A smart computer room control method based on artificial intelligence

Abstract

The present invention relates to a method for controlling a smart computer room based on artificial intelligence, comprising the following steps: S1: creating a static three-dimensional model based on the three-dimensional data of the computer room; S2: combining the power diagram and the environmental cloud diagram of the computer room into the static three-dimensional model to obtain a dynamic three-dimensional model; S3: building a visualization platform based on the dynamic three-dimensional model using WebGL; S4: detecting abnormal data nodes through an abnormal detection model based on the operation data of the equipment, and marking and reporting abnormal data nodes in the dynamic three-dimensional model; S5: combining Wi‑Fi and UWB tag technology to achieve accurate positioning of computer room personnel, and through role authorization and real-time location monitoring, based on the dynamic three-dimensional model, refined authority management is performed; S6: based on the real-time positioning of personnel and abnormal data nodes, automatically generating the optimal path for personnel, and assisting personnel to efficiently reach the designated task area. While improving the reliability of computer room management, the present invention improves the utilization efficiency of resources and the rigor of data management.

Description

Technical Field

[0001] The present invention relates to the field of computer room management and control, and in particular to an intelligent computer room management and control method based on artificial intelligence. Background Art

[0002] In today's highly digitalized and interconnected environment, all industries are constantly seeking ways to improve efficiency, reduce costs and enhance service capabilities. This is especially true for the power industry, whose traditional infrastructure management model faces many challenges, including high equipment density, strong real-time monitoring requirements, and complex failure risk management.

[0003] With the surge in computing demand and data volume, computer rooms have become one of the core assets of the power industry. The complexity and density of equipment in computer rooms make it difficult for traditional manual management methods to meet the requirements of efficient operation and maintenance and safe management. To meet this challenge, the industry is gradually turning to emerging technologies such as artificial intelligence and 3D modeling to achieve automated and intelligent management of smart computer rooms. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide an intelligent computer room management and control method based on artificial intelligence, so as to realize the intelligent, automated and efficient management and maintenance of the computer room, thereby improving the resource utilization efficiency and the rigor of data management while improving reliability.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A smart computer room management and control method based on artificial intelligence includes the following steps:

[0007] S1: Use laser scanning equipment to obtain the three-dimensional data of the computer room, and create a static three-dimensional model based on the three-dimensional data of the computer room;

[0008] S2: Obtain the operating data and environmental data of all equipment in the computer room, construct a power map and an environmental cloud map of the computer room, and combine the power map and the environmental cloud map of the computer room into a static three-dimensional model to obtain a dynamic three-dimensional model;

[0009] S3: Based on the dynamic 3D model, a visualization platform is built using WebGL, allowing personnel to view and manage all equipment in the computer room from a 3D perspective;

[0010] S4: Based on the operation data of the equipment, detect abnormal data nodes through the anomaly detection model, and mark and report the abnormal data nodes in the dynamic three-dimensional model;

[0011] S5: Combine Wi-Fi and UWB tag technology to achieve accurate positioning of personnel in the computer room, and conduct refined permission management based on dynamic 3D models through role authorization and real-time location monitoring;

[0012] S6: Based on the real-time positioning of personnel and abnormal data nodes, taking into account authority management, the nearest suitable personnel are selected, and the path planning algorithm is combined to automatically generate the optimal path for the personnel. AR navigation function is provided on mobile devices to assist personnel to reach the designated task area efficiently.

[0013] Furthermore, S1 is specifically:

[0014] Use high-precision LiDAR equipment to perform multi-angle scanning to capture detailed point cloud data of every corner and equipment in the machine room;

[0015] Based on SOR filtering, noise points of point cloud data are removed to obtain filtered multi-angle scanning point cloud data;

[0016] Use the ICP algorithm to stitch and globally align the filtered multi-angle scanning point cloud data so that all scanning data are merged into a unified point cloud;

[0017] The merged point cloud data is imported into the 3D modeling software as the basis for modeling, and surface reconstruction technology (such as Poisson Surface Reconstruction) is used to convert the point cloud into a visual 3D model.

[0018] Furthermore, based on the SOR filtering process, the noise points of the point cloud data are removed to obtain the filtered multi-angle scanning point cloud data, as follows:

[0019] Initialization parameters, number of neighborhood points k and standard deviation multiplier threshold σ;

[0020] For each point p in the point cloud i , calculate the Euclidean distance between it and other points in its neighborhood, and calculate the average neighborhood distance :

[0021] ;

[0022] in, p j for p i Any point in the neighborhood; k is the number of neighborhood points;

[0023] Calculate the mean and standard deviation of the neighborhood distances of all points , using the standard deviation multiplier Define the threshold to filter out outliers. For each point p in the point cloud i, the difference between the neighborhood average distance and the global average distance exceeds , then the point is marked as an outlier, that is, a possible noise point.

[0024] Furthermore, the ICP algorithm is used to stitch and globally align the filtered multi-angle scanning point cloud data so that all scanning data are merged into a unified point cloud, as follows:

[0025] Obtain initial alignment based on the initial positioning data of the high-precision LiDAR device;

[0026] For each point in the point cloud P , the number of points in point cloud P is N, find the nearest point in point cloud Q , forming a point pair ;

[0027] Using singular value decomposition, the optimal rigid body transformation is calculated based on the nearest point pair, including the rotation matrix R and the displacement vector t, so that the square error of the two sets of points is minimized:

[0028] ;

[0029] Update the point cloud P:

[0030] ;

[0031] in, The updated point ;

[0032] Calculate the matching error of the current iteration. If the error is lower than the preset threshold or reaches the maximum number of iterations, terminate the iteration.

[0033] Multiple aligned point clouds are fused to generate unified point cloud data.

[0034] Furthermore, S2 is specifically:

[0035] Connect various sensors based on the industrial Internet of Things to obtain real-time operation data and environmental data of equipment;

[0036] Use a database to store time series data collected at a high frequency and standardize the data to match the corresponding dimensional units in the 3D model;

[0037] Integrate equipment operating parameter data, draw power diagrams representing equipment power consumption, efficiency or status changes, and draw environmental cloud diagrams, including temperature distribution diagrams and high humidity area signs;

[0038] Import the static 3D model of the computer room into the 3D modeling software, add data interfaces and visualization components for each device and monitoring point in the model; capture and map real-time data to the static 3D model through the data interface; the visualization component defines the mapping function from data values ​​to visualization expressions, and uses dynamic texture mapping to display power diagrams and environmental cloud maps on the model surface.

[0039] Furthermore, the anomaly detection model is constructed by combining the autoencoder and OC-SVM as follows:

[0040] Collect multi-dimensional data during equipment operation, standardize the data, extract relevant features, and construct a training data set;

[0041] The autoencoder consists of an encoder and a decoder, which are used for dimensionality reduction and reconstruction of data:

[0042] Encoder: maps the training dataset data x to a low-dimensional latent space z;

[0043] ；

[0044] in, W enc and b enc are the weight matrix and bias vector of the encoder, respectively. is the activation function;

[0045] Decoder: reconstructs the input data from the latent space z;

[0046] ；

[0047] in, w dec and b dec are the weight matrix and bias vector of the decoder respectively;

[0048] The difference between the original input and the reconstructed output is minimized by the reconstruction loss function:

[0049]

[0050] Where n represents the number of data samples, It is input samples, It is Reconstructed output;

[0051] The latent space feature z output by the encoder is used to train the OC-SVM model, using the RBF kernel to account for more nonlinear distributions:

[0052] ;

[0053] in, is the kernel function, which calculates the similarity between z and z′ in the feature space; γ is the parameter in the kernel function that controls the distribution expansion; z′ is the latent space feature that is different from z;

[0054] OC-SVM objective optimization function: find a hyperplane to mark ν% of samples as normal rather than abnormal:

[0055]

[0056] Among them, w is the weight vector of the decision function; For the Slack variables; is bias;

[0057] Real-time data is passed through the autoencoder to obtain low-dimensional features z d , use OC-SVM to determine whether it is abnormal:

[0058] If the OC-SVM decision function f(z d )<0 is marked as abnormal:

[0059] ;

[0060] in, is the feature mapping function; represents the inner product operation;

[0061] During the training process, the structure of the autoencoder and the parameters of the OC-SVM are adjusted through hyperparameter optimization.

[0062] Furthermore, the structure of the autoencoder and the parameters of the OC-SVM are adjusted through hyperparameter optimization, as follows:

[0063] Use the neural network library to define the autoencoder. The autoencoder architecture includes an input layer, several hidden layers, a bottleneck layer, and a decoder.

[0064] Hyperparameters include the number of encoder layers, encoder units, and learning rate:

[0065] Under each hyperparameter combination, a preset number of training iterations are performed to enable the autoencoder to achieve the reconstruction target;

[0066] Use the trained autoencoder to obtain low-dimensional latent features of the input data;

[0067] The support vector machine library is used to read the latent features output by the autoencoder for OC-SVM training. The hyperparameter combination is adjusted using Bayesian optimization, and the hyperparameter combination with the highest performance index is selected as the final model parameters.

[0068] Furthermore, Bayesian optimization is used to adjust the hyperparameter combination as follows:

[0069] Randomly select or use Latin hypercube sampling to select several hyperparameter combinations for initial evaluation in the hyperparameter search space;

[0070] Use the initial sampling points to build a GP model and fit the objective function;

[0071] Update the mean and covariance matrix of the GP agent model using all current data samples;

[0072] Using expected improvement EI as the acquisition function, it is defined as follows:

[0073] ;

[0074] in, is the best target value currently observed, and further we get:

[0075]

[0076] in, is the predicted mean of the Gaussian process surrogate model at point x; The standard deviation of the Gaussian process prediction at point x; is the best observed in the current search process; To explore the parameters; is the value of the standard normal cumulative distribution function; is the value of the standard normal probability density function; Z is the standardization parameter;

[0077] Maximize the acquisition function EI(x) to select the next hyperparameter combination for model evaluation;

[0078] Train the model with the selected hyperparameter combination, record and calculate the objective function value;

[0079] Terminate the optimization process according to the maximum number of iterations and patience parameters;

[0080] The hyperparameters with the highest objective function value are returned as the final best combination.

[0081] Furthermore, by combining Wi-Fi and UWB tag technology, accurate positioning of personnel in the computer room can be achieved. Specifically:

[0082] The RSSI of Wi-Fi was measured at different fixed locations in the equipment room. A model between signal strength and physical distance was established through experiments, and the path loss model was used to convert signal strength into distance.

[0083] Deploy multiple UWB anchor points and use ToF measurements to convert signal round-trip time into distance;

[0084] When personnel enter and leave the computer room, they carry UWB tags, and anchor points are deployed at preset locations according to the computing area; first

[0085] Use the existing Wi-Fi access points in the computer room to make a rough location estimate using RSSI;

[0086] Collect UWB ToF data and Wi-Fi RSSI data;

[0087] The fused data is input into the Kalman filter for integration and synchronized using the current timestamp;

[0088] Convert signal strength data into preliminary distance estimates;

[0089] Synchronize the information provided by UWB precise ranging and RSSI into a unified coordinate system.

[0090] Furthermore, S6 is specifically:

[0091] Query the role permission database to determine which personnel have the authority to handle specific abnormal nodes, and use the following formula to verify permissions:

[0092]

[0093] in, is the verification formula; Indicates the authority of person u, Is an abnormal node Required permissions; is an empty set;

[0094] From the authorized personnel, select the personnel who are closest to the abnormal node and have the current available status, and use the Euclidean distance calculation to select the nearest personnel;

[0095] use Path planning algorithm, based on the current personnel location and abnormal node location, finds the optimal path;

[0096] Integrate an AR engine in mobile devices to overlay navigation instructions on the camera view;

[0097] Generate step-by-step navigation prompts based on the path planning results, and display indicator arrows and information markers in the AR view;

[0098] Update positioning information and navigation paths at any time, and dynamically adjust navigation routes based on personnel actions.

[0099] The present invention has the following beneficial effects:

[0100] 1. The present invention realizes intelligent, automated and efficient management and maintenance of the computer room, which improves the resource utilization efficiency and the rigor of data management while improving reliability;

[0101] 2. The present invention uses SOR filtering to effectively remove noise and abnormal points in point cloud data, improve the quality of point cloud data, and reduce the impact of redundant and erroneous data; based on a clear point cloud, the ICP algorithm is used to accurately align the point cloud to ensure that the point cloud scanned from multiple angles can be accurately spliced ​​into a unified model, and the data of multiple scans can be effectively integrated to achieve accurate global alignment of the point cloud;

[0102] 3. The present invention combines the feature representation of the autoencoder and OC-SVM to build an anomaly detection model through end-to-end model combination, which can effectively cope with the anomaly detection tasks in complex data sets, especially for the actual data scenarios that need to process high-dimensional and nonlinear data in computer room management;

[0103] 4. The present invention integrates advanced wireless positioning technology with precise algorithm processing to realize a comprehensive solution for real-time positioning and authority management of personnel, ensuring the safety of the computer room and efficient operation, and achieving the fastest response and resolution in abnormal situations, thereby improving overall operational efficiency and response capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 is a flow chart of the method of the present invention;

[0105] Figure 2 FIG. 1 is a schematic diagram of a system architecture in an embodiment of the present invention. DETAILED DESCRIPTION

[0106] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0107] refer to Figure 1 In this embodiment, a smart computer room management and control method based on artificial intelligence is provided, comprising the following steps:

[0108] S1: Use laser scanning equipment to obtain the three-dimensional data of the computer room, and create a static three-dimensional model based on the three-dimensional data of the computer room;

[0109] S2: Obtain the operating data and environmental data of all equipment in the computer room, construct a power map and an environmental cloud map of the computer room, and combine the power map and the environmental cloud map of the computer room into a static three-dimensional model to obtain a dynamic three-dimensional model;

[0110] S3: Based on the dynamic 3D model, a visualization platform is built using WebGL, which allows personnel to view and manage all equipment in the computer room from a 3D perspective. The information includes equipment overview, type, current status, historical data and trends, real-life images, and operating status. The power diagram and environmental cloud diagram of the computer room can be used to better understand the impact of the external environment on the computer room.

[0111] S4: Based on the operation data of the equipment, detect abnormal data nodes through the anomaly detection model, and mark and report the abnormal data nodes in the dynamic three-dimensional model;

[0112] S5: Combine Wi-Fi and UWB tag technology to achieve precise positioning of personnel in the computer room, and conduct refined permission management (such as access to buildings, computer rooms, micromodules or specific cabinets) based on dynamic three-dimensional models through role authorization and real-time location monitoring;

[0113] S6: Based on the real-time positioning of personnel and abnormal data nodes, taking into account authority management, the nearest suitable personnel are selected, and the path planning algorithm is combined to automatically generate the optimal path for the personnel. AR navigation function is provided on mobile devices to assist personnel to reach the designated task area efficiently.

[0114] In this embodiment, S1 is specifically:

[0115] Use high-precision LiDAR equipment to perform multi-angle scanning to capture detailed point cloud data of every corner and equipment in the room, including ceilings, walls, and equipment surfaces;

[0116] Based on SOR filtering, noise points of point cloud data are removed to obtain filtered multi-angle scanning point cloud data;

[0117] Use the ICP algorithm to stitch and globally align the filtered multi-angle scanning point cloud data so that all scanning data are merged into a unified point cloud;

[0118] Import the merged point cloud data into the 3D modeling software as the basis for modeling, and use surface reconstruction technology (such as Poisson Surface Reconstruction) to convert the point cloud into a visual 3D model. Model the key structures (ceilings, walls, equipment) one by one to ensure accurate comparison with the actual situation, and create a detailed 3D model based on the processed point cloud data.

[0119] In this embodiment, noise points of the point cloud data are removed based on SOR filtering to obtain filtered multi-angle scanning point cloud data, as follows:

[0120] Initialization parameters, number of neighborhood points k and standard deviation multiplier threshold σ;

[0121] For each point p in the point cloudi , calculate the Euclidean distance between it and other points in its neighborhood, and calculate the average neighborhood distance :

[0122] ;

[0123] in, p j for p i Any point in the neighborhood; k is the number of neighborhood points;

[0124] Calculate the mean and standard deviation of the neighborhood distances of all points , using the standard deviation multiplier Define the threshold to filter out outliers. For each point p in the point cloud i , the difference between the neighborhood average distance and the global average distance exceeds , then the point is marked as an outlier, that is, a possible noise point.

[0125] In this embodiment, the ICP algorithm is used to stitch and globally align the filtered multi-angle scanning point cloud data so that all the scanning data are merged into a unified point cloud, as follows:

[0126] Obtain initial alignment based on the initial positioning data of the high-precision LiDAR device;

[0127] For each point in the point cloud P , the number of points in point cloud P is N, find the nearest point in point cloud Q , forming a point pair ;

[0128] Using singular value decomposition, the optimal rigid body transformation is calculated based on the nearest point pair, including the rotation matrix R and the displacement vector t, so that the square error of the two sets of points is minimized:

[0129] ;

[0130] Update the point cloud P:

[0131] ;

[0132] in, The updated point ;

[0133] Calculate the matching error of the current iteration. If the error is lower than the preset threshold or reaches the maximum number of iterations, terminate the iteration.

[0134] Multiple aligned point clouds are fused to generate unified point cloud data.

[0135] In this embodiment, S2 is specifically:

[0136] Based on the industrial Internet of Things, various sensors are connected to obtain real-time operation data and environmental data of equipment, including temperature, power consumption, speed, status, etc.;

[0137] Use a database (such as InfluxDB, TimescaleDB) to store time series data collected at a high frequency and normalize the data to match the corresponding dimension units in the three-dimensional model;

[0138] Integrate equipment operating parameter data, draw power diagrams representing equipment power consumption, efficiency or status changes, and draw environmental cloud diagrams, including temperature distribution diagrams and high humidity area signs;

[0139] Import the static 3D model of the computer room into the 3D modeling software, add data interfaces and visualization components for each device and monitoring point in the model; capture and map real-time data to the static 3D model through the data interface; the visualization component defines the mapping function from data values ​​to visualization expressions (such as color, size, animation), and uses dynamic texture mapping to display power diagrams and environmental cloud maps on the model surface.

[0140] In this embodiment, the anomaly detection model is constructed by combining the autoencoder and OC-SVM, as follows:

[0141] Collect multi-dimensional data during equipment operation, standardize the data, extract relevant features, and construct a training data set;

[0142] The autoencoder consists of an encoder and a decoder, which are used for dimensionality reduction and reconstruction of data:

[0143] Encoder: maps the training dataset data x to a low-dimensional latent space z;

[0144]

[0145] in, W enc and b enc are the weight matrix and bias vector of the encoder, respectively. is the activation function;

[0146] Decoder: reconstructs the input data from the latent space z;

[0147] ；

[0148] in, w dec and b decare the weight matrix and bias vector of the decoder respectively;

[0149] The difference between the original input and the reconstructed output is minimized by the reconstruction loss function:

[0150]

[0151] Where n represents the number of data samples, It is input samples, It is Reconstructed output;

[0152] The latent space feature z output by the encoder is used to train the OC-SVM model, using the RBF kernel to account for more nonlinear distributions:

[0153] ;

[0154] in, is the kernel function, which calculates the similarity between z and z′ in the feature space; γ is the parameter in the kernel function that controls the distribution expansion; z′ is the latent space feature that is different from z;

[0155] OC-SVM objective optimization function: find a hyperplane to mark ν% of samples as normal rather than abnormal:

[0156]

[0157] Among them, w is the weight vector of the decision function; For the Slack variables; is bias;

[0158] Real-time data is passed through the autoencoder to obtain low-dimensional features z d , use OC-SVM to determine whether it is abnormal:

[0159] If the OC-SVM decision function f(z d )<0 is marked as abnormal:

[0160] ;

[0161] in, is the feature mapping function; represents the inner product operation;

[0162] During the training process, the structure of the autoencoder and the parameters of the OC-SVM are adjusted through hyperparameter optimization.

[0163] In this embodiment, the structure of the autoencoder and the parameters of the OC-SVM are adjusted by hyperparameter optimization, as follows:

[0164] Define an autoencoder using a neural network library such as TensorFlow or PyTorch. The autoencoder architecture consists of an input layer, several hidden layers, a bottleneck layer (the narrowest hidden layer), and a decoder.

[0165] Hyperparameters include, encoder layers and encoder units: determine the number of hidden layers and the number of neurons in each layer of the encoder and decoder, learning rate: control the training step size;

[0166] Under each hyperparameter combination, a preset number of training iterations are performed to enable the autoencoder to achieve the reconstruction target;

[0167] Use the trained autoencoder to obtain low-dimensional latent features of the input data; that is, the bottleneck layer output;

[0168] The support vector machine library is used to read the latent features output by the autoencoder for OC-SVM training. The hyperparameter combination is adjusted using Bayesian optimization, and the hyperparameter combination with the highest performance index is selected as the final model parameters.

[0169] In this embodiment, Bayesian optimization is used to adjust the hyperparameter combination, as follows:

[0170] Randomly select or use Latin hypercube sampling to select several hyperparameter combinations for initial evaluation in the hyperparameter search space;

[0171] Use the initial sampling points to build a GP model and fit the objective function (including performance indicators such as F1-score and AUC);

[0172] Update the mean and covariance matrix of the GP agent model using all current data samples (hyperparameter combinations that have been evaluated);

[0173] Using expected improvement EI as the acquisition function, it is defined as follows:

[0174] ;

[0175] in, is the best target value currently observed, and further we get:

[0176]

[0177] in, is the predicted mean of the Gaussian process surrogate model at point x; The standard deviation of the Gaussian process prediction at point x; is the best observed in the current search process; To explore the parameters; is the value of the standard normal cumulative distribution function; is the value of the standard normal probability density function; Z is the standardization parameter;

[0178] Maximize the acquisition function EI(x) to select the next hyperparameter combination for model evaluation;

[0179] Train the model with the selected hyperparameter combination and record and calculate the objective function value (performance metric).

[0180] The optimization process can be terminated based on criteria such as the maximum number of iterations, patience parameters, or when the target performance reaches a certain level;

[0181] The hyperparameters with the highest objective function value are returned as the final best combination.

[0182] In this embodiment, Wi-Fi and UWB tag technology are combined to achieve accurate positioning of the personnel in the computer room, specifically:

[0183] The RSSI of Wi-Fi was measured at different fixed locations in the equipment room. A model between signal strength and physical distance was established through experiments, and the path loss model was used to convert signal strength into distance.

[0184] Deploy multiple UWB anchor points and use ToF measurements to convert signal round-trip time into distance;

[0185] When personnel enter and leave the computer room, they carry UWB tags, and anchor points are deployed at preset locations according to the computing area; first

[0186] Use the existing Wi-Fi access points in the computer room to make a rough location estimate using RSSI;

[0187] Collect UWB ToF data and Wi-Fi RSSI data;

[0188] The fused data is input into the Kalman filter for integration and synchronized using the current timestamp;

[0189] Convert signal strength data into preliminary distance estimates;

[0190] Synchronize the information provided by UWB precise ranging and RSSI into a unified coordinate system.

[0191] In this embodiment, S6 is specifically:

[0192] Query the role permission database to determine which personnel have the authority to handle specific abnormal nodes, and use the following formula to verify permissions:

[0193]

[0194] in, is the verification formula; Indicates the authority of person u, Is an abnormal node Required permissions; is an empty set;

[0195] From the authorized personnel, select the personnel who are closest to the abnormal node and have the current available status, and use the Euclidean distance calculation to select the nearest personnel;

[0196] use Path planning algorithm, based on the current personnel location and abnormal node location, finds the optimal path;

[0197] Integrate an AR engine (such as ARKit or ARCore) in the mobile device to overlay navigation instructions on the camera view;

[0198] Generate step-by-step navigation prompts based on the path planning results, and display indicator arrows and information markers in the AR view;

[0199] Update positioning information and navigation paths at any time, and dynamically adjust navigation routes based on personnel actions.

[0200] refer to Figure 2 ,The present invention also provides another embodiment, a smart computer room management and control system based on artificial intelligence, including ,a data acquisition module, a data visualization module, a data analysis module, a personnel management module and a navigation module;

[0201] Data acquisition module, which obtains the operating data and environmental data of all equipment in the computer room;

[0202] The data visualization module uses WebGL to build a visualization platform based on a dynamic 3D model, allowing personnel to view and manage all equipment in the computer room from a 3D perspective. The information includes equipment overview, type, current status, historical data and trends, real-life images, and operating status. The power diagram and environmental cloud diagram of the computer room can be used to better understand the impact of the external environment on the computer room.

[0203] The data analysis module detects abnormal data nodes based on the operation data of the equipment through the anomaly detection model, and marks and reports the abnormal data nodes in the data visualization module;

[0204] The personnel management module combines Wi-Fi and UWB tag technology to achieve precise positioning of personnel in the computer room, and conducts refined permission management based on dynamic 3D models through role authorization and real-time location monitoring;

[0205] The navigation module selects the nearest suitable personnel based on the real-time positioning of personnel and abnormal data nodes, takes into account authority management, and combines the path planning algorithm to automatically generate the optimal path for personnel. It provides AR navigation function on mobile devices to assist personnel to reach the designated task area efficiently.

[0206] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0207] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0208] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0209] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0210] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A smart computer room management and control method based on artificial intelligence, characterized in that: The following steps are involved: S1: Use laser scanning equipment to obtain the three-dimensional data of the computer room, and create a static three-dimensional model based on the three-dimensional data of the computer room; S2: Obtain the operating data and environmental data of all equipment in the computer room, construct a power map and an environmental cloud map of the computer room, and combine the power map and the environmental cloud map of the computer room into a static three-dimensional model to obtain a dynamic three-dimensional model; S3: Based on the dynamic 3D model, a visualization platform is built using WebGL, allowing personnel to view and manage all equipment in the computer room from a 3D perspective; S4: Based on the operation data of the equipment, detect abnormal data nodes through the anomaly detection model, and mark and report the abnormal data nodes in the dynamic three-dimensional model; The anomaly detection model is constructed by combining the autoencoder and OC-SVM, as follows: Collect multi-dimensional data during equipment operation, standardize the data, extract relevant features, and construct a training data set; The autoencoder consists of an encoder and a decoder, which are used for dimensionality reduction and reconstruction of data: Encoder: maps the training dataset data x to a low-dimensional latent space z; ； in, W enc and b enc are the weight matrix and bias vector of the encoder, respectively. is the activation function; Decoder: reconstructs the input data from the latent space z; ； in, w dec and b dec are the weight matrix and bias vector of the decoder respectively; The difference between the original input and the reconstructed output is minimized by the reconstruction loss function: ； Where n represents the number of data samples, It is input samples, It is Reconstructed output The latent space feature z output by the encoder is used to train the OC-SVM model, using the RBF kernel to account for more nonlinear distributions: ； in, is the kernel function, which calculates the similarity between z and z′ in the feature space; γ is the parameter in the kernel function that controls the distribution expansion; z′ is the latent space feature that is different from z; OC-SVM objective optimization function: find a hyperplane to mark ν% of samples as normal rather than abnormal: ； Among them, w is the weight vector of the decision function; For the Slack variables; Bias; Real-time data is passed through the autoencoder to obtain low-dimensional features z d , use OC-SVM to determine whether it is abnormal: If the OC-SVM decision function f(z d )<0 is marked as abnormal: ； in, is the feature mapping function; represents the inner product operation; During the training process, the structure of the autoencoder and the parameters of the OC-SVM are adjusted through hyperparameter optimization; S5: Combine Wi-Fi and UWB tag technology to achieve accurate positioning of personnel in the computer room, and conduct refined permission management based on dynamic 3D models through role authorization and real-time location monitoring; S6: Based on the real-time positioning of personnel and abnormal data nodes, taking into account authority management, the nearest suitable personnel are selected, and the optimal path is automatically generated for the personnel in combination with the path planning algorithm. AR navigation function is provided on mobile devices to assist personnel to reach the designated task area efficiently.

2. According to the artificial intelligence-based smart computer room management and control method of claim 1, it is characterized in that: The S1 is specifically: Use high-precision LiDAR equipment to perform multi-angle scanning to capture detailed point cloud data of every corner and equipment in the machine room; Based on SOR filtering, noise points of point cloud data are removed to obtain filtered multi-angle scanning point cloud data; Use the ICP algorithm to stitch and globally align the filtered multi-angle scanning point cloud data so that all scanning data are merged into a unified point cloud; The merged point cloud data is imported into the 3D modeling software as the basis for modeling, and the surface reconstruction technology is used to convert the point cloud into a visual 3D model.

3. The method for controlling a smart computer room based on artificial intelligence according to claim 2 is characterized in that: The SOR filtering process is used to remove noise points in the point cloud data to obtain filtered multi-angle scanning point cloud data, as follows: Initialization parameters, number of neighborhood points k and standard deviation multiplier threshold σ; For each point p in the point cloud i , calculate the Euclidean distance between it and other points in its neighborhood, and calculate the average neighborhood distance : ； in, p j for p i Any point in the neighborhood; k is the number of neighborhood points; Calculate the mean and standard deviation of the neighborhood distances of all points , using the standard deviation multiplier Define the threshold to filter out outliers. For each point p in the point cloud i , the difference between the neighborhood average distance and the global average distance exceeds , then the point is marked as an outlier, that is, a possible noise point.

4. The method for controlling a smart computer room based on artificial intelligence according to claim 3 is characterized in that: The ICP algorithm is used to stitch and globally align the filtered multi-angle scanning point cloud data so that all scanning data are merged into a unified point cloud, as follows: Obtain initial alignment based on the initial positioning data of the high-precision LiDAR device; For each point in the point cloud P , the number of points in point cloud P is N, find the nearest point in point cloud Q , forming a point pair ; Using singular value decomposition, the optimal rigid body transformation is calculated based on the nearest point pair, including the rotation matrix R and the displacement vector t, so that the square error of the two sets of points is minimized: ； Update the point cloud P: ； in, The updated point ; Calculate the matching error of the current iteration. If the error is lower than the preset threshold or reaches the maximum number of iterations, terminate the iteration. Multiple aligned point clouds are fused to generate unified point cloud data.

5. The method for controlling a smart computer room based on artificial intelligence according to claim 1, characterized in that: The S2 is specifically: Connect various sensors based on the industrial Internet of Things to obtain real-time operation data and environmental data of equipment; Use a database to store time series data collected at a high frequency and standardize the data to match the corresponding dimensional units in the 3D model; Integrate equipment operating parameter data, draw power diagrams representing equipment power consumption, efficiency or status changes, and draw environmental cloud diagrams, including temperature distribution diagrams and high humidity area signs; Import the static 3D model of the computer room into the 3D modeling software, add data interfaces and visualization components for each device and monitoring point in the model; capture and map real-time data to the static 3D model through the data interface; the visualization component defines the mapping function from data values ​​to visualization expressions, and uses dynamic texture mapping to display power diagrams and environmental cloud maps on the model surface.

6. The method for controlling a smart computer room based on artificial intelligence according to claim 1, characterized in that: The structure of the autoencoder and the parameters of the OC-SVM are adjusted by hyperparameter optimization, as follows: Define autoencoder using neural network library. The autoencoder architecture includes input layer, several hidden layers, bottleneck layer, and decoder. Hyperparameters include the number of encoder layers and units, and the learning rate: Under each hyperparameter combination, a preset number of training iterations are performed to enable the autoencoder to achieve the reconstruction target; Use the trained autoencoder to obtain low-dimensional latent features of the input data; The support vector machine library is used to read the latent features output by the autoencoder for OC-SVM training. The hyperparameter combination is adjusted using Bayesian optimization, and the hyperparameter combination with the highest performance index is selected as the final model parameters.

7. The method for controlling a smart computer room based on artificial intelligence according to claim 6, characterized in that: The Bayesian optimization is used to adjust the hyperparameter combination as follows: Randomly select or use Latin hypercube sampling to select several hyperparameter combinations for initial evaluation in the hyperparameter search space; Use the initial sampling points to build a GP model and fit the objective function; Update the mean and covariance matrix of the GP agent model using all current data samples; Using expected improvement EI as the acquisition function, it is defined as follows: ； in, is the best target value currently observed, and further we get: in, is the predicted mean of the Gaussian process surrogate model at point x; The standard deviation of the Gaussian process prediction at point x; is the best observed in the current search process; To explore the parameters; is the value of the standard normal cumulative distribution function; is the value of the standard normal probability density function; Z is the standardization parameter; Maximize the acquisition function EI(x) to select the next hyperparameter combination for model evaluation; Train the model with the selected hyperparameter combination, record and calculate the objective function value; Termination of the optimization process based on the maximum number of iterations and patience parameter criteria; The hyperparameters with the highest objective function value are returned as the final best combination.

8. The method for controlling a smart computer room based on artificial intelligence according to claim 1, characterized in that: The combination of Wi-Fi and UWB tag technology can achieve accurate positioning of personnel in the computer room, specifically: The RSSI of Wi-Fi was measured at different fixed locations in the equipment room. A model between signal strength and physical distance was established through experiments, and the path loss model was used to convert signal strength into distance. Deploy multiple UWB anchor points and use ToF measurements to convert signal round-trip time into distance; When personnel enter and leave the computer room, they carry UWB tags, and anchor points are deployed at preset locations according to the computing area; first Use the existing Wi-Fi access points in the computer room to make a rough location estimate using RSSI; Collect UWB ToF data and Wi-Fi RSSI data; The fused data is input into the Kalman filter for integration and synchronized using the current timestamp; Convert signal strength data into preliminary distance estimates; Synchronize the information provided by UWB precise ranging and RSSI into a unified coordinate system.

9. The method for controlling a smart computer room based on artificial intelligence according to claim 1, characterized in that: The S6 is specifically: Query the role permission database to determine which personnel have the authority to handle specific abnormal nodes, and use the following formula to verify permissions: ； in, is the verification formula; Indicates the authority of person u, Is an abnormal node Required permissions; is an empty set; From the authorized personnel, select the personnel who are closest to the abnormal node and have the current available status, and use the Euclidean distance calculation to select the nearest personnel; use Path planning algorithm, based on the current personnel location and abnormal node location, finds the optimal path; Integrate an AR engine in mobile devices to overlay navigation instructions on the camera view; Based on the path planning results, step-by-step navigation prompts are generated, and indicator arrows and information markers are displayed in the AR view; the positioning information and navigation path are updated at any time, and the navigation route is dynamically adjusted according to the actions of the personnel.

Images