a multimodal perception module, a data fusion and processing module, a digital twin module, an analysis and prediction module, and a decision support module connected in sequence. The decision support module is electrically connected to the digital twin module.

[0076] The multimodal sensing module is responsible for the access, configuration, and management of sensors, cameras, and power environment monitoring equipment, as well as the acquisition and transmission of raw data. Its modular design facilitates the expansion to new sensor types and ensures the stability and real-time performance of data acquisition.

[0077] The data fusion and processing module receives raw data from the multimodal perception module and performs cleaning, noise reduction, alignment, and format standardization on the raw data. It then performs data fusion based on time rules, spatial rules, and device association rules to generate a dataset for analysis.

[0078] The digital twin module is responsible for maintaining the 3D geometric model, equipment asset information model, physical rule model, and business logic model of the cloud data center. It receives data from the data fusion and processing module, updates the status of the virtual entities in the digital twin in real time, and provides simulation and deduction capabilities. This module constructs a dynamic digital copy of the physical data center, which is a bridge connecting the physical world and the information world, and is also the foundation for realizing visualization and simulation.

[0079] The analysis and prediction module is responsible for performing real-time streaming computation on the data transmitted from the digital twin, detecting anomalies, periodically mining historical data, training and optimizing prediction models, and performing diagnostic inference.

[0080] The decision support module is used to intuitively display the global status, early warning information, and analysis results of the cloud data center (using 2D / 3D charts, 3D scenes, etc.), manage alarm rules and send alarm notifications, and automatically generate operation and maintenance work orders based on the prediction results. This module provides an intuitive and user-friendly interactive experience, transforming complex analysis results into actionable guidelines.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud data center intelligent predictive monitoring method based on digital twin and multimodal perception, characterized in that, Includes the following steps: S1. Multimodal data acquisition and fusion: Deploy various sensors, cameras, and power and environmental monitoring equipment in the cloud data center to collect data in real time, and process and fuse the collected data to form a dataset with contextual information. S2. Construct and synchronize a digital twin model. Based on the 3D model of the cloud data center, equipment asset information, and logical relationships, construct a digital twin and drive it with the collected real-time data to achieve real-time synchronization mapping with the physical data center. S3, Analysis, Judgment and Prediction: Utilizes fused data and digital twin models to monitor multiple indicators in real time, perform trend analysis and correlation analysis; and quickly conduct root cause analysis based on anomalies, and predict the future health status of equipment based on historical data and machine learning models. S4. Early warning and decision support: Based on the analysis and prediction results, generate early warning information of different levels and notify the operation and maintenance personnel. S5. Feedback and Model Self-Optimization: The actions taken by operations and maintenance personnel based on the early warning information, and the effects of those actions, serve as new feedback data to evaluate and optimize the accuracy of predictions.

2. The intelligent predictive monitoring method for cloud data centers based on digital twins and multimodal perception according to claim 1, characterized in that: In step S1, various sensors include temperature sensors, humidity sensors, vibration sensors, and noise sensors. The cloud data center also deploys management interfaces for data access, including IPMI, SNMP, MQTT, CoAP, Modbus, BACnet, Redfish, Kubernetes API, and cloud platform API. Data processing includes: Data cleaning, handling missing data and outliers; Data denoising involves using filtering algorithms to smooth the data while preserving the true trend. Data standardization, data with the same units and dimensions; Data fusion includes any one or more of the following: data-level fusion, feature-level fusion, and decision-level fusion.

3. The intelligent predictive monitoring method for cloud data centers based on digital twins and multimodal perception according to claim 1, characterized in that: In step S2, the 3D model includes the 3D shape and layout of the device. By embedding the physical rules of the device, the device asset information is understood. The physical rules of the device include the power-heat conversion model of the server, the cooling capacity model of the air conditioner, and the airflow dynamics model. The digital twin defines the behavioral logic of the device in a specific state and accurately describes the dependencies between devices.

4. The intelligent predictive monitoring method for cloud data centers based on digital twins and multimodal perception according to claim 1, characterized in that: In step S3, when analyzing the device status, multiple indicators for each device automatically learn normal behavior patterns to form a dynamic baseline that changes over time and date. Furthermore, the device does not view individual indicators in isolation but automatically analyzes the correlation between different indicators of the device. When determining the cause of equipment malfunctions, the methods used include root cause localization based on logical and physical topology and causal inference engines; When predicting the future state of equipment, the remaining service life, performance, and capacity of the equipment are predicted based on the equipment's historical data.

5. The intelligent predictive monitoring method for cloud data centers based on digital twins and multimodal perception according to claim 1, characterized in that: In step S4, the warning information is notified to the operation and maintenance personnel through methods such as large screen, SMS, email, voice, and APP push. Based on the warning information level and the business impact of the root cause of the fault, the warning information is automatically sent to the relevant department heads. At the same time, the IT service management tools integrated in the cloud data center automatically create and assign operation and maintenance work orders, and track the overall processing lifecycle of the operation and maintenance work orders.

6. The intelligent predictive monitoring method for cloud data centers based on digital twins and multimodal perception according to claim 1, characterized in that: In step S5, the feedback data includes the direct operation and marking of the early warning information by the operation and maintenance personnel, the accuracy of the prediction of the future health status of the equipment, and the digital twin model continuously calculates the prediction accuracy, recall rate, and false alarm rate.

7. A cloud data center intelligent predictive monitoring system based on digital twin and multimodal perception, used to implement the cloud data center intelligent predictive monitoring method based on digital twin and multimodal perception as described in any one of claims 1-6, characterized in that, It includes a multimodal sensing module, a data fusion and processing module, a digital twin module, an analysis and prediction module, and a decision support module connected in sequence, wherein the decision support module is electrically connected to the digital twin module.

8. The cloud data center intelligent predictive monitoring system based on digital twin and multimodal perception according to claim 7, characterized in that: The multimodal sensing module is responsible for the access, configuration and management of sensors, cameras and power environment monitoring equipment, as well as the acquisition and transmission of raw data; The data fusion and processing module receives raw data from the multimodal perception module and performs cleaning, noise reduction, alignment, and format standardization on the raw data. It then performs data fusion based on time rules, spatial rules, and device association rules.

9. The cloud data center intelligent predictive monitoring system based on digital twin and multimodal perception according to claim 8, characterized in that: The digital twin module is responsible for maintaining the 3D geometric model, equipment asset information model, physical rule model, and business logic model of the cloud data center, receiving data from the data fusion and processing module, and updating the status of the virtual entities in the digital twin in real time. The analysis and prediction module is responsible for performing real-time streaming computation on the data transmitted from the digital twin, detecting anomalies, periodically mining historical data, training and optimizing prediction models, and performing diagnostic inference.

10. The cloud data center intelligent predictive monitoring system based on digital twin and multimodal perception according to claim 9, characterized in that: The decision support module is used to intuitively display the global status, early warning information, and analysis results of the cloud data center, manage alarm rules and send alarm notifications, and automatically generate operation and maintenance work orders based on the prediction results.

Images