Data center digital twinborn simulation and decision-making system oriented to intelligent management

Abstract

The invention relates to the technical field of data center management, and discloses a data center digital twinborn simulation and decision-making system oriented to intelligent management. The system comprises a data center physical feature sensing module, a virtual space reconstruction module, an operation situation deduction engine, an abnormal behavior recognition module and a decision instruction generation module. Wherein the physical feature sensing module collects multi-dimensional operation parameters of the infrastructure in real time; the virtual space reconstruction module dynamically constructs a three-dimensional virtual model based on the collected parameters; running a situation deduction engine to simulate a resource scheduling and energy flow process; the abnormal behavior recognition module analyzes the analog data stream to detect an abnormal operation mode; and the decision instruction generation module integrates the abnormal information and generates an optimization regulation and control instruction for the physical equipment. According to the system, intelligent management of the data center is realized through real-time mapping, dynamic deduction and intelligent decision making of physical and virtual spaces, and the management precision and timeliness are improved.

Description

Technical Field

[0001] The present invention relates to the field of data center management technology, and specifically to a data center digital twin simulation and decision-making system for intelligent management. Background Art

[0002] With the rapid development of information technology, data centers, as core hubs for information storage and processing, continue to grow in size and complexity. Modern data centers often incorporate a wide range of infrastructure, including massive servers, network equipment, cooling systems, and power supply and distribution facilities. The multi-dimensional parameters generated by these devices during operation are interconnected and dynamically changing, posing significant challenges to traditional management models.

[0003] Traditional data center management relies heavily on manual inspections and decentralized monitoring systems, which present significant limitations. The operational status of physical equipment is lagging, making it difficult to capture transient changes such as temperature fluctuations, energy consumption peaks, and resource load imbalances in real time. Monitoring data from each subsystem is fragmented, lacking the ability to conduct collaborative analysis from a global perspective, making it difficult for managers to accurately predict potential risks. For example, when a server cluster experiences performance degradation due to insufficient cooling, traditional monitoring may only display the temperature anomaly of a specific device in isolation, failing to correlate the operating parameters of the air conditioning system with the cabinet layout, thus delaying troubleshooting.

[0004] Data center resource scheduling and energy optimization have long relied on empirical decision-making, lacking precise simulation verification. In scenarios with dynamically changing business loads, static resource allocation strategies often lead to wasted computing power or bottlenecks, while also maintaining high energy consumption, contradicting the need for green and low-carbon development. With the advancement of the digital economy, data centers are increasingly in need of intelligent management, requiring an integrated solution that enables real-time interaction between physical and virtual spaces, dynamic deduction, and intelligent decision-making. Summary of the Invention

[0005] The purpose of the present invention is to provide a data center digital twin simulation and decision-making system for intelligent management to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention provides a data center digital twin simulation and decision-making system for intelligent management, the system comprising: Data center physical characteristics perception module, used to collect multi-dimensional operating parameters of data center infrastructure in real time; A virtual space reconstruction module, configured to dynamically construct a three-dimensional virtual model of the data center based on the multi-dimensional operating parameters acquired by the data center physical feature perception module; An operating situation deduction engine is used to simulate the resource scheduling and energy flow process within the data center based on the three-dimensional virtual model generated by the virtual space reconstruction module; An abnormal behavior identification module is used to analyze the simulated data stream output by the operation situation deduction engine in real time and detect abnormal operation modes that deviate from a preset threshold; The decision instruction generation module is used to integrate the abnormal operation mode identified by the abnormal behavior identification module and generate optimization control instructions for physical equipment.

[0007] Preferably, the data center physical feature perception module includes: Environmental parameter acquisition unit, which obtains the temperature distribution, humidity distribution and air flow velocity distribution of each area of ​​the data center in real time; Equipment status monitoring unit, which continuously tracks the power consumption data of the server cluster, calculates the load rate and the operating status of the cooling equipment; The energy consumption feature extraction unit performs feature fusion on the original data acquired by the environmental parameter acquisition unit and the equipment status monitoring unit, and calculates the regional energy efficiency feature vector.

[0008] Preferably, the virtual space reconstruction module includes: The spatial topology analyzer identifies the spatial relationship between cold aisles, hot aisles, cabinet clusters, and cooling units based on the data center's architectural drawings and equipment layout information. A dynamic attribute mapper that associates and binds the regional energy efficiency feature vector calculated by the energy consumption feature extraction unit to the corresponding spatial position identified by the spatial topology parser; A multi-scale model generator generates a hierarchical three-dimensional virtual model including thermodynamic properties based on the spatial position relationship output by the spatial topology analyzer and the energy efficiency characteristics bound by the dynamic property mapper.

[0009] Preferably, the operation situation deduction engine includes: A thermal field dynamic tracking engine simulates the spatiotemporal evolution of airflow organization and temperature field based on the hierarchical three-dimensional virtual model output by the multi-scale model generator; Load migration simulator, simulating the dynamic scheduling path and resource usage changes of computing tasks between different server nodes; The energy consumption coupling analyzer correlates the thermal field spatial distribution output by the thermal field dynamic tracking engine with the resource occupancy change output by the load migration simulator to calculate a global energy consumption hotspot map.

[0010] Preferably, the abnormal behavior identification module includes: A threshold deviation detection unit compares the global energy consumption hotspot map generated by the energy consumption coupling analyzer with a preset energy efficiency benchmark interval in real time; a pattern drift analysis unit for identifying the aggregation characteristics of the deviation data found by the threshold deviation detection unit in the time dimension and the diffusion trend in the space dimension; The anomaly root cause locator associates the aggregation characteristics and diffusion trends identified by the pattern drift analysis unit, marks the anomaly source physical component and associates the affected link.

[0011] Preferably, the decision instruction generation module includes: The policy knowledge base retriever matches the optimization policy template in the historical handling case library based on the abnormal source physical components and associated impact links marked by the abnormal root cause locator; A multi-objective optimization calculator calculates adjustment amounts of control parameters based on the optimization strategy template retrieved by the strategy knowledge base retriever and in combination with the hierarchical three-dimensional virtual model state in the current virtual space reconstruction module; An instruction sequence assembler converts the control parameter adjustment amount output by the multi-objective optimization calculator into an executable control instruction sequence.

[0012] Preferably, the system further comprises: a fault propagation deduction module, configured to simulate a transmission path of a single point fault in the three-dimensional virtual model of the virtual space reconstruction module based on the associated impact links marked by the abnormal root cause locator; a risk level assessor, configured to calculate a system-level risk entropy value based on the depth and breadth of the conduction path simulated by the fault propagation deduction module; The plan matching engine is used to activate the disaster recovery plan of the corresponding level according to the risk entropy value calculated by the risk level assessor.

[0013] Preferably, the fault propagation deduction module includes: A dependency mining unit analyzes the physical connections and logical dependencies between the cabinet cluster, the cooling unit, and the power supply circuit in the virtual space reconstruction module; A failure impact simulator injects the abnormal source fault signal marked by the abnormal root cause locator into the three-dimensional virtual model constructed by the multi-scale model generator to track the propagation trajectory of the state abnormality; The critical path identifier marks the device node set with the highest fault conduction relationship strength according to the abnormal state propagation trajectory output by the failure impact simulator.

[0014] Preferably, the risk level assessor includes: An impact range measurement unit calculates the spatial coverage ratio of the set of device nodes marked by the critical path identifier in the three-dimensional virtual model; a service interruption assessment unit, associated with the resource scheduling path of the load migration simulator, and quantifying the delay increment and interruption probability of the affected computing task; The risk entropy value aggregator weightedly aggregates the spatial coverage ratio calculated by the impact range measurement unit and the delay increment and interruption probability quantified by the business interruption assessment unit, and outputs a comprehensive risk level label.

[0015] Preferably, the plan matching engine includes: A level response mapper, which indexes a pre-configured emergency response level directory according to the comprehensive risk level label output by the risk entropy value fusion device; A resource scheduling preconfigurator generates a backup refrigeration equipment startup priority and a computing task migration target matrix based on the emergency response level determined by the level response mapper; The multi-source decision instruction integrator fuses the executable control instruction sequence of the decision instruction generation module with the migration target matrix generated by the resource scheduling preconfigurator to generate a final execution instruction package.

[0016] Compared with the prior art, the present invention has the following beneficial effects: Through the collaborative operation of multiple modules, a real-time mapping and interaction mechanism between physical facilities and virtual models has been established. The multi-dimensional operating parameters collected in real time by the data center's physical characteristic perception module provide precise data input for virtual space reconstruction, enabling the 3D virtual model to dynamically reflect the true state of physical equipment, including operating parameters, spatial layout relationships, and resource load. This breaks down the information barriers between physical and virtual spaces in traditional management.

[0017] The operational simulation engine simulates resource scheduling and energy flow using a three-dimensional virtual model. It can replicate various data center operational scenarios in a virtual environment, including resource allocation effects, energy consumption paths, and potential bottlenecks under varying business loads. This simulation capability allows managers to rehearse various control strategies without disrupting physical system operations, thus mitigating the risks of blind decision-making and providing a flexible testing ground for exploring optimization solutions.

[0018] The Abnormal Behavior Identification module analyzes simulated data streams in real time, promptly detecting unusual patterns that deviate from preset thresholds. This module identifies not only parameter anomalies at a single device but also system-wide coordination disorders, such as chain reactions caused by resource scheduling imbalances and localized overloads caused by abnormal energy flows. This comprehensive anomaly detection capability, compared to traditional single-point monitoring, better reflects the system's integrity and interdependence, helping to identify potential risks in advance and reduce the probability of failures.

[0019] The decision-making instruction generation module integrates the optimization and control instructions generated by anomaly identification results and can directly apply them to physical devices, achieving a closed loop from virtual simulation to physical control. These instructions are based on an analysis of the overall system operation status, balancing resource utilization efficiency and energy consumption. For example, they can dynamically adjust the operating status of server clusters when business load fluctuates, reducing energy consumption while ensuring service quality, or optimize the operating parameters of air conditioning systems based on cooling requirements to improve cooling efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a timing diagram of the data center digital twin simulation and decision-making system for intelligent management according to the present invention; Figure 2 Flowchart of the work of the virtual space reconstruction module; Figure 3 A flowchart for running the situational simulation engine; Figure 4 A flowchart of the work of the decision instruction generation module; Figure 5 Flowchart of the fault propagation simulation module. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] See also Figure 1 The present invention provides a data center digital twin simulation and decision-making system for intelligent management, the system comprising: By establishing a real-time mapping relationship between physical and virtual space, intelligent monitoring, simulation, and optimized control of the data center are achieved. The system includes a data center physical feature perception module, a virtual space reconstruction module, an operational situation deduction engine, an abnormal behavior identification module, and a decision-making instruction generation module. The physical feature perception module collects multi-dimensional operational parameters of the infrastructure in real time, including environmental conditions, equipment load, and energy consumption characteristics; the virtual space reconstruction module dynamically generates a three-dimensional virtual model based on the collected data; the operational situation deduction engine simulates resource scheduling and energy flow processes; the abnormal behavior identification module detects abnormal patterns that deviate from preset thresholds; and the decision-making instruction generation module generates optimization and control instructions based on the abnormal analysis results, forming a closed-loop management from data collection to decision execution.

[0023] Example 1: See Figure 2The environmental parameter acquisition unit of the data center's physical feature perception module is deployed using a distributed sensor network. The temperature sensor uses a platinum resistance PT1000 model to monitor temperature changes at cabinet air inlets, air outlets, and key points in the channel with an accuracy of 0.1°C. The humidity sensor is based on a capacitive measurement principle, with a range covering 10% to 90% relative humidity and a sampling interval of 5 seconds. A hot wire anemometer is used to monitor air flow velocity, with a cross-shaped array of measurement points arranged at the junction of hot and cold channels to record vector data of air flow. All sensors are networked via the RS-485 bus and use the Modbus RTU protocol to transmit data in real time to the edge computing gateway.

[0024] The device status monitoring unit collects heterogeneous device data through multiple interface protocols. Server power consumption data is obtained via the SNMP protocol of the intelligent rack PDU, with a sampling frequency of 1Hz. Calculated load data queries the BMC controller through the IPMI interface to obtain indicators such as CPU core temperature, utilization, and memory occupancy. Storage device status collects disk IOPS and latency data through the Redfish API. Refrigeration equipment monitoring connects to the chiller controller via the BACnet protocol to read parameters such as chilled water flow, supply and return water temperatures, and compressor operating frequency. After collection, all device data is uniformly affixed with a timestamp and device location tag, forming a raw data stream with time and space identification.

[0025] The energy consumption feature extraction unit performs multi-stage processing on the raw data. The data cleaning stage uses a sliding window anomaly detection algorithm to remove outliers caused by transient sensor failures. The feature calculation stage calculates the 90th percentile of temperature data, the coefficient of variation of humidity data, and the standard deviation of airflow velocity for each cabinet area. The energy efficiency index calculation stage integrates equipment power consumption and environmental parameters to calculate the cooling efficiency factor and heat accumulation index for each cabinet area. The resulting regional energy efficiency feature vector includes a dynamic weight adjustment mechanism that automatically increases the weight coefficient of temperature-related features in high-temperature areas.

[0026] The spatial topology parser in the virtual space reconstruction module processes multiple input data sources. Building structural data is imported into the BIM model in IFC format to analyze the structural relationships between the building's load-bearing columns, anti-static flooring, and ceiling space. Equipment layout data is parsed from cabinet CAD drawings to identify cabinet U-position numbers, network patch panel locations, and the spatial coordinates of air conditioning vents. During the topology construction phase, graph theory algorithms are used to calculate the Euclidean distance from each cabinet to the nearest air conditioning vent and establish a "cabinet-air conditioning" association matrix. Spatial indexing is managed using an octree structure, enabling rapid querying of equipment distribution around any spatial location.

[0027] A dynamic attribute mapper enables the association and binding of physical data with virtual objects. Attribute mapping tables are stored in a distributed key-value database, with the primary key being a hash of the device's location coordinates and the value domain consisting of real-time collected data and calculated features. Mapping updates utilize a publish-subscribe model. When physical sensor data changes, the corresponding virtual model components are notified via the messaging middleware. Spatial attribute binding supports dynamic adjustment. When the air conditioner adjusts the airflow rate, the airflow velocity attributes of the affected cabinet areas are automatically updated.

[0028] The multi-scale model generator constructs hierarchical 3D visualization models. The macro model displays the entire data center's architectural outline and key equipment distribution, using lightweight rendering technology to smoothly display large-scale scenes. The meso model focuses on a single computer room area, using physically based rendering technology to depict the temperature distribution and airflow organization on the cabinet surface. The micro model supports device-level viewing, displaying the hotspot distribution and component temperatures within a single server. Model updates utilize a differential synchronization mechanism, transmitting only the modified data to reduce network transmission load. Model interaction supports zooming, cross-section viewing, and time-space playback, allowing operations and maintenance personnel to observe the data center's operating status from multiple angles.

[0029] The model rendering engine utilizes real-time global illumination technology to accurately represent the thermal radiation effects of different material surfaces. Thermal image rendering of the cabinet's metal surfaces utilizes fragment shader programming, dynamically adjusting color mapping based on temperature data. Airflow visualization utilizes a particle system to plot the movement of hot and cold airflows in three-dimensional space. All visualization effects support VR device integration, providing an immersive data center inspection experience. Model data storage utilizes a time-series database architecture to fully record historical state changes and support time-based playback of operational scenarios at any point in time.

[0030] During system implementation, the physical perception module and the virtual modeling module maintain consistency through high-precision time synchronization. A network clock protocol ensures that all collected data timestamps are accurate within milliseconds. A data consistency check mechanism regularly compares physical sensor readings with virtual model attributes, triggering a recalibration process when discrepancies are detected. In the event of sensor failure, the fault-tolerance module performs short-term predictions based on historical data patterns to maintain the continuous operation of the virtual model.

[0031] The entire implementation process strictly adhered to data center operation and maintenance specifications. All data collection operations were performed through standard device interfaces to avoid interference with normal system operations. The accuracy of the virtual model was controlled to within ±1cm of the actual physical dimensions, and the resolution of the temperature field simulation reached 0.5°C. The system deployment adopted a modular architecture, allowing for flexible expansion of sensing nodes and modeling computing power based on data center scale. The operation and maintenance interface provides a standardized data access API, facilitating integration and data exchange with existing monitoring systems.

[0032] During system operation, the physical feature perception module continuously generates data streams, which are then cleaned and feature extracted before being fed into the virtual space reconstruction module. The reconstruction module dynamically updates the 3D model state, keeping pace with changes in the physical world. This real-time mapping mechanism provides an accurate virtual environment foundation for subsequent situational analysis and anomaly detection. The entire data processing process utilizes a pipeline architecture, keeping end-to-end latency from data acquisition to model update within 3 seconds, meeting real-time requirements.

[0033] The interactive interface of the virtual model provides a variety of analysis tools. The thermodynamic analysis tool calculates the heat load density of any area, and the airflow analysis tool simulates changes in airflow distribution after adjusting air conditioning parameters. The spatial query function supports filtering and viewing by device type, temperature range, or energy efficiency level. The comparative analysis function displays the operating status at different time points side by side, visually illustrating parameter change trends. All analysis results can be generated into standardized reports to assist in operation and maintenance decision-making.

[0034] Various practical constraints were considered during system implementation. For sensor deployment, a hybrid wireless and wired network was adopted, with low-power wireless sensor nodes used in areas where cabling was difficult. For model accuracy, adaptive level-of-detail technology was employed for different regions, maintaining high-precision modeling in critical areas while appropriately reducing model complexity in less important areas. For computing resource allocation, a load balancing mechanism was established to dynamically adjust the priority of simulation computing tasks during peak business hours.

[0035] Example 2: See Figure 3 The thermal field dynamic tracking engine of the operation situation simulation engine uses computational fluid dynamics methods to simulate the temperature field. The simulation area is discretized into hexahedral grid units, the side length of each unit is set to 0.5 meters, and the total number of grids is dynamically adjusted according to the actual scale of the data center. In the boundary condition setting, the air supply outlet of the air conditioner is defined as the velocity inlet boundary, and the air outlet of the cabinet is defined as the pressure outlet boundary. The solver uses the SIMPLE algorithm to iteratively solve the three-dimensional unsteady heat transfer equation, and the time step is set to 5 seconds. During the thermal field calculation process, the server power consumption is loaded as a volume heat source onto the grid unit corresponding to the cabinet position. The thermal field simulation results are output in the form of a temperature cloud map, and the color mapping range is from blue at 20°C to red at 45°C, which intuitively shows the distribution and evolution of the hot zone.

[0036] Accuracy control of thermal field simulations is achieved through mesh independence verification. A coarse mesh is initially used for calculations, and the mesh is gradually refined until the temperature variation in critical areas is less than 0.3°C. A mesh adaptation algorithm automatically adjusts local mesh density based on temperature gradients, using a denser mesh in areas of high gradients. The computational convergence criterion is set to reduce the residual error to below 10^-4 to ensure the reliability of the simulation results. Thermal field data is stored in a hierarchical structure, with the original mesh data retained in cache and the simplified model used for rendering stored in graphics memory.

[0037] The load migration simulator constructs a task scheduling model based on queuing theory. Computing tasks are abstracted as tuples representing the number of CPU cores, memory size, and storage I / O requirements. Physical hosts are modeled as a multi-server queuing system, with each server corresponding to a CPU core. The task arrival process is simulated using a Poisson distribution, and service times follow a negative exponential distribution. The migration decision algorithm comprehensively considers the target node's remaining resources, network bandwidth, and migration overhead to select the optimal target host. The migration process simulation includes realistic influencing factors such as memory page transfer delays and packet retransmissions caused by network congestion. The simulator outputs resource utilization curves and task queue waiting time statistics for each physical host.

[0038] The energy consumption coupling analyzer establishes a correlation model between the thermal field and the load. It receives temperature distribution data from the thermal field dynamic tracking engine and power consumption data from the load migration simulator. Using a spatial interpolation algorithm, the analyzer maps the discrete point data into a unified coordinate system. The coupling analysis is performed using a sliding time window with a 15-minute window length and a 1-minute sliding step. Within each analysis cycle, the correlation coefficient between the average temperature and average power consumption of each cabinet area is calculated to identify areas with abnormal thermal-electrical coupling strength.

[0039] The key indicator calculation formula used in the energy consumption coupling analysis process is:

[0040] in: represents the correlation coefficient between the power consumption of the i-th cabinet and the j-th temperature monitoring point, is the power consumption of the i-th cabinet at time t, is the reading of the jth temperature monitoring point at time t, and represents the time average of the corresponding variable, and T is the number of sampling points in the analysis time window. This coefficient quantifies the spatiotemporal correlation between device power consumption and ambient temperature. Its value range is [-1, 1]. Positive values ​​indicate a strong positive correlation, while negative values ​​indicate a strong negative correlation.

[0041] The threshold deviation detection unit of the abnormal behavior identification module utilizes a dynamic threshold mechanism. The base threshold is derived from the safe operating range specified in the device's technical manual, with dynamic adjustments calculated based on the statistical characteristics of historical data. For temperature data, a sliding window is used to calculate the mean and standard deviation, with the dynamic threshold set at ±3 times the standard deviation. Threshold detection for power consumption data considers differences in daytime and nighttime patterns, establishing separate reference benchmarks for daytime and nighttime. The detection algorithm utilizes incremental calculations, retaining only summary statistics within the window, significantly reducing computational complexity.

[0042] The pattern drift analysis unit uses time series pattern mining technology. The analysis window is set to 30 minutes to detect the persistence of abnormal patterns. Temporal dimension analysis uses a change point detection algorithm to identify moments when statistical characteristics suddenly change. Spatial dimension analysis constructs an anomaly propagation graph, where nodes represent devices or areas, and edges represent the strength of anomaly associations. Propagation velocity calculation uses time delay correlation analysis to determine the time difference between anomalies spreading from one area to another. The analysis results output the diffusion direction and velocity predictions for anomaly hotspots.

[0043] The anomaly root cause locator implements an analysis process based on causal reasoning. It first constructs a device association graph, where nodes include devices such as air conditioners, cabinets, and power distribution units, and edges represent physical connections or logical dependencies. Anomaly propagation simulation uses a message passing mechanism, starting from the anomaly detection point and transmitting the anomaly signal along the associated edges to adjacent nodes. A root cause scoring algorithm comprehensively considers anomaly intensity, propagation path length, and device criticality to rank potential root cause devices. The location results are output as probabilistic estimates of the likelihood of each device being the fault source.

[0044] In the system implementation, the operational scenario simulation engine utilizes a distributed computing architecture. Thermal field simulation tasks are divided into multiple sub-areas and assigned to different compute nodes for parallel processing. The load migration simulator utilizes an event-driven mechanism, processing migration requests in chronological order. The energy consumption coupling analyzer is implemented as a stream processing pipeline, continuously receiving input data and outputting analysis results in real time. A dynamic computing resource scheduling algorithm allocates CPU cores and memory resources based on task priority to ensure the timely completion of critical simulation tasks.

[0045] The real-time processing capabilities of the abnormal behavior identification module are achieved through pipeline optimization. The data preprocessing stage utilizes a parallel filtering algorithm to simultaneously clean and extract features from multiple data streams. The threshold detection stage is implemented as a stateless service, supporting horizontal scalability to cope with data spikes. The pattern analysis stage utilizes an incremental learning algorithm to continuously update model parameters to adapt to system state changes. The root cause location stage employs a heuristic search strategy to locate the most likely fault source within a limited timeframe.

[0046] In the visualization interface, operational status simulation results are displayed as dynamic heat maps and resource utilization dashboards. The heat map supports dragging the timeline to view historical status, and key parameters can be overlaid on the 3D model. Anomaly detection results are color-coded, with minor anomalies displayed as yellow warnings and severe anomalies as red alerts. Root cause analysis paths are highlighted in the device association diagram, making it easier to track fault propagation chains. All visualization elements support interactive querying; click on any device to view detailed parameters and historical trends.

[0047] The data storage solution is optimized for different data types. Thermal simulation results are stored using block compression, with data for each time step saved as a separate file. Load migration records are stored in a time series database, supporting efficient time range queries. Abnormal event data is stored in a document database, fully preserving event context. All data is accompanied by rich metadata, including acquisition time, data source, and quality identification.

[0048] During system operation, the operational situation simulation engine continuously consumes real-time data provided by the physical feature perception module and generates predictive state evolution trends through simulation calculations. The abnormal behavior identification module analyzes discrepancies between simulation results and measured data, detecting potential issues and locating their root causes. These two modules form a closed-loop analysis chain, from observation to fundamental understanding, providing a comprehensive state assessment for subsequent decision-making. The entire processing process maintains a time delay of less than 10 seconds, meeting the requirements of real-time monitoring.

[0049] Fault-tolerance mechanisms ensure stable system operation. The data validation module checks the input data for acceptable ranges and filters out obviously erroneous readings. The computation monitoring module tracks the progress of each simulation task and automatically restarts or downgrades tasks that time out. The exception handling module captures runtime errors, isolates faulty components, and activates backup services. The state persistence module regularly saves system checkpoints, enabling rapid recovery after unexpected interruptions.

[0050] The system deployment plan takes into account practical environmental constraints. When computing resources are limited, the grid density or simulation frequency of thermal field simulations can be reduced. When network bandwidth is limited, data compression and differential transmission technologies are used to reduce communication traffic. When integrating with existing monitoring systems, standardized data interfaces and protocol conversion adapters are provided. System configuration parameters are centrally managed through a graphical interface, supporting online adjustments and immediate effectiveness.

[0051] This implementation builds a comprehensive digital image of the data center's operating status through sophisticated physical process modeling and intelligent anomaly analysis algorithms. Thermal field simulation recreates the dynamic changes in temperature distribution, load migration simulation predicts the effectiveness of resource scheduling, and coupled analysis reveals the complex relationship between energy consumption and heat dissipation. Anomaly detection not only identifies surface symptoms but also delves deeper into potential root causes. This approach provides in-depth status awareness and problem diagnosis capabilities for intelligent data center management.

[0052] Example 3: See Figure 4 The policy knowledge base retriever of the decision instruction generation module adopts a semantic-based case matching mechanism. The knowledge base is structured into a three-layer index system: the top layer is classified by device type, the middle layer is clustered by fault phenomenon, and the bottom layer stores specific disposal plans. The case representation is described in ontology language, which includes a set of fault characteristics, environmental context information, execution operation sequence and effect evaluation indicators. The retrieval algorithm combines exact matching and fuzzy search. When the fault description provided by the abnormal root cause locator is input, the key features are first accurately matched, and then similar historical cases are found through similarity calculation. The similarity calculation considers the weighted comparison of multiple dimensions such as fault location, equipment model, environmental parameters, etc., and returns the top K candidate solutions with the highest matching degree.

[0053] The multi-objective optimization calculator processes the candidate solutions output by the policy knowledge base searcher and constructs a decision model that incorporates multiple conflicting objectives. Optimization variables include adjustable parameters such as air conditioner setpoint temperature, fan speed, and virtual machine distribution ratio. The objective function simultaneously considers temperature uniformity, energy efficiency, and business continuity. Temperature uniformity is measured by the standard deviation of cabinet temperatures, energy efficiency is calculated by total power consumption, and business continuity is assessed by the number of task migrations. Constraints include equipment safe operation thresholds, resource capacity limits, and service level agreement requirements.

[0054] The trade-off decision formula used in the multi-objective optimization process is:

[0055] in: represents the comprehensive score of decision plan x, is the weight coefficient of the kth target, is the value of plan x on the kth objective, and The weight coefficients are determined using the Analytic Hierarchy Process (AHP) and dynamically adjusted based on the data center's current operating mode, prioritizing business continuity during peak hours and prioritizing energy efficiency during energy-saving periods. The optimization solution uses a genetic algorithm with an elite retention strategy, with a population size of 100 and 50 iterations. The final Pareto optimal solution set is output for selection by operations personnel.

[0056] The instruction sequence assembler converts optimization results into executable control commands. The conversion rule base defines the device control protocols and command formats corresponding to various parameter adjustments. For example, air conditioner temperature adjustment is converted to the BACnet protocol's WriteProperty command, and virtual machine migration is converted to the OpenStack API's live-migration request. The timing orchestrator analyzes the dependencies between operations and generates a topologically sorted execution sequence to ensure that cooling system adjustments precede computing load migration. The security verification module verifies the parameter range of each command to prevent the issuance of dangerous instructions. The resulting instruction package, containing complete operation steps, expected results, and a rollback plan, is sent to the device controller via an encrypted channel.

[0057] The dependency mining unit of the fault propagation deduction module analyzes the multi-layer dependency structure of the data center. Physical layer dependencies, including power supply links, cooling lines, and network cabling, are constructed by parsing the power distribution system single-line diagram and piping layout diagram. Logical layer dependencies, including virtual machine affinity rules, storage replication relationships, and network traffic paths, are extracted from monitoring system configuration information. Dependencies are modeled as a weighted directed graph, where nodes represent devices or software components, edges represent dependency relationships, and weights indicate dependency strength. The graph structure uses an adjacency table to store dependencies, enabling fast traversal and subgraph extraction.

[0058] The Failure Impact Simulator implements rule-based fault propagation simulation. Taking the fault source node provided by the anomaly root cause locator as input, the simulator performs a breadth-first search along the dependency graph edges, marking all potentially affected nodes. The propagation rule base defines the failure probability of different types of dependencies, such as setting the failure probability of power dependencies to 0.9 and network dependencies to 0.6. The state transition model simulates the state transition of a device from normal to faulty, taking into account practical factors such as fault detection latency and redundancy switchover time. The simulation results record the failure time and state change trajectory of each node, generating a complete spatiotemporal diagram of the fault propagation.

[0059] The impact measurement unit of the risk level assessor calculates the proportion of physical and logical space affected. Physical space measurement measures the percentage of cabinets affected by the failure, while logical space measurement measures the proportion of affected virtual machines in the cluster. The service interruption assessment unit analyzes the degree of service degradation and calculates the expected deviation from core business metrics, such as the increase in database query latency and the decrease in video streaming bitrate. The assessment model distinguishes between the impact of brief interruptions and persistent failures, using cumulative damage theory to quantify the hidden costs of prolonged low-performance operation.

[0060] The risk entropy aggregator implements a multi-dimensional risk assessment algorithm. The input layer receives raw data on three dimensions: impact scope, business interruption, and service level. Normalization transforms each dimension linearly to the [0, 1] range. The entropy weighting method calculates the objective weight of each dimension, reflecting the degree of data dispersion. This weighted fusion outputs a risk score from 0 to 100, which is categorized into four levels: caution, warning, critical, and emergency based on preset thresholds. The assessment results are accompanied by a detailed explanation of the basis, listing the main risk factors and key impact paths.

[0061] The plan matching engine's level response mapper maintains an emergency response strategy matrix. Matrix rows represent risk levels, columns represent device types, and cells store corresponding response processes. The resource scheduling preconfigurer dynamically allocates backup resources based on risk levels. For example, under emergency levels, 15% of computing resources are automatically reserved for fault recovery. The plan validation module test-runs the plan process in a virtual environment to check for resource conflicts and operational inconsistencies, ensuring plan feasibility. Version management tracks plan modification history and supports rapid rollback to stable versions.

[0062] A multi-source decision-making command integrator coordinates optimization commands with emergency response plans. A conflict detection algorithm identifies resource contention and conflicting objectives between commands. For example, if cooling optimization requires reducing air conditioning power while a fault response plan requires increasing cooling capacity, the coordination strategy employs a priority mechanism, prioritizing fault recovery commands over routine optimization commands. The resulting execution command package includes detailed timelines, resource allocation, and exception handling instructions, creating a complete, traceable, and auditable record of operations.

[0063] In the system implementation, the decision-making instruction generation module adopts a microservices architecture. The policy knowledge base service is deployed independently, supporting concurrent queries from multiple clients. The optimization computing service runs on GPU accelerators to shorten solution time. The instruction conversion service is deployed on edge nodes close to the device to reduce control latency. All services communicate asynchronously through message queues to improve system resiliency.

[0064] The fault propagation simulation module is implemented as a stateful service, maintaining an up-to-date dependency graph. The graph update service listens for configuration change events and adjusts dependencies in real time. The simulation executor utilizes an event-driven architecture to efficiently handle large numbers of parallel simulation requests. The risk assessment service caches intermediate results to accelerate the continuous assessment process. The visualization service generates an interactive fault tree, visually displaying propagation paths and key nodes.

[0065] During system operation, the decision-making instruction generation module receives anomaly analysis results and generates a control plan through case retrieval, optimization calculations, and instruction conversion. The fault propagation deduction module predicts potential risks, assesses the impact, and develops a response plan. These two modules work together to resolve the current anomaly and prevent cascading failures, forming a comprehensive decision support system. The processing process maintains a complete logical trajectory, supporting post-event review and analysis.

[0066] Fault-tolerance mechanisms ensure the reliability of the decision-making process. Input validation filters out invalid parameters and prevents error propagation. Solution monitoring monitors the convergence of optimization algorithms and switches to alternative methods when anomalies occur. Instruction validation ensures that control commands comply with safety regulations. A rollback mechanism automatically restores the previous state in the event of instruction execution failure. State persistence regularly saves decision context, allowing execution to resume from a checkpoint after an interruption.

[0067] The system deployment plan takes actual operational and maintenance needs into consideration. Decision services can be deployed independently and integrated with existing management systems through standard interfaces. The deduction module supports distributed deployment, allowing large-scale data centers to be divided into zones for parallel evaluation. The resource monitoring module dynamically adjusts computing resource allocation, prioritizing critical services during peak hours. The configuration interface provides intuitive policy editing tools, lowering the barrier to entry.

[0068] Example 4: See Figure 5 The dependency mining unit of the fault propagation deduction module first constructs a device association graph for Area A. This module analyzes the cooling system piping layout and identifies CRAC-03, which is connected to the main pipe via chilled water branch pipe VP-12 and is responsible for cooling four cabinet areas, A12-A15. The electrical system diagram shows that these cabinets are powered by PDU-A2, and the network topology indicates that the order processing system's database cluster runs within these cabinets. The system automatically generates a device dependency table. See Table 1.

[0069] Table 1: Device dependency table.

[0070]

[0071] The Failure Impact Simulator received the CRAC-03 fault signal from the root cause locator and initiated multiple rounds of propagation simulations. The first round of simulations revealed that a drop in chilled water flow caused the temperature of cabinet A12 to rise by 5°C over 8 minutes, triggering the server fans to operate at full speed. The second round of simulations revealed that the surge in power consumption in cabinet A12 triggered the overload protection of PDU-A2, which in turn caused cabinet A13 to lose power. The third round of simulations captured the database master node triggering a failover due to the power outage in cabinet A13, but network congestion extended the failover time to 45 seconds. The simulations recorded the time of each node's state change, forming a complete failure timeline.

[0072] The critical path identifier (CPI) analyzed the simulation results and identified three critical transmission paths: cooling path CRAC-03 → A12 → database service outage; power path CRAC-03 → A12 → PDU-A2 → A13; and network path A13 → DB-Master → Core-SW1 → web service degradation. Based on the transmission speed and impact range, the system labeled PDU-A2 and DB-Master as critical bottleneck devices.

[0073] The risk level assessor's impact measurement unit calculates the physical space impact. Area A has 20 cabinets, of which four are affected, accounting for 20%. Logical space measurement revealed that seven of the 12 microservices in the order processing system depend on the faulty database, accounting for 58.3%. The space heat map shows that the temperature in areas A12-A15 has exceeded the upper limit recommended by ASHRAE.

[0074] The business interruption assessment unit analyzed the service impact details. The average response time for the order query API increased from 120ms to 2100ms, and the payment transaction success rate decreased from 99.98% to 91.23%. The assessment model identified the most serious business impact as an increase in the rate of shopping cart timeouts, resulting in a 35% increase in the probability of customers adding duplicate items. The system automatically correlated the SLA terms and calculated the service credit compensation clauses that might be triggered by this outage.

[0075] The risk entropy amalgamator performs a comprehensive multi-dimensional assessment. Input data includes: physical impact range (20%), logical impact range (58.3%), core business indicator deviation (40%), and SLA default probability (72%). Using the entropy weighting method, the weights for each dimension are calculated as 0.35, 0.25, 0.2, and 0.2, respectively. This weighted amalgamation outputs a comprehensive risk value of 68, corresponding to a "critical" risk rating. The assessment report details the contribution of each influencing factor, with the SLA default probability contributing the most, at 38%.

[0076] After receiving the risk assessment results, the plan matching engine initiates the response process. The level response mapper retrieves the plan corresponding to the "Severe" level, which includes: immediately activating the backup air conditioner CRAC-03B, migrating the load from cabinets A12-A15 to Area B, and enabling database read-write splitting mode. The resource scheduling preconfigurer checks that Area B currently has 25% remaining cooling capacity, which meets the migration requirements but requires limiting the amount of new load. The system generates a detailed operation schedule, requiring the air conditioner switchover to be completed within 5 minutes and the computing load migration to be completed within 15 minutes.

[0077] The multi-source decision-making command integrator coordinates operational commands from multiple systems. First, it sends a command to the building management system to power on the CRAC-03B and set the supply air temperature to 18°C. It then notifies the virtualization management platform to migrate the database node from cabinet A12 to cabinet B09 and configures a load balancing policy to direct 30% of query traffic to the backup database. Finally, it requests priority bandwidth for database synchronization traffic from the network management system. All commands are accompanied by execution condition checks; if the temperature in area B is abnormal, the migration is immediately halted.

[0078] During system implementation, the fault propagation simulation module adopted a layered simulation strategy. The foundation layer simulates fault transmission between physical devices, the middle layer handles resource scheduling logic, and the application layer analyzes the impact on business services. This layered structure ensures simulation efficiency while ensuring that critical business impacts are not missed. The simulation process supports pause and single-step execution, allowing operations personnel to observe the evolution details of specific links.

[0079] The risk level assessor implements a dynamic weighting mechanism. It automatically increases the weight of service indicators during peak business hours and prioritizes equipment safety parameters during nighttime maintenance windows. The assessment model incorporates industry benchmark data, such as comparing temperature rise rates with typical values ​​for similar data centers, to enhance the objectivity of the assessment results.

[0080] The visual interface presents the complete analysis process. The 3D model displays the fault propagation path with a red pulse effect, and a gradient bar chart indicates the risk level for different areas. The business impact dashboard updates key performance indicators in real time, comparing normal ranges to current values. The operation instruction list displays execution status, with completed steps marked green and ongoing steps displayed with a progress bar.

[0081] System deployment takes actual operational scenarios into consideration. The simulation service is deployed on independent computing nodes to avoid impacting production system performance. The assessment engine utilizes a distributed architecture, supporting parallel analysis of multiple fault scenarios. Integration with existing monitoring systems is achieved through adapters, automatically converting alarm information into simulation input. Operational instructions are routed through an approval workflow, and critical operations require secondary confirmation before execution.

[0082] In actual use at Data Center Area A, the system successfully predicted the potential cascading effects of a CRAC-03 failure. Based on the prediction results, the operations team preemptively reinforced the PDU-A2 power supply and adjusted the failover timeout settings for the database cluster. When a refrigerant leak actually occurred at CRAC-03, the system triggered a contingency plan that limited service interruption to under two minutes, significantly shorter than historically reported handling times for similar incidents. After troubleshooting, the system automatically generated an analysis report documenting the deviation between the actual propagation path and the predicted results, which was used to optimize subsequent prediction models.

[0083] This implementation demonstrates the complete process of fault propagation simulation and risk assessment using specific equipment failure cases. From dependency building to multiple rounds of state simulation, critical path identification, and comprehensive risk calculation, it ultimately forms an actionable emergency plan. Practical applications have proven that this method can effectively predict chain reactions in complex systems, providing forward-looking decision support for data center operations and maintenance.

[0084] Example 5: Demonstrating a closed-loop management process from risk assessment to command execution using an actual data center operation and maintenance scenario. When a voltage fluctuation occurred in the power distribution system in the eastern area of ​​the data center, the system fully executed the emergency response process. The risk level assessor detected that the UPS input voltage was continuously outside the normal range, with a frequency of 3-5 fluctuations per minute. A comprehensive assessment determined this to be a Level 3 risk event. The level response mapper immediately initiated a search of the corresponding emergency plan library, matching the pre-configured power anomaly response plan based on the risk characteristics.

[0085] The plan library is organized in a multi-level index structure. The first level is classified by risk type, the second level is divided by equipment area, and the third level stores specific operation steps. The current event is classified into the "Abnormal Power Quality / Eastern District / UPS Input Fluctuation" plan branch. The system retrieves three related plans: backup circuit switching plan, load level reduction plan, and generator start-up preparation plan. Each plan is accompanied by a description of the applicable conditions, including the voltage fluctuation amplitude threshold, duration requirements, and equipment status prerequisites. The plan selection algorithm comprehensively considers the current environmental parameters and automatically selects the backup circuit switching plan as the main execution plan. The remaining plans enter the activation queue.

[0086] The resource scheduling preconfigurator analyzes the operational requirements of the selected plan and begins resource preparation. The computing module assesses the current load distribution in the East Zone, identifies a list of VMs eligible for migration, and generates a migration plan based on business priorities. The cooling system checks the status of backup air conditioning units and calculates the cooling capacity balance after the switchover. The network module reserves emergency communication bandwidth to ensure smooth transmission of control commands. The status of all resource preparations is continuously updated via a heartbeat mechanism, and any resource shortage triggers dynamic adjustments to the plan.

[0087] The plan verification phase simulates the entire execution process within a digital twin environment. The system first clones the current operating state to create a sandbox environment, then executes the plan steps step by step. The simulation reveals that a direct switchover could temporarily cause the storage system to lose connectivity. The plan optimization module then inserts instructions to warm up the storage controller. After verification, an execution timeline is generated, noting the ideal execution time and maximum tolerable latency for each step. Checks are implemented at key operation points, such as requiring confirmation of the physical switch status before powering on.

[0088] The multi-source decision-making command integrator began coordinating the multi-system command sequence. The power management system received the first command: Close the static switch of the backup circuit. This command's execution condition was that the input voltage fluctuated for more than 10 seconds. The virtualization management platform received the second command: Prepare to migrate the virtual machines in Group A of the East District, but do not actually perform the migration. The building control system received the third command: Start the backup air conditioning unit but keep it in standby mode. All commands are bound together through a transaction mechanism; the failure of any command triggers a rollback of the entire system.

[0089] The instruction packet generation module consolidates discrete operations into traceable task units. Each task unit includes an operation objective, execution conditions, expected results, timeout settings, and compensation measures. A dependency graph is established between tasks, ensuring, for example, that load migration can only begin after the backup circuit is stable. The system automatically generates human-readable instructions detailing each step's procedures and precautions, providing graphic guidance to field engineers as they work with the automation system.

[0090] During the plan execution phase, a dual-channel control mechanism is employed. The automated channel directly controls the system through a standard interface to execute pre-set operations, while the manual channel pushes key instructions to the operations and maintenance personnel's console for confirmation. For steps involving physical equipment manipulation, such as manually switching a circuit breaker in a power distribution cabinet, the system generates a work ticket with a QR code. On-site personnel scan the code to obtain operational instructions and feedback on the execution results. The status of each system is monitored in real time during execution, and abnormalities trigger dynamic adjustments.

[0091] During the East District voltage fluctuation incident, the system completed all contingency plan preparations within 90 seconds. When a voltage fluctuation was detected lasting 15 seconds, a backup circuit switchover was automatically triggered. During the switchover process, if a storage node failed to switch properly, the system immediately initiated a compensation plan: restoring high-priority virtual machines first while isolating the faulty storage volume. The entire switchover process was ultimately completed within 2 minutes and 30 seconds, keeping the service interruption time within the SLA.

[0092] The post-event analysis module collects complete execution logs and compares the discrepancies between the planned and actual results. The voltage fluctuation event analysis report reveals that the storage node switchover delay was primarily due to firmware version incompatibility. The system automatically updates this information to the device knowledge base. The plan library dynamically adjusts parameters based on the execution results, optimizing the backup circuit switchover threshold for similar events from 10 seconds to 8 seconds. Operational records during execution are transformed into new case examples, enriching the system's response experience.

[0093] The system implementation utilizes a microservices architecture to ensure scalability. The plan management service is independently deployed, supporting concurrent editing of the plan library by multiple users. The resource scheduling service utilizes a distributed lock mechanism to coordinate concurrent requests. The command conversion service features a built-in protocol adaptation layer, enabling integration with device control systems from different manufacturers. The execution engine service implements a transaction compensation model to ensure atomicity. All services communicate via an event bus, with status changes pushed to the monitoring center in real time.

[0094] The operation and maintenance interface provides a comprehensive view of emergency plan management. A risk map displays the activation status of emergency plans in each area, using color-coded risk levels. The emergency plan editor supports graphically orchestrating operational processes, allowing users to drag and drop device icons to construct response plans. The execution monitoring screen displays command completion progress in real time, highlighting abnormal operations. The analysis dashboard summarizes historical emergency plan execution results, displaying trend indicators such as success rate and average execution time.

[0095] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0096] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data center digital twin simulation and decision-making system for intelligent management, characterized by: include: Data center physical characteristics perception module, used to collect multi-dimensional operating parameters of data center infrastructure in real time; A virtual space reconstruction module, configured to dynamically construct a three-dimensional virtual model of the data center based on the multi-dimensional operating parameters acquired by the data center physical feature perception module; An operating situation deduction engine is used to simulate the resource scheduling and energy flow process within the data center based on the three-dimensional virtual model generated by the virtual space reconstruction module; An abnormal behavior identification module is used to analyze the simulated data stream output by the operation situation deduction engine in real time and detect abnormal operation modes that deviate from a preset threshold; The decision instruction generation module is used to integrate the abnormal operation mode identified by the abnormal behavior identification module and generate optimization control instructions for physical equipment.

2. The data center digital twin simulation and decision-making system for intelligent management according to claim 1, characterized in that: The data center physical feature perception module includes: Environmental parameter acquisition unit, which obtains the temperature distribution, humidity distribution and air flow velocity distribution of each area of ​​the data center in real time; Equipment status monitoring unit, which continuously tracks the power consumption data of the server cluster, calculates the load rate and the operating status of the cooling equipment; The energy consumption feature extraction unit performs feature fusion on the original data acquired by the environmental parameter acquisition unit and the equipment status monitoring unit, and calculates the regional energy efficiency feature vector.

3. The data center digital twin simulation and decision-making system for intelligent management according to claim 2, characterized in that: The virtual space reconstruction module includes: The spatial topology analyzer identifies the spatial relationship between cold aisles, hot aisles, cabinet clusters, and cooling units based on the data center's architectural drawings and equipment layout information. A dynamic attribute mapper that associates and binds the regional energy efficiency feature vector calculated by the energy consumption feature extraction unit to the corresponding spatial position identified by the spatial topology parser; A multi-scale model generator generates a hierarchical three-dimensional virtual model including thermodynamic properties based on the spatial position relationship output by the spatial topology analyzer and the energy efficiency characteristics bound by the dynamic property mapper.

4. The data center digital twin simulation and decision-making system for intelligent management according to claim 3, characterized in that: The operation situation deduction engine includes: A thermal field dynamic tracking engine simulates the spatiotemporal evolution of airflow organization and temperature field based on the hierarchical three-dimensional virtual model output by the multi-scale model generator; Load migration simulator, simulating the dynamic scheduling path and resource usage changes of computing tasks between different server nodes; The energy consumption coupling analyzer correlates the thermal field spatial distribution output by the thermal field dynamic tracking engine with the resource occupancy change output by the load migration simulator to calculate a global energy consumption hotspot map.

5. The data center digital twin simulation and decision-making system for intelligent management according to claim 4, characterized in that: The abnormal behavior identification module includes: A threshold deviation detection unit compares the global energy consumption hotspot map generated by the energy consumption coupling analyzer with a preset energy efficiency benchmark interval in real time; a pattern drift analysis unit for identifying the aggregation characteristics of the deviation data found by the threshold deviation detection unit in the time dimension and the diffusion trend in the space dimension; The anomaly root cause locator associates the aggregation characteristics and diffusion trends identified by the pattern drift analysis unit, marks the anomaly source physical component and associates the affected link.

6. The data center digital twin simulation and decision-making system for intelligent management according to claim 5, characterized in that: The decision instruction generation module includes: The policy knowledge base retriever matches the optimization policy template in the historical handling case library based on the abnormal source physical components and associated impact links marked by the abnormal root cause locator; A multi-objective optimization calculator calculates adjustment amounts of control parameters based on the optimization strategy template retrieved by the strategy knowledge base retriever and in combination with the hierarchical three-dimensional virtual model state in the current virtual space reconstruction module; An instruction sequence assembler converts the control parameter adjustment amount output by the multi-objective optimization calculator into an executable control instruction sequence.

7. The data center digital twin simulation and decision-making system for intelligent management according to claim 6, characterized in that: The system further comprises: a fault propagation deduction module, configured to simulate a transmission path of a single point fault in the three-dimensional virtual model of the virtual space reconstruction module based on the associated impact links marked by the abnormal root cause locator; a risk level assessor, configured to calculate a system-level risk entropy value based on the depth and breadth of the conduction path simulated by the fault propagation deduction module; The plan matching engine is used to activate the disaster recovery plan of the corresponding level according to the risk entropy value calculated by the risk level assessor.

8. The data center digital twin simulation and decision-making system for intelligent management according to claim 7, characterized in that: The fault propagation deduction module includes: A dependency mining unit analyzes the physical connections and logical dependencies between the cabinet cluster, the cooling unit, and the power supply circuit in the virtual space reconstruction module; A failure impact simulator injects the abnormal source fault signal marked by the abnormal root cause locator into the three-dimensional virtual model constructed by the multi-scale model generator to track the propagation trajectory of the state abnormality; The critical path identifier marks the device node set with the highest fault conduction relationship strength according to the abnormal state propagation trajectory output by the failure impact simulator.

9. The data center digital twin simulation and decision-making system for intelligent management according to claim 8, characterized in that: The risk level assessor comprises: An impact range measurement unit calculates the spatial coverage ratio of the set of device nodes marked by the critical path identifier in the three-dimensional virtual model; a service interruption assessment unit, associated with the resource scheduling path of the load migration simulator, and quantifying the delay increment and interruption probability of the affected computing task; The risk entropy value aggregator weightedly aggregates the spatial coverage ratio calculated by the impact range measurement unit and the delay increment and interruption probability quantified by the service interruption assessment unit, and outputs a comprehensive risk level label.

10. The data center digital twin simulation and decision-making system for intelligent management according to claim 9, characterized in that: The plan matching engine includes: A level response mapper, which indexes a pre-configured emergency response level directory according to the comprehensive risk level label output by the risk entropy value fusion device; A resource scheduling preconfigurator generates a backup refrigeration equipment startup priority and a computing task migration target matrix based on the emergency response level determined by the level response mapper; The multi-source decision instruction integrator fuses the executable control instruction sequence of the decision instruction generation module with the migration target matrix generated by the resource scheduling preconfigurator to generate a final execution instruction package.

Images