Intelligent machine room multi-dimensional dynamic monitoring management method and system based on digital twinning

Abstract

The application provides a multi-dimensional dynamic monitoring and management method and system for intelligent machine rooms based on digital twinning, relates to the technical field of machine room monitoring, and comprises the following steps: collecting a three-dimensional point cloud set and an RGB image set of a machine room space, and establishing a space mapping table; mapping access control event data to a corresponding position in a digital twinning space to generate an access control event space positioning set, identifying access control abnormal events, and constructing an access control abnormal event space positioning set; performing abnormal behavior identification to construct an abnormal event graph structure; based on the abnormal event graph structure, reconstructing and visualizing abnormal trajectories; combining the access control abnormal event space positioning set, abnormal behavior category prediction probability distribution, and trajectory risk level to execute a hierarchical early warning mechanism and generate an access control abnormal event log; and the application realizes intelligent positioning, trajectory visualization, and risk hierarchical early warning of access control abnormal events in intelligent machine rooms, and is suitable for digital operation and maintenance management systems such as intelligent machine rooms.

Description

Technical Field

[0001] This invention relates to the field of data center monitoring technology, specifically to a method and system for multi-dimensional dynamic monitoring and management of smart data centers based on digital twins. Background Technology

[0002] With the rapid development of digital twin technology and intelligent sensing systems, smart data centers, as a new generation of information infrastructure, have become crucial for ensuring the continuity of critical business operations and efficient operation and maintenance management through their security visualization monitoring capabilities. How to achieve spatial localization, semantic understanding, and intelligent response to key anomalies such as personnel behavior and access control events within the data center based on dynamic sensing data from the real physical space has become a core issue that urgently needs to be addressed in the construction of smart data centers.

[0003] Chinese patent application CN118466258B, authorized by CN118466258B, discloses a smart data center (BA) automation management system based on big data analysis. The system includes: a real-time monitoring module, which is communicatively connected to a controller and a power-on management module, which is communicatively connected to a power-on control module, which is communicatively connected to the controller; the real-time monitoring module is used to perform real-time monitoring and analysis of the environment within the smart data center; the power-on management module is used to manage and analyze the power-on environment and energy consumption of the smart data center; and the power-on control module is used to perform environmental regulation and analysis during the power-on state of the smart data center.

[0004] However, existing technologies still face numerous challenges in the security management of smart data centers. Most current smart data center access control systems have relatively limited functionality, only recording basic log information such as card swipes and door openings. They lack the ability to integrate with systems like video surveillance and personnel positioning. This information silo phenomenon makes it impossible to accurately reconstruct the course of events and personnel movement in the event of a critical security incident. Especially in cases of unauthorized personnel trespassing, abnormal card swipes late at night, or continuous access control failures, the system cannot use 3D visualization to instantly trace behavioral trajectories and identify risks, easily leading to missed or misjudged alarms. Furthermore, because the incident alarm and maintenance handling processes are not effectively integrated, the system often only issues alarms and cannot further link and record the maintenance response process and results. This results in critical incident response and tracing relying on manual operation, which is inefficient, error-prone, and restricts the overall closed-loop security response capability of the data center. If unauthorized entry is not detected and handled in a timely manner, unauthorized operators may enter core areas, potentially damaging servers, power supply systems, or communication equipment, leading to data leaks, equipment downtime, or even widespread business interruptions, affecting network security. Summary of the Invention

[0005] To overcome the aforementioned shortcomings of existing technologies, this invention provides a multi-dimensional dynamic monitoring and management method and system for smart data centers based on digital twins. This method constructs a digital twin spatial model of the smart data center, enabling accurate spatial positioning of access control events within the digital twin space. It breaks through the limitations of traditional planar event recording methods, generates a device space mapping table, and makes spatial information expression more intuitive. This effectively solves the challenges of multi-source data fusion and access control anomaly event identification in complex smart data center environments. It assists maintenance personnel in quickly identifying key areas and critical time periods, accurately conducting spatiotemporal tracing and security control of access control anomalies, and ensuring closed-loop management of the entire risk control process.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A multi-dimensional dynamic monitoring and management method for smart data centers based on digital twins includes:

[0008] Acquisition of 3D point cloud set of the data acquisition room space and RGB image collection A digital twin space model representing the physical state of the computer room and supporting monitoring and control is constructed, and a spatial mapping table between the digital twin space and the physical space is established. ;

[0009] Based on spatial mapping table This method maps personnel entry and exit event data generated by the access control system to corresponding spatial locations in a digital twin space, continuously monitors personnel behavior in real time to identify abnormal events, and constructs a spatial location set of abnormal access control events. ;

[0010] Spatial location set of access control abnormal events Perform abnormal behavior identification and correlation analysis, and construct an abnormal event graph structure. ;

[0011] Based on anomaly event graph structure Dynamic reconstruction of abnormal trajectories is performed, and visual control decision support is achieved through the three-dimensional scene of the control interface;

[0012] Combined with spatial location set of access control abnormal events Probability distribution of abnormal behavior categories and trajectory risk level The system determines the corresponding control strategy and issues linkage control commands accordingly, constructs a hierarchical early warning and response control mechanism for access control anomalies, and automatically generates access control anomaly event logs to achieve closed-loop feedback control of the monitoring system; where k is the trajectory index and N' is the total number of trajectories.

[0013] Furthermore, the spatial mapping table The methods for establishing it include:

[0014] Based on 3D point cloud set and RGB image collection Building multimodal fusion data ;

[0015] Based on multimodal fusion data set By combining Building Information Modeling (BIM), a spatial index structure is established. ;

[0016] Based on device set Spatial models with semantic labels Spatial index structure Spatial mapping is performed, a time-series index update mechanism is introduced, and a spatial mapping table is established. ;

[0017] The set of devices Automatically obtained through installation records, RFID tags, QR codes, or cameras;

[0018] The time-series index update mechanism uses a KD-Tree to dynamically maintain the spatial index structure.

[0019] Furthermore, the spatial index structure The methods for establishing it include:

[0020] Based on multimodal fusion data set A voxel meshing algorithm is used to perform structured transformation on the 3D point cloud data to obtain a 3D point cloud voxel set. ;

[0021] Combining structural component information provided by Building Information Modeling (BIM) with a set of 3D point cloud voxels Spatial alignment is performed, and a multimodal semantic segmentation process is executed to obtain a spatial model with semantic labels. ;

[0022] Spatial Model Based on Semantic Labels Constructing a spatial index structure .

[0023] Furthermore, the spatial location set of access control abnormal events The construction methods include:

[0024] Get the original access control event set And based on the original access control event set Building a high-quality access control event set ;

[0025] Based on high-quality access control event set and space mapping table Obtain the spatial location set of access control events. ;

[0026] Based on the spatial localization set of access control events Construct a spatial location set for access control anomaly events .

[0027] Furthermore, the abnormal event graph structure The construction methods include:

[0028] Spatial location set of access control abnormal events Obtain the feature representation of access control abnormal events and video behavior segment feature representation Constructing multimodal features ;

[0029] Based on multimodal features Temporal convolutional networks are introduced to construct the probability distribution for predicting abnormal behavior categories. ;

[0030] Predicting probability distribution based on abnormal behavior categories With historical operation and maintenance event data set Construct an exception event graph structure .

[0031] Furthermore, the abnormal event graph structure The construction methods also include:

[0032] Define the initial exception event graph , where the set of nodes ; Indicates the first Each node corresponds to a historical or current abnormal event. The total number of nodes; ≤ ;

[0033] Based on each event node Construct the initial feature matrix ;in, For one The set of real numbers in rows and columns d; R represents real numbers;

[0034] Based on historical operation and maintenance event data sets Define the initial adjacency matrix based on event correlation. ;

[0035] The initial adjacency matrix elements Indicates an event Determining whether an edge exists;

[0036] A dynamic graph update mechanism is introduced to adapt to new events in real time and dynamically update the adjacency matrix. ;

[0037] Graph Convolutional Networks (GCNs) are used for inter-event information propagation and embedding learning to obtain the first... High-order semantic feature embedding of layers;

[0038] pass After the layer graph convolution operation, extract the first layer. High-order semantic embedding vectors Combined with node set With the final dynamic adjacency matrix Construct a structured exception event graph. ;

[0039] Furthermore, the steps of dynamic trajectory reconstruction and 3D scene visualization of the abnormal trajectory include:

[0040] Based on anomaly event graph structure Spatial positioning set of access control events Reconstruct the movement trajectory of abnormal personnel and output the trajectory point sequence. ;

[0041] Based on trajectory point sequence A trajectory prediction algorithm based on the fusion of Kalman filter and Long Short-Term Memory (LSTM) network is used to obtain the predicted trajectory curve. ;

[0042] Based on predicted trajectory curve Spatial location set of access control abnormal events and probability distribution of abnormal behavior categories It enables visualization of 3D trajectory mapping and display of abnormal brightness.

[0043] Furthermore, the steps for visualizing the three-dimensional trajectory mapping and displaying abnormal brightness include:

[0044] Predict the trajectory curve Each trajectory point Mapping to the coordinate system of the digital twin 3D simulation scene, constructing the corresponding 3D spatial trajectory line. ;in, Indicates the first Each trajectory point records the corresponding timestamp. Indicates the first Each trajectory point records the corresponding timestamp. The corresponding spatial location based on the fusion of Kalman filter and Long Short-Term Memory (LSTM) network;

[0045] All According to timestamp The sequential connections form a dynamic trajectory line, which is then played in a time-driven manner on a 3D visualization platform.

[0046] Spatial location set of access control abnormal events and probability distribution of abnormal behavior categories For each predicted trajectory curve Each trajectory point Rendering colors Make a judgment.

[0047] The determination of the rendering color for each trajectory point of each predicted trajectory curve includes:

[0048] Set probability distribution thresholds, including upper threshold, middle threshold and lower threshold;

[0049] The risk level is... For high-risk warning, ; Predict the probability distribution of the abnormal behavior category for the k-th trajectory point;

[0050] ,but This is a medium-risk warning. ;

[0051] ,but Low risk warning. ;

[0052] ,but This is normal. .

[0053] Furthermore, when High-risk warning, The system immediately triggered a Level 1 high-risk warning event; when Medium risk warning The system immediately triggered a Level 2 medium-risk warning event; when Low risk warning or normal. The system does not trigger alarms proactively.

[0054] Furthermore, the access control anomaly event log includes event number, event time, spatial location, anomaly category and risk level, and anomaly warning record.

[0055] A digital twin-based intelligent data center multi-dimensional dynamic monitoring and management system is used to implement the aforementioned digital twin-based intelligent data center multi-dimensional dynamic monitoring and management method. The system includes:

[0056] Digital twin spatial modeling module: used to collect 3D point cloud sets of the computer room space. With RGB image collection A digital twin space representing the physical state of the computer room and supporting monitoring and control is constructed, and a spatial mapping table between the digital twin space and the physical space is established. ;

[0057] Access control anomaly event localization module: based on spatial mapping table This method maps personnel entry and exit event data generated by the access control system to corresponding spatial locations in a digital twin space, continuously monitors personnel behavior in real time to identify abnormal events, and constructs a spatial location set of abnormal access control events. ;

[0058] Abnormal Behavior Recognition and Abnormal Event Graph Construction Module: Based on Spatial Localization Set of Access Control Abnormal Events Perform abnormal behavior identification and correlation analysis, and construct an abnormal event graph structure. .

[0059] Abnormal trajectory reconstruction and 3D visualization control module: Based on the abnormal event graph structure, it performs dynamic reconstruction of abnormal trajectories and realizes visualization control decision support through the 3D scene of the control interface.

[0060] Linked early warning control and log generation module: It is used to combine the spatial location set of access control abnormal events, the predicted probability distribution of abnormal behavior categories and trajectory risk level to determine the hierarchical early warning linkage response control mechanism and issue closed-loop linkage control commands to the controlled equipment in the target area accordingly. It also collects the execution status of the controlled equipment in real time as status feedback and generates a closed-loop access control abnormal event log containing equipment execution feedback and response status.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] This invention constructs a digital twin spatial model with spatial semantics by fusing 3D point cloud and RGB image data, and establishes a dynamically updatable spatial index structure to achieve accurate mapping and real-time updates of access control anomalies in 3D space. This avoids the problems of unclear device positioning and unclear event response in traditional 2D records, and improves the spatial visualization guidance capability and rapid response efficiency of anomalies.

[0063] This invention constructs a high-precision mapping mechanism between access control anomaly events and digital twin space, and introduces a Bayesian filtering model and an abnormal behavior recognition model to perform semantic parsing and spatial reasoning on the original access control data. This enables the identification and location of access control anomaly events such as multiple consecutive card swipe failures and access during off-duty periods in three-dimensional space, avoiding the problems of abstract event description and ambiguous location in traditional access control systems, and improving the spatial visibility and response efficiency of maintenance personnel when dealing with sudden anomalies.

[0064] This invention constructs a multimodal alignment mechanism between access control anomaly events and video behavior information, integrates contextual semantics and video spatiotemporal features, and breaks down spatiotemporal information gaps between multimodal data sources. It introduces graph convolution and dynamic graph update strategies to establish spatiotemporal semantic associations between abnormal behaviors and historical events, forming a dynamically evolving anomaly event graph. This avoids the problems of isolated behavior recognition and evolutionary relationship defects in traditional systems, and provides an accurate graph model foundation for subsequent event tracing, multi-source early warning linkage, and behavior reasoning.

[0065] This invention addresses the issue of missing trajectories in access control records by constructing a spatial mapping relationship between an anomaly event graph structure and access control events, and by introducing a fusion of Kalman filtering and LSTM networks to dynamically repair and predict trajectory data. Through joint visualization of predicted trajectories, access control anomaly location points, and the predicted probabilities and risk levels of anomaly behavior categories, it achieves dynamic reconstruction and risk classification of anomaly individuals in a digital twin 3D space, providing maintenance personnel with intuitive, continuous, and traceable analytical data.

[0066] This invention combines access control anomaly location results, behavior category prediction probability, and trajectory risk level to determine the risk level of each trajectory point and automatically trigger the corresponding level of early warning response. At the same time, it generates anomaly event logs in a structured manner, avoiding problems such as delayed early warning triggering, ambiguous risk levels, and lack of traceability of event records in traditional systems, thereby realizing hierarchical alarms, full-process recording, and precise spatial positioning of anomaly events in smart data centers. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a flowchart illustrating the principle of the intelligent data center multi-dimensional dynamic monitoring and management method based on digital twins of the present invention.

[0069] Figure 2This is a flowchart of the method for constructing a spatial mapping table according to the present invention;

[0070] Figure 3 This is a flowchart of the method for constructing an exception event graph structure according to the present invention;

[0071] Figure 4 This is a flowchart of the method for constructing an exception event graph structure according to the present invention;

[0072] Figure 5 This is a flowchart of the method for trajectory reconstruction and visualization of abnormal trajectories according to the present invention;

[0073] Figure 6 This is a flowchart of the method for generating access control abnormal event logs according to the present invention;

[0074] Figure 7 This is an example diagram of the access control abnormal event log of the present invention;

[0075] Figure 8 This is a functional module diagram of the intelligent data center multi-dimensional dynamic monitoring and management system based on digital twins, as presented in this invention. Detailed Implementation

[0076] 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.

[0077] Example 1

[0078] Please see Figure 1 As shown, this embodiment provides a multi-dimensional dynamic monitoring and management method for smart data centers based on digital twins, including:

[0079] Step S1000: Collect a 3D point cloud set of the computer room space. and RGB image collection A digital twin space model representing the physical state of the computer room and supporting monitoring and control is constructed, and a spatial mapping table between the digital twin space and the physical space is established. .

[0080] Furthermore, such as Figure 2 As shown, step S1000 includes:

[0081] Step S1100: Obtain the 3D point cloud set of the computer room space. and RGB image collection And based on a 3D point cloud set and RGB image collection Building multimodal fusion data .

[0082] Specifically, LiDAR sensors are deployed in the computer room to scan and collect high-density 3D point cloud data in the space, recording the precise 3D coordinates of each sampling point, resulting in a set of 3D point clouds of the computer room space. Represented as:

[0083] ;

[0084] in, Indicates the first Each point cloud sampling point contains three-dimensional coordinate information; Indicates the first Point cloud sampling points Spatial three-dimensional coordinate information, It is the horizontal coordinate. It is the vertical coordinate. These are depth coordinates; This represents the total number of sampling points in the three-dimensional point cloud set of the computer room space.

[0085] To improve the quality of point cloud data, a dynamic threshold function is set. The original 3D point cloud set of the computer room space Preprocessing is performed to filter out point cloud data that may contain anomalies. Specifically, this includes:

[0086] First, calculate each point cloud sampling point. neighborhood center mean point The specific process formula is as follows:

[0087] ;

[0088] in, Indicates the first Point cloud sampling points neighborhood center mean point Spatial three-dimensional coordinate information, yes The horizontal coordinates, It is the vertical coordinate. These are depth coordinates.

[0089] Subsequently, based on Euclidean distance and dynamic threshold function Determine point cloud sampling points The specific formula for determining whether a point is an outlier is as follows:

[0090] ;

[0091] in, Represents the Euclidean distance, used to calculate point cloud sampling points. Its neighborhood center mean point The Euclidean distance between them.

[0092] Finally, the retained valid point cloud sampling points are used to construct a filtered 3D point cloud set for the computer room space. The specific formula is as follows:

[0093] ;

[0094] This represents the filtered set of 3D point clouds of the computer room space, used to retain a new set of valid point cloud sampling points; This represents a dynamic threshold function used to determine the sampling points in a point cloud. Is it abnormal? Indicates the current time frame; This represents the estimated environmental noise level, used to dynamically adjust the threshold value.

[0095] Simultaneously, multiple fixed-view cameras are deployed in the server room, employing a unified time-synchronized acquisition mechanism to ensure the synchronous acquisition of RGB image frames from various perspectives. This enriches the texture and color feature information within the 3D geometric data, resulting in a collection of RGB images of the server room space. Represented as:

[0096] ;

[0097] in, Indicates the first Image frames captured by a camera; This indicates the total number of cameras.

[0098] To ensure the consistency of multi-source data, a time synchronization mechanism is used during data acquisition to ensure that the data collected by the lidar and camera are consistent in time frames, providing a reliable foundation for subsequent multimodal fusion.

[0099] Finally, based on the intrinsic and extrinsic parameters of the LiDAR and camera obtained from sensor calibration, a projection mapping relationship between 3D point cloud coordinates and image pixel coordinates is established. Specifically, each filtered point cloud sampling point... Project onto the corresponding RGB image plane and obtain the color value corresponding to its pixel coordinates. This generates a multimodal fusion data set of the computer room space. The specific formula is as follows:

[0100] ;

[0101] in, This represents the color value of that pixel in the image. (i.e., RGB three-channel values) The value for the red channel. This is the green channel value; This is the value for the blue channel.

[0102] Step S1200, based on the multimodal fusion data set By combining Building Information Modeling (BIM), a spatial index structure is established. .

[0103] Specifically, this step aims to integrate the multimodal fusion data set from step S1100. It is converted into a structured three-dimensional voxel mesh representation and integrated with semantic information from Building Information Modeling (BIM) to realize a three-dimensional digital model with spatial structure and semantic attributes, providing a foundation for subsequent equipment mapping and spatial analysis.

[0104] First, based on multimodal fusion datasets A voxel grid filtering algorithm is used to perform structured transformation on 3D point cloud data, realizing the discretization of continuous space. The specific process formula is as follows:

[0105] ;

[0106] in, Represents a set of three-dimensional point cloud voxels. The center coordinates of each voxel are used to discretize the point cloud data onto a regular grid. Indicates the index of a voxel unit. The total number of voxel units; This represents a voxelization algorithm used to convert point clouds into voxel representations; The voxel size is represented by meters (m) and is used to divide the spatial grid cells, controlling the spatial discretization accuracy.

[0107] Next, the structural component information (such as walls, cabinets, doors, etc.) provided in the BIM is combined with the 3D point cloud voxel set. Spatial alignment is performed, and multimodal semantic segmentation is carried out. The specific process formula is as follows:

[0108] ;

[0109] in, This represents the semantically labeled spatial model of the output, used for each voxel unit. Assigning semantic label categories , voxel unit The corresponding semantic tag category, For a predefined set of semantic categories (such as walls, ceilings, cabinets, pipes, etc.); This represents a semantic segmentation model used to combine voxelized spatial data with BIM information and output semantic labeling results. It represents a building information model, which includes prior knowledge such as spatial structure, component type, and semantic labels, and provides additional semantic and structural information to assist or guide semantic segmentation.

[0110] Finally, based on the spatial model with semantic labels Constructing a spatial index structure The specific process formula is as follows:

[0111] ;

[0112] in, Represented as a spatial index structure, it is used for efficient querying, locating, and matching of the structure of semantic voxel units in the space; This indicates the function to build a K-Dimensional Tree, which is used to construct a tree-like index structure from voxel center points to support fast searching.

[0113] For example, taking a smart data center as an example, a spatial model with semantic tags is constructed. , , ,but will with , , For each node, construct a 3D KD-Tree for easy and fast searching.

[0114] Step S1300, based on the device set Spatial models with semantic labels Spatial index structure Spatial mapping is performed, a time-series index update mechanism is introduced, and a spatial mapping table is established. .

[0115] Specifically, this step aims to map the set of device entities to the generated spatial model with semantic labels, thereby enabling precise spatial positioning and management of devices in the digital twin system.

[0116] Equipment Collection Acquired automatically by installation records, RFID, QR codes, or cameras; each It refers to a physical device (such as a server, switch, UPS power supply, air conditioner, access control system, monitoring equipment, and other intelligent terminal devices). This represents the total number of physical devices. Mapping to a spatial model with semantic labels Establish the correspondence between equipment and spatial location to obtain the equipment space mapping table. .

[0117] To support dynamic updates of the device set (such as adding new devices, relocating devices, removing devices, etc.), a time-series index update mechanism is introduced, using a KD-Tree to dynamically maintain the spatial index structure. The specific expression formula is as follows:

[0118] ;

[0119] in, Indicates the current time step Spatial index structure; This represents a function for dynamically updating the index, used to efficiently insert or delete nodes in an existing index structure while maintaining spatial search efficiency. This represents the spatial index structure of the previous time step; This represents the set of device changes that are added, moved, or deleted at the current moment.

[0120] Step S2000, based on the spatial mapping table This method maps personnel entry and exit event data generated by the access control system to corresponding spatial locations in a digital twin space, continuously monitors personnel behavior in real time to identify abnormal access control events, and constructs a spatial location set for abnormal access control events. .

[0121] This step aims to collect personnel entry and exit event data through the access control system, and combine it with spatial mapping to achieve spatial positioning of access control events and detection of abnormal behavior, providing basic data support for subsequent personnel trajectory analysis and security management.

[0122] Furthermore, such as Figure 3 As shown, step S2000 includes:

[0123] Step S2100: Obtain the original access control event set. And based on the original access control event set Building a high-quality access control event set .

[0124] Specifically, this step aims to collect raw access control event data from the access control device. The data was cleaned to ensure quality. Each event record included a timestamp, personnel ID, access control point ID, and card swipe status. The collected data was then deduplicated and filtered for outliers to remove invalid data, resulting in a high-quality set of access control events for subsequent analysis. .

[0125] First, raw access control event data is collected in real time through access control data acquisition devices (such as IC card readers, fingerprint scanners, facial recognition terminals, etc.) deployed in the computer room. The specific formula is expressed as follows:

[0126] ;

[0127] in, Represents the original set of access control events; Indicates a single access control event. Indicates the first The time of each access control event is used to record the specific time the event occurred. This serves as a unique identifier for each employee, representing the employee number corresponding to a card swipe or fingerprint scan. This is a unique identifier (ID) for the access control point, indicating the specific location of the access control device where the incident occurred. Indicates the card swipe status (success or failure); This represents the total number of original access control events.

[0128] Next, based on the original access control event set The system performs deduplication and outlier filtering to remove invalid data and obtain a high-quality set of access control events. The specific steps are as follows:

[0129] First, deduplication. For the same person and the same access control point, at time intervals... Multiple card swipes within the same period are considered duplicate events; the earliest event is retained. If an event pair exists... satisfy: , , , If the event is repeated, then the earliest event is selected. Ultimately, the set of events after removing duplicates is represented as: .

[0130] Secondly, outlier filtering. For the deduplicated data, outlier filtering is performed to remove invalid data. If... For exception removal, invalid user IDs will be removed. Represents the set of legitimate user IDs; if Invalid access control point IDs will be removed. Represents the set of valid access control point IDs; if For exception removal, timestamps that are abnormal will be removed. Indicates a reasonable range of events.

[0131] Finally, after the above two steps, a high-quality access control event set is obtained. The specific formula is as follows:

[0132] ;

[0133] Step S2200, based on a high-quality access control event set and space mapping table Obtain the spatial location set of access control events. .

[0134] Specifically, this step aims to locate each access control event to its physical spatial coordinates by mapping access control point IDs to spatial coordinates. Simultaneously, time series modeling and Bayesian filtering methods are introduced to infer the possible true spatial location of personnel.

[0135] First, for the current number Access control incident Its corresponding spatial coordinates The specific formula is as follows:

[0136] ;

[0137] in, This represents a mapping function used to uniquely identify access control points. Mapped to coordinates in space;

[0138] Next, to improve positioning accuracy, a positioning model based on Bayesian filtering is constructed by further combining the time series characteristics and spatial transfer patterns of access control events. The specific process formula is as follows:

[0139] ;

[0140] in, Indicates the first The employee's location corresponding to this access control incident; Indicates that the employee is in The spatial location at the time of the event is used to describe the employee's movement path; Indicates from item 1 to item 2. A time series set of access control events; Indicates from item 1 to item 2. A time series set of access control events is used to predict the current location; This represents the posterior probability, used for the current inference target (the probability distribution of the employee's spatial state). Indicates direct proportion; Indicates the observed likelihood, used for events With position The degree of matching, such as whether the access control point belongs to this location; This represents the state transition probability, used to describe the employee movement model (such as floor corridor structure, passageway accessibility). This represents the prior probability, used to infer the distribution of people at the previous position from historical data; Represents all possible Perform summation;

[0141] Ultimately, the spatial location set of access control events is obtained. .

[0142] Step S2300, based on the access control event spatial location set Construct a spatial location set for access control anomaly events .

[0143] This step is based on the spatial localization set of access control events. This study comprehensively employs a rule base and a learning model to detect access control anomalies, identify potential violations or abnormal access control behaviors, and construct a spatial localization set of access control anomaly events. The specific basic rules are as follows:

[0144] First, multiple card swipes failed. If the employee... In a continuous time period Memory exists If a single card swipe is recorded as a potential risk of tailgating, impersonation, or identity theft, it will be deemed abnormal.

[0145] ;

[0146] in, Indicates the number of events, used to represent the number of consecutive events; This indicates the threshold for the number of failed card swipes; exceeding this number will be considered an error. The existence symbol indicates the existence of a subset of events that satisfy a certain condition; Indicates the first To the A continuous set of access control events; Indicates the first The card swipe status of each access control event, such as "success" or "failure"; Indicates the first The card swipe status for this access control event is "failed"; Indicates the first The user ID of the access control event and the first If two access control events have the same user ID, it means that the user is the same. Indicates the length of the time window, used to limit the continuous time range in which card swipe failure events occur; This indicates that the time interval between consecutive events does not exceed the set time window.

[0147] Secondly, entry outside of designated duty hours. If employees... Attempting to enter a restricted area outside of authorized hours is considered a risky act such as nighttime boundary trespassing, unauthorized overtime work, or unauthorized entry. Guard duty periods should be set. ,but:

[0148] ;

[0149] in, This indicates the legal time period for guard duty, i.e., the time range during which passage is permitted; Indicates the start time of the duty period; Indicates the end time of the duty shift; Indicates the first The timestamp of the access control event (i.e., the time the card was swiped); It indicates "if and only if", representing a logical equivalence relation; It means "does not belong to".

[0150] Third, illegal passage through areas. If employees Permission areas It can be obtained from the permission mapping table, if If so, it is considered abnormal:

[0151] ;

[0152] in, Indicates employees The set of authorized areas (e.g., the floor of their workstation, the rooms they are allowed to enter, etc.).

[0153] Finally, through filtering using the aforementioned multiple anomaly rules, a spatial location set of access control anomaly events is obtained. The specific formula is as follows:

[0154] ;

[0155] Step S3000: Based on the spatial location set of access control abnormal events Perform abnormal behavior identification and correlation analysis, and construct an abnormal event graph structure. .

[0156] This step aims to collect personnel entry and exit event data through the access control system, and combine it with spatial mapping to achieve spatial positioning of access control events and detection of abnormal behavior, providing basic data support for subsequent personnel trajectory analysis and security management.

[0157] Furthermore, such as Figure 4 As shown, step S3000 includes:

[0158] Step S3100, based on the spatial location set of access control abnormal events Obtain the feature representation of access control abnormal events and video behavior segment feature representation Constructing multimodal features .

[0159] Specifically, this step aims to construct a multi-modal feature vector fused from multiple sources to support subsequent abnormal behavior identification.

[0160] First, based on the spatial location set of access control abnormal events constructed in step S2300. A bidirectional long short-term memory (BiLSTM) network is used to extract the contextual semantic features of access control anomaly events, resulting in a feature representation of access control anomaly events. .

[0161] Next, based on the spatial location set of access control abnormal events constructed in step S2300... Locating abnormal segments in video surveillance streams The specific process formula is as follows:

[0162] ;

[0163] in, This indicates that a function is called to extract a time segment containing access control anomaly events from the complete video surveillance stream; This indicates the time period of the access control anomaly event, starting from the time the anomaly event begins. Until the end time of access control abnormal events ; Indicates the first Access control point ID.

[0164] For example, if The abnormal access control event occurred between 08:01 and 08:06 on March 26, 2025, with the abnormal access point ID being "West Gate of Building 1". Therefore, the abnormal segment from 08:01 to 08:06 on March 26, 2025, is extracted from the video stream of the camera corresponding to the West Gate of Building 1. .

[0165] Based on abnormal segments in video surveillance streams An Inflated 3D ConvNet model is used to model the actions in the video, resulting in feature representations of video action segments. .

[0166] Finally, based on the feature representation of access control anomaly events and video behavior segment feature representation Multi-source data, constructing multi-modal features The specific process formula is as follows:

[0167] ;

[0168] in, Indicates the process The unified feature vector of the access control mode obtained after mapping; The linear mapping matrix representing the access control mode will Transform it into a unified-dimensional embedded representation; Indicates the process The unified feature vector of the video modality obtained after mapping; The linear mapping matrix representing the video modality will Transform it into a unified-dimensional embedded representation.

[0169] ;

[0170] in, Indicates the first Attention weights for each modality, satisfying ; Indicates the first A unified feature representation of each modality; Indicates the first A unified feature representation of each modality; This represents a trainable modality weight query vector, used to measure the modality weights of the first modality. The importance of each modal feature determines the attention score; express transpose, This represents an exponential function used to map scores to non-negative weights, which helps to highlight significant modal features.

[0171] Based on attention weights To obtain multimodal features The specific process formula is as follows:

[0172] ;

[0173] Step S3200, based on multimodal features Temporal convolutional networks are introduced to construct the probability distribution for predicting abnormal behavior categories. .

[0174] Specifically, this step aims to build upon the multimodal features constructed in step S3100. The system utilizes a deep neural network classifier to determine the specific type of abnormal behavior, including but not limited to accidental entry, tailgating, and malicious card swiping. By introducing a Temporal Convolutional Network (TCN), it can accurately identify the types of abnormal behaviors that may occur in access control scenarios.

[0175] First, based on multimodal features Temporal convolutional network (TCN) is introduced to extract temporal features. The specific process formula is as follows:

[0176] ;

[0177] in, This represents the Temporal Convolutional Network (TCN) function, used to capture temporal information such as the occurrence pattern, duration, and dependencies of anomalous behavior.

[0178] Next, based on time series characteristics Construct the probability distribution for predicting abnormal behavior categories The specific process formula is as follows:

[0179] ;

[0180] in, Indicates the total number of categories of abnormal behavior; This represents the activation function, which is used to transform the linear output into a probability distribution so that the sum of the predicted probabilities of all categories is 1, facilitating classification decisions. This represents the classification weight matrix, used to classify time-series features. Mapped to the classification space; This represents the bias term, which corresponds to the bias for each category and is used to help adjust the classification boundaries.

[0181] For example, if the system identifies three types of abnormal behavior ,in Indicates accidentally entering, Indicates following, This indicates malicious card swiping. The model's output probability distribution for the abnormal behavior category is as follows: If the value is 75%, it means that the model believes the current behavior has the highest probability of "following" (up to 75%), and the system can respond accordingly.

[0182] Step S3300: Predict the probability distribution based on the abnormal behavior category. With historical operation and maintenance event data set Construct an exception event graph structure .

[0183] Specifically, this step aims to predict the probability distribution of abnormal behavior categories based on the abnormal behavior category constructed in step S3200. Construct an anomaly event graph structure using the historical operation and maintenance event data obtained in step S3100. This system aims to model the spatiotemporal relationships and semantic behavior associations between anomalous events. By introducing a Graph Convolutional Network (GCN) and a dynamic graph update mechanism, the system can deeply mine the potential correlations between anomalous events and adaptively evolve and learn, providing structured graph support for subsequent trajectory reconstruction and anomalous behavior tracing.

[0184] The historical operation and maintenance event data set The access control abnormal event feature representation obtained from step S3100 and video behavior segment feature representation .in Indicates the first Historical operation and maintenance event data for abnormal events may include information such as access control event characteristics and abnormal segment characteristics extracted from surveillance videos; This indicates that the set has a total of One record of historical abnormal data.

[0185] First, define the initial exception event graph. ,in Represents a set of nodes, each node This corresponds to a historical or current abnormal event; Represents the set of edges, each edge It is an event and Potential temporal, spatial, or semantic connections between them. Based on each event node. Construct the initial feature matrix The specific formula is as follows:

[0186] ;

[0187] in, The initial feature matrix includes the feature set of all nodes (i.e., anomalous events) in the initial anomalous event graph at the initial time. Rows represent different events, and columns represent features of each dimension. Represent a OK, The set of real numbers in a column; This represents the feature dimension, the feature dimension of each event node (e.g., 5-dimensional access control features + 10-dimensional video features, for a total of 15 dimensions). Indicates the first The feature vectors of each event node have a dimension of . ; This represents the transpose of a vector.

[0188] ;

[0189] in, Indicates the first Access control behavior characteristics of each event node (such as the time of illegal card swiping and card number type); Indicates the first Video surveillance behavior characteristics of each event node (such as abnormal trajectory distribution and time spent in the area); Indicates feature concatenation operation; .

[0190] Subsequently, based on the historical operation and maintenance event data set Define the initial adjacency matrix based on event correlation. The adjacency matrix elements Indicates an event The formula for determining whether an edge exists is as follows:

[0191] ;

[0192] The aforementioned The criteria for judgment may include: 1. Time proximity (e.g., the two events occur within 5 minutes). Second, spatial proximity (e.g., occurring on the same floor or in the same area); third, similarity in the type of abnormal behavior (e.g., both being tailing behaviors). .

[0193] in, Represents the initial adjacency matrix The first in , For each item, a value of 1 indicates a related item, while a value of 0 indicates no related item. and Indicates the time of the abnormal event, which is the [number]th [event]. , No. The occurrence time of an event (which can be a timestamp, minutes, etc.) is used to determine time proximity. This represents a time threshold, used to determine time proximity. and It is the abnormal behavior category prediction probability distribution constructed in step S3200, which is used to indicate which type of abnormality the event is judged to be (e.g., tailing, forced entry, etc.). This indicates consistency in behavior categories, determining whether the predicted behavior categories of two events are consistent. If they are consistent, it suggests that they may belong to the same type of abnormal behavior.

[0194] Next, a dynamic graph update mechanism is introduced to adapt to new events in real time and dynamically update the adjacency matrix. The specific process formula is as follows:

[0195] ;

[0196] in, Indicates the first The adjacency matrix after the second propagation; Represents a non-linear activation function; This indicates that all nodes in the graph are at the th . The feature matrix at the next iteration; The trainable weight matrix represents graph attention; Representation of the characteristic matrix The transpose operation.

[0197] Based on this, a Graph Convolutional Network (GCN) is used for inter-event information propagation and embedding learning. The specific formula for layer propagation is as follows:

[0198] ;

[0199] in, Indicates the first The higher-order semantic feature embeddings of the layer, after being updated by the graph convolution of this layer, have the following shape: , It is the number of nodes in the graph. It is the first The feature dimension of the layer; This represents the adjacency matrix of self-loops after normalization preprocessing, with the shape as follows: ,Right now , It is the original adjacency matrix. It is the identity matrix; express The degree matrix is ​​a diagonal matrix with the shape of . ; express The inverse square root of the matrix is ​​used to normalize the adjacency matrix and prevent features from being too large or too small in scale during propagation. Indicates the first The weight matrix of the layer has the following shape: .

[0200] Ultimately, through After the layer graph convolution operation, extract the first layer. High-order semantic embedding vectors Combined with node set-based With the final dynamic adjacency matrix Construct a structured exception event graph. The specific formula is as follows:

[0201] ;

[0202] Step S4000, based on the exception event graph structure It performs dynamic reconstruction of abnormal trajectories and provides visual control decision support through the 3D scene of the control interface.

[0203] This step aims to collect personnel entry and exit event data through the access control system, and combine it with spatial mapping to achieve spatial positioning of access control events and detection of abnormal behavior, providing basic data support for subsequent personnel trajectory analysis and security management.

[0204] Furthermore, such as Figure 5 As shown, step S4000 includes:

[0205] Step S4100, based on the exception event graph structure Spatial positioning set of access control events Reconstruct the movement trajectory of abnormal personnel and output the trajectory point sequence. .

[0206] Specifically, this step aims to build upon the exception event graph structure constructed in step S3300. The spatial location set of access control events obtained in step S2200 The movement trajectories of abnormal individuals are reconstructed over time, ultimately outputting a continuous sequence of trajectory points. This provides basic data support for subsequent analysis of abnormal behavior chains.

[0207] To address the issue of sparse or missing access control event location points, an adaptive time interpolation weighted trajectory interpolation algorithm is adopted. This algorithm considers the distance differences between location values ​​at different time points to achieve smooth completion of missing trajectory segments.

[0208] First, regarding the current time point that needs to be estimated. Based on historical trajectory points The corresponding spatial location is calculated using a weighted average interpolation function. The specific process formula is as follows:

[0209] ;

[0210] in, Indicates a point in time The spatial location is the interpolated position at any time. The spatial location of personnel on the data (i.e., interpolation points); This indicates the target time point for interpolation, which is the time point at which the spatial location needs to be estimated. The summation formula is used to perform weighted calculations on all historical points (numbered as follows). ); Represents the interpolation weight, which is the first... Historical trajectory recording points relative to the target time point The degree of influence of interpolation; Representing the spatial coordinates of a known time point, it is the first... The coordinates of each historical trajectory record point.

[0211] ;

[0212] in, This indicates a known time point, which is the [number]th [time point]. The timestamps corresponding to each historical trajectory record point; This represents the interpolation weight index, which controls the sensitivity of time distance to the weights. The larger the index, the smaller the weight of distant points. The summation symbol is used to sum over all trajectory points and to normalize each point. Ensure the total weight is 1; This represents the total number of historical trajectory points.

[0213] Furthermore, to adapt to the changing behavioral characteristics of different abnormal individuals, an adaptive interpolation weighting index is proposed. The dynamic adjustment mechanism. Adaptive interpolation weight index. The specific process formula is as follows:

[0214] ;

[0215] in, This represents the initial interpolation weight index baseline value; A coefficient representing the control adjustment range; The standard deviation of the historical velocity of a trajectory (a volatility index) is used to quantify the volatility of a trajectory (i.e., whether the trajectory is stable or fluctuates drastically).

[0216] Based on the volatility index of historical trajectory speed Automatic adjustment will be performed. If... A large value indicates drastic changes in personnel speed, jumpy behavior, and irregular movement; therefore, a smaller interpolation weight index should be set. To improve interpolation responsiveness; if A small value indicates a stable and linear trajectory, thus a larger interpolation weight exponent should be set. Increase local weighting. Volatility indicator The specific process formula is as follows:

[0217] ;

[0218] ;

[0219] in, Indicates shared ownership The velocity calculation interval for each trajectory segment (i.e., the interval between two trajectory points is one trajectory segment); Indicates an index variable; From the first segment of the trajectory to the... Summation operation of segment trajectories; express The average velocity of the segment trajectory; Represents trajectory points and The Euclidean distance between them; This represents the time interval between two adjacent trajectory points; This represents the average velocity across all trajectory segments;

[0220] Finally, the output is a continuous sequence of trajectory points. , is used to represent continuous trajectory points obtained by interpolation on a complete time axis (including missing segments). Wherein, Indicates the first Each trajectory point records the corresponding timestamp; Indicates the first Each trajectory point records the corresponding timestamp. The corresponding spatial location of the interpolation estimate, i.e., in time The spatial coordinates of the personnel are calculated using an interpolation algorithm.

[0221] Step S4200, based on the trajectory point sequence A trajectory prediction algorithm based on the fusion of Kalman filter and Long Short-Term Memory (LSTM) network is used to obtain the predicted trajectory curve. .

[0222] Specifically, this step is based on the trajectory point sequence output in step S4100. This paper employs a trajectory prediction algorithm based on the fusion of Kalman filter and Long Short-Term Memory (LSTM) network. By smoothing the trajectory and predicting missing points, it outputs a continuous and high-confidence predicted trajectory curve. .

[0223] In the specific process, firstly based on the trajectory point sequence Kalman filtering is applied to smooth the trajectory. Kalman filtering is an optimal recursive filtering algorithm used for dynamic system state estimation. In this application, it is used to estimate the true spatial position of the target individual at each time point, filtering out noise and abrupt changes in the trajectory. The specific process formula is as follows:

[0224] ;

[0225] in, Indicates the first Each trajectory point records the corresponding timestamp. The corresponding spatial location after Kalman filtering; This indicates the spatial location corresponding to the previous time step after Kalman filtering; ' represents the state transition matrix, used to describe the transition from... arrive State evolution; The Kalman gain matrix is ​​used to weigh the confidence level between measured and predicted values. Indicates the current time point Observed values; This represents the observation matrix, used to map the state to the observable space.

[0226] Next, for cases with continuous missing segments or complex motion patterns, a Long Short-Term Memory (LSTM) network is introduced as an auxiliary prediction module to enhance the Kalman filter state and learn the dynamic patterns of the time series in the historical trajectory. The specific process formula is as follows:

[0227]

[0228] ;

[0229] in, This represents the timestamp corresponding to the k-th trajectory point record. The corresponding spatial location based on the fusion of Kalman filter and Long Short-Term Memory (LSTM) network; This indicates the weights of the LSTM predictions in the current state update; Indicates the first Each trajectory point records the corresponding timestamp. The corresponding spatial location predicted by LSTM; This represents the LSTM model function trained based on historical time series trajectory points.

[0230] Finally, the output is a smooth and continuous predicted trajectory curve obtained from the fusion model. .

[0231] Step S4300, based on the predicted trajectory curve Spatial location set of access control abnormal events and probability distribution of abnormal behavior categories It enables visualization of 3D trajectory mapping and display of abnormal brightness.

[0232] Specifically, this step is based on the predicted trajectory curve output in step S4200. Step S2300 constructs a spatial location set for access control anomaly events. and the probability distribution of abnormal behavior categories predicted in step S3200 The trajectory data is mapped to the digital twin 3D simulation environment, and the abnormal visuals are highlighted by the color coding mechanism, so as to realize the spatial visualization of the target individual's movement path and the spatiotemporal tracing of risk events.

[0233] First, predict the trajectory curve Each trajectory point Mapping this to the coordinate system of the digital twin 3D simulation scene, a corresponding 3D spatial trajectory line is constructed. The specific process formula is as follows:

[0234] ;

[0235] in, Indicates the first The spatial coordinates of a trajectory point after mapping in the digital twin 3D environment; This represents the mapping transformation function that maps each trajectory point to the 3D visualization scene.

[0236] Next, all of them According to timestamp The data are sequentially connected to form dynamic trajectory lines, which are then played in a time-driven manner on a 3D visualization platform (such as one based on Cesium or Unity) to reproduce the movement process of the target individual, realizing the dynamic playback and spatial restoration of dynamic behaviors in real scenes in a virtual environment.

[0237] Furthermore, based on the spatial location set of access control anomaly events and probability distribution of abnormal behavior categories For each predicted trajectory curve Each trajectory point Rendering colors Detailed judgments are made to visualize the risk level based on confidence level. The specific color rendering rules and trajectory risk levels are as follows:

[0238] Set probability distribution thresholds, including upper threshold, middle threshold and lower threshold;

[0239] The risk level is... For high-risk warning, ;

[0240] ,but This is a medium-risk warning. ;

[0241] ,but Low risk warning. ;

[0242] ,but This is normal. .

[0243] In this embodiment, the upper threshold is set to 0.85, the middle threshold is set to 0.6, and the lower threshold is set to 0.4.

[0244] in, Indicates the first Risk level of each trajectory point; Indicates the first The rendering color of each trajectory point; Indicates the first The probability distribution of predicted abnormal behavior categories for each trajectory point.

[0245] Step S5000: Combine the spatial location set of access control abnormal events. Probability distribution of abnormal behavior categories and trajectory risk level The system establishes a tiered early warning and linkage response control mechanism and issues closed-loop linkage control commands to the controlled equipment in the target area accordingly. It also collects real-time data on the execution status of the controlled equipment as status feedback and generates a closed-loop access control abnormal event log that includes equipment execution feedback and response status.

[0246] Specifically, this step is based on the spatial location set of access control abnormal events constructed in step S2300. The probability distribution of abnormal behavior categories output by step S3200 Trajectory risk level This method achieves a complete closed-loop control process, the core of which lies in: first, determining a tiered early warning and linkage response control mechanism based on risk analysis results; second, issuing closed-loop linkage control commands to the controlled devices in the physical world based on this mechanism; subsequently, collecting real-time data on the execution status of these devices as status feedback; and finally, generating a closed-loop access control anomaly event log containing device execution feedback and response status. Through this series of operations, this method achieves a complete implementation from intelligent analysis to closed-loop management, effectively improving the proactive security defense capabilities of intelligent data centers.

[0247] Furthermore, such as Figure 6 As shown, step S5000 includes:

[0248] Step S5100: Set the trajectory-based risk level. A tiered early warning and coordinated response control mechanism.

[0249] This step involves real-time, point-by-point assessment of all trajectory points and their risk levels. The system automatically triggers a pre-defined, differentiated, tiered early warning and response control mechanism to ensure timely and accurate system responses. The specific rules are as follows:

[0250] Firstly, high-risk early warning and mandatory intervention response. When High-risk warning, Upon such event, the system immediately triggers a Level 1 high-risk warning event. This not only triggers a Level 1 red alert notification but also automatically generates a set of control parameters for the target equipment and issues closed-loop linkage control commands to the controlled equipment in the target area to proactively intervene in the situation.

[0251] The closed-loop linkage control commands include, but are not limited to, driving at least one of the following operations: 1. Forcibly locking the relevant access control mechanism, changing the door lock to a closed or prohibited state, and preventing the target personnel from going further; 2. Commanding the pan-tilt unit in the video surveillance system to automatically focus on and continuously track the abnormal target personnel, ensuring uninterrupted recording of key images; 3. Activating the security warning lights and audible and visual alarms in the target area, switching to a high-intensity alarm mode to warn the person involved and remind surrounding personnel.

[0252] Secondly, medium-risk alerts and manual decision-making response. When Medium risk warning When the abnormal trajectory and related information are detected, the system will immediately trigger a Level 2 orange medium-risk warning, highlight the abnormal trajectory and related information on the management and control interface, and push a notification. Managers can choose whether to take further action, such as issuing inspection instructions.

[0253] Thirdly, low-risk monitoring and silent recording response. When Low risk warning or normal. At this time, the system does not actively trigger alarms, but it still silently records the relevant trajectory points and their anomaly prediction probabilities as a data source for potential risk trend analysis.

[0254] Step S5200: Generate a closed-loop access control anomaly event log containing device execution feedback and response status.

[0255] Specifically, based on the completion of tiered early warning, this step further realizes the institutional recording and archiving management of abnormal events. By automatically generating closed-loop access control abnormal event logs, the key feature of which is that the log content includes physical-level device execution feedback and system-level response status, thereby constructing a complete and irrefutable traceability evidence chain.

[0256] The access control anomaly event log adopts a structured data format with a unified format and globally unique identifier, including seven key information elements: First, the event number: serving as a unique identifier for each access control anomaly event log, facilitating efficient system indexing and precise location. This number is automatically generated by "timestamp + spatial location code + behavior category code + sequence number," ensuring the global uniqueness and traceability of the event record. For example, the number "EVT20250317-A13-B05" indicates that the event occurred on March 17, 2025, located in area A13 (such as a certain access control floor area in a smart server room), and is the 5th Class B anomaly event identified in that area that day (where "B" represents a certain anomaly category, such as tailgating), possessing good uniqueness and traceability. Second, the event time: recording the specific time when the abnormal trajectory point was determined by the system to be abnormal behavior, adopting the international standard ISO8601 format, serving as an important time benchmark for behavior pattern analysis and response timeliness assessment. Third, the spatial location: containing the relative coordinate information of the trajectory point in the building space. The system includes the access control point ID number and its associated ID number, enabling precise identification of the area where anomalies occur in the smart data center or on a visualized map, supporting fine-grained regional risk analysis and precise dispatch response. Fourthly, it records the anomaly categories and risk levels: including the categories of abnormal behavior identified by the system (such as tailgating, forced entry, prolonged loitering, malicious card swiping, etc.), and simultaneously recording the predicted probability distribution of their corresponding abnormal behavior categories. The system includes seven key features: 1) Track risk level (high / medium / low), providing information management personnel with a reference for determining the nature of anomalies and prioritizing responses. 2) Anomaly warning level recording: Detailed records of whether the system triggers the warning mechanism; if so, the warning level is recorded simultaneously. 3) Equipment execution feedback: Recording execution results collected from each controlled device, such as door lock status "confirmed locked," camera "tracking open," and alarm "sounded." 4) Response loop status: Integrating instructions and feedback information, automatically determining and recording the handling status of this event, such as "successfully closed loop," "execution failed," and "awaiting manual intervention."

[0257] For example, such as Figure 7As shown, it presents an example of the access control exception event log in a visual monitoring scenario of a smart computer room. This log presents five key pieces of information about the event in a visual panel, including the event number, event time, spatial location, exception category and risk level, exception warning record, device execution feedback, and response closed-loop status, etc. For example, the event number is EVT20250317-A13-B01, recorded at 08:22:57 on March 17, 2025, occurring in area A13, the access control number is D-17, and the three-dimensional coordinates are (12.3, 8.7, 0.0). The system determines that the exception category is "tailgating in", the confidence level is 0.92, and the risk level is high. The system has triggered a high-level warning. At the same time, the "device execution feedback" field of this log further records the actual results of the system's linkage control, specifically: "Access control D-17: Status confirmed locked; Camera: Focus tracking has been started; Acoustic-optic alarm: Ringing has been activated". Its "response closed-loop status" field automatically determines and records as "successfully closed loop" based on the above successful instruction execution and status collection, thus forming a complete traceability evidence chain covering the entire process of perception, decision-making, action, and feedback.

[0258] Specifically, this step has the following technical advantages: First, it supports the transformation of computer room security management from static auditing to dynamic perception. In the past, computer room security management relied more on regular manual auditing of access control records and video materials, with a large number of blind spots and delays. This invention endows the system with real-time dynamic perception and intelligent archiving capabilities, promoting the operation and maintenance system to transform from "post-event auditing" to "in-process perception + immediate response", meeting the trend requirements of the future data center's evolution towards intelligent and autonomous security management. Second, it enhances the ability to prevent access risks in important protected areas. In high-security-level areas, the traditional mechanism based on "door recognition + log recording" cannot meet the system's tracking requirements for intrusion paths, behavior patterns, and post-event evidence chains. This invention establishes a complete semantic chain of "time + space + behavior + response" for abnormal behaviors through automatically archived access control exception event logs with high confidence levels and spatial annotation capabilities, enabling managers to have clear, traceable, and controllable decision-making bases during emergency responses. Third, the structured expression of access control abnormal behaviors and cross-platform linkage tracking. Through the unified format standard of access control exception event logs, multi-angle and quantifiable records of access control behaviors are achieved from the data dimension, spatial dimension to the time dimension, endowing the video monitoring system and the dispatching response platform with the capabilities of accurate indexing, fast playback, and intelligent comparison, supporting the whole-process analysis and closed-loop control of abnormal events.

[0259] Embodiment 2

[0260] Based on Embodiment 1, this embodiment provides a multi-dimensional dynamic monitoring and management system for a smart computer room based on digital twins, as Figure 8 shown, including:

[0261] Digital twin spatial modeling module: used to collect 3D point cloud sets of the computer room space. With RGB image collection A digital twin space representing the physical state of the computer room and supporting monitoring and control is constructed, and a spatial mapping table between the digital twin space and the physical space is established. ;

[0262] Access control anomaly event localization module: based on spatial mapping table This method maps personnel entry and exit event data generated by the access control system to corresponding spatial locations in a digital twin space, continuously monitors personnel behavior in real time to identify abnormal events, and constructs a spatial location set of abnormal access control events. ;

[0263] Abnormal Behavior Recognition and Abnormal Event Graph Construction Module: Based on Spatial Localization Set of Access Control Abnormal Events Perform abnormal behavior identification and correlation analysis, and construct an abnormal event graph structure. .

[0264] Abnormal trajectory reconstruction and 3D visualization control module: Based on the abnormal event graph structure, it performs dynamic reconstruction of abnormal trajectories and realizes visualization control decision support through the 3D scene of the control interface.

[0265] Linked early warning control and log generation module: It is used to combine the spatial location set of access control abnormal events, the predicted probability distribution of abnormal behavior categories and trajectory risk level to determine the hierarchical early warning linkage response control mechanism and issue closed-loop linkage control commands to the controlled equipment in the target area accordingly. It also collects the execution status of the controlled equipment in real time as status feedback and generates a closed-loop access control abnormal event log containing equipment execution feedback and response status.

[0266] The parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0267] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-dimensional dynamic monitoring and management method for smart data centers based on digital twins, characterized in that: include: Collect a set of 3D point clouds and RGB images of the computer room space, construct a digital twin space that represents the physical state of the computer room and supports monitoring and control, and establish a spatial mapping table between the digital twin space and the physical space. Based on the spatial mapping table, the personnel entry and exit event data generated by the access control system are mapped to the corresponding spatial locations in the digital twin space. Continuous real-time status detection of personnel behavior is performed to identify abnormal event behaviors and construct a spatial location set of abnormal access control events. Based on the spatial location set of access control abnormal events, perform abnormal behavior identification and correlation analysis, and construct an abnormal event graph structure; Based on the abnormal event graph structure, the abnormal trajectory is dynamically reconstructed, and the 3D scene of the control interface is used to realize visual control decision support. By combining the spatial location set of access control abnormal events, the predicted probability distribution of abnormal behavior categories and the trajectory risk level, a graded early warning linkage response control mechanism is determined, and a closed-loop linkage control command is issued to the controlled equipment in the target area accordingly. The execution status of the controlled equipment is collected in real time as status feedback, and a closed-loop access control abnormal event log containing equipment execution feedback and response status is generated. The method for constructing the abnormal event graph structure includes: Based on the spatial localization set of access control abnormal events, feature representations of access control abnormal events and video behavior segments are obtained to construct multimodal features; Based on multimodal features, a temporal convolutional network is introduced to construct the probability distribution for predicting abnormal behavior categories; An abnormal event graph structure is constructed based on the predicted probability distribution of abnormal behavior categories and the historical operation and maintenance event data set.

2. The method for multi-dimensional dynamic monitoring and management of intelligent data centers based on digital twins according to claim 1, characterized in that, The method for establishing the spatial mapping table includes: Multimodal fusion data is constructed based on 3D point cloud sets and RGB image sets; Based on a multimodal fusion dataset and combined with a building information model, a spatial index structure is established; Spatial mapping is performed based on a device set, a spatial model with semantic tags, and a spatial index structure. A temporal index update mechanism is introduced to establish a spatial mapping table. The time-series index update mechanism uses a KD-Tree to dynamically maintain the spatial index structure.

3. The method for multi-dimensional dynamic monitoring and management of intelligent data centers based on digital twins according to claim 2, characterized in that, The method for establishing the spatial index structure includes: Based on the multimodal fusion dataset, a voxel meshing algorithm is used to perform structured transformation on the 3D point cloud data to obtain a 3D point cloud voxel set; By combining the structural component information provided by the building information model with the three-dimensional point cloud voxel set for spatial alignment, and performing a multimodal semantic segmentation process, a spatial model with semantic labels is obtained. A spatial index structure is constructed based on a spatial model with semantic labels.

4. The method for multi-dimensional dynamic monitoring and management of intelligent data centers based on digital twins according to claim 1, characterized in that, The method for constructing the spatial location set of access control abnormal events includes: Obtain the original set of access control events, and construct a high-quality set of access control events based on the original set of access control events; Based on the high-quality access control event set and spatial mapping table, a spatial location set of access control events is obtained; Based on the spatial location set of access control events, construct the spatial location set of access control abnormal events.

5. The method for multi-dimensional dynamic monitoring and management of intelligent data centers based on digital twins according to claim 4, characterized in that, The method for constructing the abnormal event graph structure also includes: Define an initial anomaly event graph, and construct an initial feature matrix based on each event node; Based on the historical operation and maintenance event data set, an initial adjacency matrix is ​​defined according to the event correlation. The elements of the initial adjacency matrix indicate whether there is an edge between events; A dynamic graph update mechanism is introduced to adapt to new events in real time and dynamically update the adjacency matrix; Graph convolutional networks are used for inter-event information propagation and embedding learning to obtain the first... High-order semantic feature embedding of layers; pass After the layer graph convolution operation, extract the first layer. The higher-order semantic embedding vectors are combined with the node set and the final dynamic adjacency matrix to construct a structured abnormal event graph structure.

6. The method for multi-dimensional dynamic monitoring and management of intelligent data centers based on digital twins according to claim 1, characterized in that, The steps for dynamic reconstruction and 3D scene visualization of the abnormal trajectory include: Each trajectory point of the predicted trajectory curve is mapped to the coordinate system of the digital twin 3D simulation scene to construct the corresponding 3D spatial trajectory line; Connect all three-dimensional spatial trajectory points in time stamp order to form a dynamic trajectory line, and play it in a time-driven manner in the three-dimensional visualization platform; Based on the spatial location set of access control abnormal events and the predicted probability distribution of abnormal behavior categories, the rendering color of each trajectory point of each predicted trajectory curve is determined; The determination of the rendering color for each trajectory point of each predicted trajectory curve includes: Set probability distribution thresholds, including upper threshold, middle threshold and lower threshold; If the trajectory point belongs to the spatial location set of abnormal access control events and the predicted probability distribution of abnormal behavior categories is greater than the upper threshold, then the risk level is high-risk warning and the rendering color is red. If the trajectory point belongs to the spatial location set of abnormal access control events and the predicted probability distribution of the abnormal behavior category is greater than the medium threshold and less than or equal to the upper threshold, then the risk level is medium risk warning and the rendering color is orange. If the trajectory point belongs to the spatial location set of abnormal access control events and the predicted probability distribution of the abnormal behavior category is greater than or equal to the lower threshold and less than or equal to the middle threshold, then the risk level is low risk and the rendering color is yellow. If the trajectory point does not belong to the spatial location set of access control abnormal events, the risk level is normal and the rendering color is green.

7. The method for multi-dimensional dynamic monitoring and management of intelligent data centers based on digital twins according to claim 6, characterized in that, When the trajectory risk level is high-risk warning and the rendering color is red, the system immediately triggers a level one high-risk warning event; when the trajectory risk level is medium-risk warning and the rendering color is orange, the system immediately triggers a level two medium-risk warning. When the trajectory risk level is low risk or normal, and the rendering color is yellow or green, the system will not actively trigger an alarm.

8. A multi-dimensional dynamic monitoring and management system for intelligent data centers based on digital twins, used to implement the multi-dimensional dynamic monitoring and management method for intelligent data centers based on digital twins as described in any one of claims 1-7, characterized in that, The system includes: Digital Twin Space Modeling Module: Used to collect 3D point cloud sets and RGB image sets of the computer room space, construct a digital twin space that represents the physical state of the computer room and supports monitoring and control, and establish a spatial mapping table between the digital twin space and the physical space; Access control abnormal event location module: Based on the spatial mapping table, the access control system maps personnel entry and exit event data to the corresponding spatial location in the digital twin space, continuously monitors personnel behavior in real time to identify abnormal event behavior, and constructs a set of spatial locations for access control abnormal events; Abnormal Behavior Recognition and Abnormal Event Graph Construction Module: Based on the spatial location set of abnormal access control events, it performs abnormal behavior recognition and correlation analysis, and constructs an abnormal event graph structure; Abnormal trajectory reconstruction and 3D visualization control module: Based on the abnormal event graph structure, it performs dynamic reconstruction of abnormal trajectories and realizes visualization control decision support through the 3D scene of the control interface; Linked early warning control and log generation module: It is used to combine the spatial location set of access control abnormal events, the predicted probability distribution of abnormal behavior categories and trajectory risk level to determine the hierarchical early warning linkage response control mechanism and issue closed-loop linkage control commands to the controlled equipment in the target area accordingly. It also collects the execution status of the controlled equipment in real time as status feedback and generates a closed-loop access control abnormal event log containing equipment execution feedback and response status.

Images