Machine room equipment state monitoring method and device based on multiple parameters

Abstract

The invention discloses a machine room equipment state monitoring method and device based on multiple parameters, and relates to the technical field of equipment state monitoring. Acquiring multi-parameter data of each device in the machine room, and inputting the multi-parameter data into the device monitoring model for state monitoring to obtain state data; infrared images of all devices are obtained through an infrared sensor, and feature extraction is carried out on the infrared images to obtain temperature data; and performing state evaluation on each device in the machine room according to the temperature data and the state data, and if an abnormal device proportion of each device in the machine room is greater than a preset proportion, determining that the device in the machine room is abnormal. By fusing multi-parameter data and utilizing the LSTM network, the machine room equipment monitoring and operation and maintenance efficiency is improved, the system triggers infrared monitoring when the abnormal equipment proportion exceeds a threshold value, overheated equipment is accurately positioned, the misjudgment risk is reduced, after heat distribution characteristics are quantified, the equipment load and refrigeration intensity are dynamically adjusted, the anomaly detection accuracy is improved, the data processing cost is reduced, and the system is suitable for large-scale popularization and application. Energy consumption management is optimized, and the service life of hardware is prolonged.

Description

Technical Field

[0001] The present invention belongs to the technical field of equipment status monitoring, and in particular relates to a method and device for monitoring the equipment status of a computer room based on multiple parameters. Background Art

[0002] The equipment in the computer room is developing towards intelligence and integration. By introducing advanced sensor technology, the Internet of Things (IoT) and artificial intelligence (AI), real-time monitoring of equipment status and predictive maintenance can be achieved.

[0003] Patent number CN118521556A discloses a computer room equipment status monitoring method and system based on image recognition. The system collects and preprocesses images of the equipment's operating status. The system compares the processed images with pre-stored reference images to identify changes in the status of each U unit in the cabinet, changes in equipment status, and equipment labels. The system then outputs the identified photos and equipment label information. The system uses the SIFT algorithm to identify recognition points in the image to be tested, compares them to a standard image, and uses the Euclidean distance between the key feature vectors in the two images to determine whether the equipment's operating status is abnormal.

[0004] In the existing technology, the process of detecting the status of equipment in the computer room is prone to data silos, large data volumes, false alarms and missed reports, and the equipment status monitoring lacks accuracy and efficiency, resulting in an abnormal increase in management energy consumption and operation and maintenance costs. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of data islands, large data volume, false alarms and missed alarms in the process of equipment status detection in the computer room, lack of accuracy and monitoring efficiency of equipment status monitoring, and abnormal increase in management energy consumption and operation and maintenance costs, and propose a computer room equipment status monitoring method and device based on multiple parameters.

[0006] In a first aspect of the present invention, a method for monitoring the status of equipment in a computer room based on multiple parameters is first proposed, the method comprising:

[0007] Obtain multi-parameter data of each device in the computer room, and input the multi-parameter data into the device monitoring model for status monitoring to obtain status data; the multi-parameter data includes: environmental parameters, power parameters and device parameters; the status data is the power consumption and resource occupancy rate of the device;

[0008] Acquire infrared images of each device through an infrared sensor, and perform feature extraction on the infrared image to obtain temperature data;

[0009] The status of each device in the computer room is evaluated according to the temperature data and the status data. If the abnormal device ratio of each device in the computer room is greater than a preset ratio, it is determined that the device in the computer room is abnormal.

[0010] Optionally, performing feature extraction on the infrared image to obtain temperature data includes:

[0011] Obtaining visible light images of each device, and inputting the visible light images and the infrared images into the improved YOLOv8 to extract temperature data;

[0012] The improvements to the improved YOLOv8 model include:

[0013] An improved YOLOv8 model is obtained by adding an adaptive weight module between the zeroth and first layers of the YOLOv8 model, replacing the C2f modules in the fifteenth and twenty-first layers with spatial attention modules, replacing the C2f modules in the twelfth and eighteenth layers with VoVGSCSP modules, and replacing the Conv modules in the third, fifth, and seventh layers with Axconv modules.

[0014] The optional adaptive weight module works as follows:

[0015] Inputting the visible light image into a 3×3 convolutional layer for convolution to obtain a first feature map, and inputting the first feature into a 3×3 convolutional layer and a 5×5 convolutional layer for convolution operations to obtain a first feature map and a second feature map;

[0016] Adding the first feature map and the second feature map to obtain a third feature map, performing a global average pooling operation on the third feature map to obtain a fourth feature map, multiplying the fourth feature map by the first eigenvector and the second feature map respectively to obtain a fifth feature map and a sixth feature map, and adding the fifth feature map and the sixth feature map to obtain a seventh feature map;

[0017] Determine the maximum temperature T in the area according to the infrared image max and the regional minimum temperature T min , a temperature gradient ΔT is calculated based on the maximum temperature and the minimum temperature of the region, and a visible light weight and an infrared weight are calculated based on the temperature gradient ΔT; the maximum temperature and the minimum temperature of the region are respectively the average values of the temperature intervals in which the temperature area in the infrared image exceeds a preset ratio;

[0018] Visible light weight: W Vis =(ΔT max -ΔT) / (ΔT max -ΔT min );

[0019] Infrared weight: W IR =(ΔT-ΔT min ) / (ΔT max-ΔT min );

[0020] Among them, W Vis is the visible light weight, W IR is the infrared weight, ΔT max is the preset maximum temperature difference, ΔT min is the preset minimum temperature difference, ΔT is the temperature gradient;

[0021] The visible light image and the seventh feature map are multiplied according to the visible light weight to obtain a first fused image, the seventh feature map and the infrared image are multiplied according to the infrared weight to obtain a second fused image, and the first fused image and the second fused image are added to obtain a third fused image.

[0022] Optional, the working principle of the spatial attention module, including:

[0023] Inputting the target image into the residual attention module to obtain a first target image, performing a width global average pooling operation and a height global average pooling operation on the first target image to obtain a width enhanced image and a height enhanced image;

[0024] splicing the width-enhanced image and the height-enhanced image to obtain a second target image, convolving the second target image through a 1×1 convolutional layer to obtain a third target image, and performing a batch normalization operation on the third target image to obtain a fourth target image;

[0025] The fourth target image is input into a 1×1 convolutional layer for convolution operation to obtain a fifth target image, and the fifth target image is activated by Sigmoid to obtain a sixth target image. The target image and the sixth target image are element-wise multiplied to obtain an enhanced target image.

[0026] Optionally, performing a status assessment on each device in the computer room based on the temperature data and the status data includes:

[0027] determining a target power consumption and a target resource occupancy rate of the target device per unit time based on the status data, and determining that the target device is an abnormal device if the temperature change of the target device per unit time is less than a temperature drop threshold, the target power consumption is greater than a power consumption threshold, and the target resource occupancy rate is greater than a resource occupancy rate threshold;

[0028] If the ratio of abnormal devices in the equipment room is greater than a preset ratio, it is determined that the equipment in the equipment room is abnormal. The preset ratio includes a first preset ratio and a second preset ratio, including:

[0029] Determine the ratio of abnormal devices in the computer room. If the first preset ratio is less than the abnormal device ratio and less than the second preset ratio, then perform resource computing power scheduling on each device in the computer room.

[0030] If the second preset ratio is less than or equal to the abnormal equipment ratio, the temperature in the computer room will be lowered and a fault warning will be issued.

[0031] In a second aspect of the present invention, a multi-parameter-based equipment status monitoring device for a computer room is proposed, comprising: a data acquisition module, a feature extraction module, and an abnormality judgment module:

[0032] The data acquisition module is used to obtain multi-parameter data of each device in the computer room, and input the multi-parameter data into the device monitoring model for status monitoring to obtain status data; the multi-parameter data includes: environmental parameters, power parameters and device parameters; the status data is the power consumption and resource utilization rate of the device;

[0033] The feature extraction module is used to obtain infrared images of each device through an infrared sensor, and perform feature extraction on the infrared image to obtain temperature data;

[0034] The abnormality judgment module is used to evaluate the status of each device in the computer room according to the temperature data and the status data, and determine that the equipment in the computer room is abnormal if the abnormal device ratio of each device in the computer room is greater than a preset ratio.

[0035] Optionally, the feature extraction module is further configured to: obtain a visible light image of each device, and input the visible light image and the infrared image into an improved YOLOv8 to extract temperature data;

[0036] The improvements to the improved YOLOv8 model include:

[0037] An improved YOLOv8 model is obtained by adding an adaptive weight module between the zeroth and first layers of the YOLOv8 model, replacing the C2f modules in the fifteenth and twenty-first layers with spatial attention modules, replacing the C2f modules in the twelfth and eighteenth layers with VoVGSCSP modules, and replacing the Conv modules in the third, fifth, and seventh layers with Axconv modules.

[0038] The optional adaptive weight module works as follows:

[0039] Inputting the visible light image into a 3×3 convolutional layer for convolution to obtain a first feature map, and inputting the first feature into a 3×3 convolutional layer and a 5×5 convolutional layer for convolution operations to obtain a first feature map and a second feature map;

[0040] Adding the first feature map and the second feature map to obtain a third feature map, performing a global average pooling operation on the third feature map to obtain a fourth feature map, multiplying the fourth feature map by the first eigenvector and the second feature map respectively to obtain a fifth feature map and a sixth feature map, and adding the fifth feature map and the sixth feature map to obtain a seventh feature map;

[0041] Determine the maximum temperature T in the area according to the infrared image max and the regional minimum temperature T min , a temperature gradient ΔT is calculated based on the maximum temperature and the minimum temperature of the region, and a visible light weight and an infrared weight are calculated based on the temperature gradient ΔT; the maximum temperature and the minimum temperature of the region are respectively the average values of the temperature intervals in which the temperature area in the infrared image exceeds a preset ratio;

[0042] Visible light weight: W Vis =(ΔT max -ΔT) / (ΔT max -ΔT min );

[0043] Infrared weight: W IR =(ΔT-ΔT min ) / (ΔT max -ΔT min );

[0044] Among them, W Vis is the visible light weight, W IR is the infrared weight, ΔT max is the preset maximum temperature difference, ΔT min is the preset minimum temperature difference, ΔT is the temperature gradient;

[0045] The visible light image and the seventh feature map are multiplied according to the visible light weight to obtain a first fused image, the seventh feature map and the infrared image are multiplied according to the infrared weight to obtain a second fused image, and the first fused image and the second fused image are added to obtain a third fused image.

[0046] Optional, the working principle of the spatial attention module, including:

[0047] Inputting the target image into the residual attention module to obtain a first target image, performing a width global average pooling operation and a height global average pooling operation on the first target image to obtain a width enhanced image and a height enhanced image;

[0048] splicing the width-enhanced image and the height-enhanced image to obtain a second target image, convolving the second target image through a 1×1 convolutional layer to obtain a third target image, and performing a batch normalization operation on the third target image to obtain a fourth target image;

[0049] The fourth target image is input into a 1×1 convolutional layer for convolution operation to obtain a fifth target image, and the fifth target image is activated by Sigmoid to obtain a sixth target image. The target image and the sixth target image are element-wise multiplied to obtain an enhanced target image.

[0050] Optionally, the abnormality judgment module is configured to: determine a target power consumption and a target resource occupancy rate of the target device within a unit time according to the status data; if a temperature change of the target device within a unit time is less than a temperature drop threshold, the target power consumption is greater than a power consumption threshold, and the target resource occupancy rate is greater than a resource occupancy rate threshold, then determine that the target device is an abnormal device;

[0051] The abnormality judgment module further includes: a first judgment module and a second judgment module:

[0052] The first judgment module is used to determine the ratio of abnormal devices in the computer room, and if the first preset ratio is less than the abnormal device ratio and less than the second preset ratio, then perform resource computing power scheduling on each device in the computer room;

[0053] The second judgment module is configured to cool the computer room and issue a fault warning if the second preset ratio is less than or equal to the abnormal equipment ratio.

[0054] Beneficial effects of the present invention:

[0055] This paper proposes a multi-parameter computer room equipment status monitoring method. The method obtains multi-parameter data from each device in the computer room and inputs this data into a device monitoring model for status monitoring to obtain status data. Infrared images of each device are acquired using infrared sensors, and feature extraction is performed on the infrared images to obtain temperature data. The status of each device in the computer room is evaluated based on the temperature and status data. If the proportion of abnormal devices in the computer room exceeds a preset ratio, the device in the computer room is deemed abnormal. By integrating multi-parameter data and utilizing an LSTM network, the efficiency of computer room equipment monitoring and operation and maintenance is improved. The system triggers infrared monitoring when the proportion of abnormal devices exceeds a threshold, accurately locating overheating devices and reducing the risk of misjudgment. After quantifying the thermal distribution characteristics, the system dynamically adjusts the device load and cooling intensity to improve anomaly detection accuracy, reduce data processing costs, optimize energy management, and extend hardware life. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present invention will be further described below with reference to the accompanying drawings.

[0057] Figure 1 A flowchart of a method for monitoring the status of equipment in a computer room based on multiple parameters provided by an embodiment of the present invention;

[0058] Figure 2 A schematic structural diagram of a multi-parameter-based computer room equipment status monitoring device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments represent only a portion of the embodiments of the present invention, not all of them. The term "and / or" herein simply describes an association relationship between associated objects, indicating that three possible relationships exist. For example, "A" and "B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, references to "first," "second," and so on in the present invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include at least one of these features. Furthermore, the technical solutions of the various embodiments may be combined, but only if they are achievable by a person of ordinary skill in the art. If a combination of technical solutions contradicts or is unachievable, such combination shall be deemed non-existent and outside the scope of protection claimed by the present invention.

[0060] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0061] The embodiment of the present invention provides a method for monitoring the status of equipment in a computer room based on multiple parameters. Figure 1 , Figure 1 A flowchart of a method for monitoring the status of equipment in a computer room based on multiple parameters is provided in an embodiment of the present invention. The method comprises the following steps:

[0062] S101, obtaining multi-parameter data of each device in the computer room, inputting the multi-parameter data into the device monitoring model to perform status monitoring to obtain status data;

[0063] S102, acquiring infrared images of each device through an infrared sensor, and extracting features from the infrared images to obtain temperature data;

[0064] S103, evaluating the status of each device in the computer room according to the temperature data and the status data. If the abnormal device ratio of each device in the computer room is greater than a preset ratio, it is determined that the device in the computer room is abnormal.

[0065] Multi-parameter data includes: environmental parameters, power parameters and equipment parameters; status data includes equipment power consumption and resource utilization;

[0066] A multi-parameter computer room equipment status monitoring method provided by an embodiment of the present invention improves the efficiency of computer room equipment monitoring and operation and maintenance by fusing multi-parameter data and utilizing an LSTM network. The system triggers infrared monitoring when the proportion of abnormal equipment exceeds a threshold, accurately locates overheating equipment, reduces the risk of misjudgment, and dynamically adjusts equipment load and cooling intensity after quantifying heat distribution characteristics to improve anomaly detection accuracy, reduce data processing costs, optimize energy consumption management, and extend hardware life.

[0067] In one implementation, during equipment status monitoring in a computer room, multi-parameter data, including environmental parameters (specific heat capacity of the environment within the computer room, i.e., specific heat capacity of air, e.g., 1.006 joules / gram·degree Celsius (J / g·°C)), power parameters (power consumption of equipment), and equipment parameters (resource utilization, e.g., CPU utilization and internal layer utilization), is acquired and fed into an equipment monitoring model (e.g., a long-short-term memory network model). This allows for comprehensive monitoring of equipment status within the computer room. This multi-dimensional data monitoring approach provides more accurate equipment status information, helping to promptly identify potential problems and anomalies.

[0068] In one implementation, in the training process of a device monitoring model (e.g., a long short-term memory network model), training data is input into the long short-term memory network model for training to obtain model update parameters, and the weight matrix and bias in the input gate, output gate, and forget gate of the long short-term memory network model are updated according to the model update parameters to obtain a device monitoring model; the device monitoring model is used to learn the power consumption of each device in the computer room and the resource occupancy rate of the device, and the output of the model is the power consumption and resource occupancy rate of the device in the future, that is, the status data; the heat dissipation capacity of the current indoor air is determined according to the environmental parameters, and when the proportion of abnormal devices is greater than the preset proportion, the device status is determined to be abnormal, and infrared images of each device are obtained through infrared sensors for feature extraction (determining the actual temperature of each device); the status of each device in the computer room is evaluated according to the temperature data to determine the heat dissipation capacity of each device, and when the heat dissipation capacity of the device cannot meet the requirements, early warning, resource computing power scheduling, and cooling system (air conditioning) operations are performed.

[0069] In one implementation, when the proportion of devices with abnormal status detected in the computer room exceeds a preset proportion, the system can automatically determine that the equipment in the computer room is abnormal. This judgment mechanism enables the computer room management to respond quickly to possible equipment failures, reduce business interruptions and service interruptions caused by equipment failures, and improve the reliability of computer room operations.

[0070] In one implementation, the status of each device in the computer room is evaluated based on temperature data extracted from infrared images, which can more accurately determine whether the equipment is in an abnormal state. If the proportion of devices with abnormal status in the computer room is greater than a preset proportion, that is, there are 100 devices in the computer room, and 60 of them are abnormal, the proportion of abnormal devices at this time is 60%, and the preset proportion is 50%, then the equipment in the current computer room is determined to be abnormal. The status assessment method based on temperature data can provide a scientific basis for computer room management, help achieve refined management, and reduce unnecessary maintenance costs.

[0071] In one embodiment, extracting features from infrared images to obtain temperature data includes:

[0072] Obtain visible light images of each device, and input the visible light images and infrared images into the improved YOLOv8 to extract temperature data;

[0073] Improvements to the YOLOv8 model include:

[0074] An improved YOLOv8 model is obtained by adding an adaptive weight module between the zeroth and first layers of the YOLOv8 model, replacing the C2f modules in the fifteenth and twenty-first layers with spatial attention modules, replacing the C2f modules in the twelfth and eighteenth layers with VoVGSCSP modules, and replacing the Conv modules in the third, fifth, and seventh layers with Axconv modules.

[0075] In one implementation, by introducing the adaptive weight module, spatial attention module, VoVGSCSP module, and Axconv module into the YOLOv8 model, the improved model can more effectively extract and process temperature data from visible light and infrared images. These improvements enhance the model's ability to recognize image features and improve the accuracy and robustness of object detection.

[0076] In one implementation, an adaptive weight module is introduced between the zeroth and first layers of the model to dynamically adjust the weights based on the features of the input image. This module helps the model better adapt to different types of image data (visible light and infrared), thereby improving the accuracy and efficiency of feature extraction.

[0077] In one implementation, the spatial attention module replaces the C2f module in the 15th and 21st layers of the model. This module enhances the model's focus on spatial information in the image, allowing it to focus more on key areas within the image, thereby improving object detection performance. The spatial attention mechanism helps the model more accurately identify objects in complex backgrounds.

[0078] In one implementation, the Axconv module is a lightweight convolution module designed to improve resource efficiency. This module achieves model lightweighting by replacing traditional fixed-size convolution kernels with kernels of varying sizes, significantly reducing the number of parameters and computational requirements while maintaining optimal detection performance. The Axconv module first uses a 1×1 convolution kernel for dimensionality reduction to reduce parameters and computational complexity. It then uses a 3×3 convolution kernel to convolve the reduced feature map to extract richer feature information. Finally, a 1×1 convolution kernel is used to restore the original dimension of the feature map and enhance its features, thereby reducing model complexity while maintaining performance.

[0079] In one implementation, the VoVGSCSP module is an image processing module that enhances the model's learning capabilities by employing GSConv (lightweight convolution) and GSBottleneck (bottleneck structure), and designs cross-layer partial networks through a one-shot aggregation method. The design goal of this module is to reduce computational complexity and network structure while maintaining sufficient accuracy. The VoVGSCSP module replaces the Cf2 module at the neck of the model, using lightweight convolution (GSConv) instead of standard convolution, thereby reducing computational cost. At the same time, GSConv also provides model learning capabilities comparable to standard convolution. The introduction of the VoVGSCSP module can reduce model complexity without compromising recognition accuracy.

[0080] In one implementation, the aforementioned improvements enable the YOLOv8 model to more effectively extract temperature data when processing visible light and infrared images. These improvements enable the model to demonstrate greater accuracy and robustness in object detection tasks, particularly when processing multimodal image data. The improved model is better suited to diverse application scenarios, such as security monitoring and equipment fault detection, providing more reliable detection results.

[0081] In one embodiment, the working principle of the adaptive weight module includes:

[0082] The visible light image is input into the 3×3 convolution layer for convolution to obtain the first feature map, and the first feature is input into the 3×3 convolution layer and the 5×5 convolution layer for convolution operations to obtain the first feature map and the second feature map;

[0083] The first feature map and the second feature map are added to obtain a third feature map, the third feature map is subjected to a global average pooling operation to obtain a fourth feature map, the fourth feature map is multiplied by the first feature vector and the second feature map respectively to obtain a fifth feature map and a sixth feature map, and the fifth feature map and the sixth feature map are added to obtain a seventh feature map;

[0084] Determine the maximum temperature T in the area based on the infrared imagemax and the regional minimum temperature T min , the temperature gradient ΔT is calculated based on the regional maximum temperature and the regional minimum temperature, and the visible light weight and infrared weight are calculated based on the temperature gradient ΔT; the regional maximum temperature and the regional minimum temperature are the average values of the temperature intervals in the infrared image whose temperature area accounts for more than a preset ratio;

[0085] Visible light weight: W Vis =(ΔT max -ΔT) / (ΔT max -ΔT min );

[0086] Infrared weight: W IR =(ΔT-ΔT min ) / (ΔT max -ΔT min );

[0087] Among them, W Vis is the visible light weight, W IR is the infrared weight, ΔT max is the preset maximum temperature difference, ΔT min is the preset minimum temperature difference, ΔT is the temperature gradient;

[0088] The visible light image and the seventh feature map are multiplied according to the visible light weight to obtain a first fused image, the seventh feature map and the infrared image are multiplied according to the infrared weight to obtain a second fused image, and the first fused image and the second fused image are added to obtain a third fused image.

[0089] In one implementation, the temperature gradient ΔT = T max -T min The highest temperature in the region and the lowest temperature in the region are the average values of the temperature intervals in which the temperature area in the infrared image exceeds the preset ratio. For example, the highest temperature in the infrared image is 65.9°C (in the temperature interval [65.0-66.0)°C), but the temperature area in the temperature interval [65.0-66.0)°C does not exceed the preset ratio, so it is not the highest temperature. Only when the area in the temperature interval exceeds the preset ratio, the temperature interval average value in the above temperature interval is calculated (temperature average value calculation is performed by random sampling in the interval).

[0090] In one implementation, by inputting the visible light image into a 3×3 convolutional layer and a 5×5 convolutional layer for convolution operations, feature maps at different scales can be obtained. This multi-scale feature extraction method can capture the details and texture information of the image at different resolutions, enhance the richness of the feature representation, and help improve the performance of subsequent image processing tasks; through the global average pooling operation, the spatial dimension of the third feature map can be compressed to obtain the fourth feature map; global average pooling helps to reduce the size of the feature map while retaining important feature information, providing a compact representation for subsequent feature map multiplication and fusion.

[0091] In one implementation, the fourth feature map is multiplied with the first feature vector and the second feature map respectively to obtain the fifth and sixth feature maps. This feature map enhancement method utilizes the correlation between feature maps to further highlight important features and provide richer information for subsequent feature fusion.

[0092] In one implementation, the introduction of adaptive weights (infrared and visible light) allows the image fusion process to dynamically adjust the contributions of visible light (the room's lighting) and infrared images based on temperature fluctuations. When the infrared weight is high, the fused image relies more on the infrared image, and temperature fluctuations have less impact on feature extraction of equipment within the room. Conversely, when the visible light weight is high, the fused image relies more on the visible light image, and temperature fluctuations have a greater impact on feature extraction. This adaptive adjustment mechanism facilitates effective monitoring of equipment within the room under varying temperature fluctuations.

[0093] In one implementation, by calculating the temperature gradient ΔT in the infrared image and dynamically adjusting the fusion weights of the visible light and infrared images accordingly, namely the visible light weight WVis and the infrared weight WIR, this method can intelligently perform differentiated processing on different areas within the image. In areas where the temperature difference is not obvious, by reducing the weight of the infrared image, its impact on the fused image can be reduced, thereby enhancing the distinction between the temperature characteristics of the frame and equipment in the computer room. This method of adaptively adjusting the fusion ratio based on temperature difference not only improves the visibility of temperature characteristics, but also helps to more accurately monitor and evaluate the status of each device in the computer room, especially in scenarios where the temperature changes are subtle or the temperature difference is not large, it can improve the accuracy and reliability of fault detection (temperature monitoring).

[0094] In one embodiment, the working principle of the spatial attention module includes:

[0095] Input the target image into the residual attention module to obtain a first target image, and perform width global average pooling operation and height global average pooling operation on the first target image to obtain a width enhanced image and a height enhanced image;

[0096] The width-enhanced image and the height-enhanced image are concatenated to obtain a second target image, the second target image is convolved with a 1×1 convolutional layer to obtain a third target image, and the third target image is batch normalized to obtain a fourth target image.

[0097] The fourth target image is input into a 1×1 convolutional layer for convolution operation to obtain a fifth target image, and the fifth target image is activated by Sigmoid to obtain a sixth target image. The target image and the sixth target image are element-wise multiplied to obtain an enhanced target image.

[0098] In one implementation, the low-level features of the original image are retained through residual connections to avoid the gradient vanishing problem, while allowing the attention module to focus on learning residual features, providing a multi-scale feature basis for subsequent pooling operations, and enhancing the module's compatibility with deep and shallow information; the first target image (C×H×W, i.e., C: number of channels, H: height, and W: width) is subjected to width global average pooling operations and height global average pooling operations respectively to obtain a width-enhanced image (C×1×W, such as the position of a long strip target) and a height-enhanced image (C×H×1, such as the key area of a vertical structure), and global average pooling is performed on the first target image along the width dimension and height dimension respectively to compress the height dimension and width dimension, and extract the global spatial features of each row and column.

[0099] In one implementation, the width and height global features are concatenated along the spatial dimension (C×1×(H+W)) to provide directional complementary information for subsequent convolutions, enhancing the comprehensiveness of spatial attention modeling; 1×1 convolution: adjusts the channel dimension (such as dimensionality reduction or dimensionality increase), reduces redundant features and fuses cross-channel information (adjusts the channel dimension through lightweight convolution to reduce the amount of computation).

[0100] In one implementation, a batch normalization operation is performed on the third target image to obtain the fourth target image, and the convolved features are normalized (BatchNorm) to eliminate feature distribution offset and suppress abnormal gradient fluctuations during training; improve feature stability and avoid subsequent nonlinear activations (such as ReLU) from saturating or failing due to input dimension differences.

[0101] In one implementation, the sixth target image is multiplied element-by-element with the target image, thereby highlighting task-related details (such as classification targets or detection box key points) while maintaining the original feature space structure, thereby improving the model's judgment ability.

[0102] In one embodiment, the status of each device in the computer room is evaluated based on temperature data and status data, including:

[0103] Determine the target power consumption and target resource utilization rate of the target device per unit time based on the status data. If the temperature change of the target device per unit time is less than the temperature drop threshold, the target power consumption is greater than the power consumption threshold, and the target resource utilization rate is greater than the resource utilization rate threshold, then the target device is determined to be an abnormal device.

[0104] If the ratio of abnormal devices in the equipment room is greater than a preset ratio, the equipment in the equipment room is determined to be abnormal. The preset ratio includes a first preset ratio and a second preset ratio, including:

[0105] Determine the ratio of abnormal devices in the computer room. If the first preset ratio is less than the abnormal device ratio and less than the second preset ratio, then perform resource computing power scheduling on each device in the computer room.

[0106] If the second preset ratio is less than or equal to the abnormal equipment ratio, the temperature in the computer room will be lowered and a fault warning will be issued.

[0107] In one implementation, the output of the device monitoring model is the power consumption and resource occupancy rate of the device in a future period of time. The power consumption and resource occupancy rate per unit time in the future period of time are selected, and the future period of time > unit time; if the temperature change of the target device in unit time is < the temperature drop threshold, that is, the temperature change of the target device (the temperature drop value) is < the temperature drop threshold, for example: the original target device temperature is 75°C, and the temperature of the target device drops by 10°C (temperature change). At this time, the target device temperature is 65°C, and the temperature drop threshold is 20°C. Since 10°C < 20°C, it is determined that the temperature change is < the temperature drop threshold.

[0108] In one implementation, by analyzing the target device's power consumption, resource utilization, and temperature change per unit time, it is possible to accurately determine whether the device is in an abnormal state. This method can promptly identify devices that exceed power consumption and resource utilization thresholds despite a temperature drop, thereby preventing potential failures and performance issues. By determining the proportion of abnormal devices in the computer room, the overall health of the computer room can be assessed. If the proportion of abnormal devices is between a first preset ratio and a second preset ratio, the system will schedule resource computing power to optimize resource allocation and improve energy efficiency. If the proportion of abnormal devices reaches or exceeds the second preset ratio, cooling and fault warning measures are triggered to prevent the spread of the fault.

[0109] In one implementation, when the proportion of abnormal devices in the computer room is between two preset ratios, the system will schedule resource computing power for each device. This scheduling can balance the load, avoid resource waste, and ensure the stable operation of critical tasks, thereby improving the operating efficiency and reliability of the entire computer room; if the proportion of abnormal devices in the computer room reaches or exceeds the second preset ratio, the system will take cooling measures and issue a fault warning. This helps to quickly respond to possible equipment failures, reduce the risk of damage caused by equipment overheating, and protect other equipment in the computer room from the impact of failures; the equipment monitoring model can predict the power consumption and resource utilization of equipment in the future. This predictive capability allows computer room management to plan and adjust resource allocation in advance to cope with upcoming high load conditions, thereby achieving more effective energy management and fault prevention.

[0110] Based on the same inventive concept, the embodiment of the present invention also provides a device for monitoring the status of equipment in a computer room based on multiple parameters. Figure 2 , Figure 2 A schematic diagram of the structure of a multi-parameter computer room equipment status monitoring device provided by an embodiment of the present invention includes: a data acquisition module, a feature extraction module and an abnormality judgment module:

[0111] The data acquisition module is used to obtain multi-parameter data of each device in the computer room and input the multi-parameter data into the device monitoring model for status monitoring to obtain status data; the multi-parameter data includes: environmental parameters, power parameters and device parameters; the status data is the power consumption and resource utilization rate of the device;

[0112] The feature extraction module is used to obtain infrared images of each device through infrared sensors and extract features from the infrared images to obtain temperature data;

[0113] The abnormality judgment module is used to evaluate the status of each device in the computer room based on temperature data and status data. If the abnormal device ratio of each device in the computer room is greater than a preset ratio, the device in the computer room is judged to be abnormal.

[0114] A multi-parameter computer room equipment status monitoring device provided by an embodiment of the present invention improves the efficiency of computer room equipment monitoring and operation and maintenance by fusing multi-parameter data and utilizing an LSTM network. The system triggers infrared monitoring when the proportion of abnormal equipment exceeds a threshold, accurately locates overheating equipment, reduces the risk of misjudgment, and dynamically adjusts equipment load and cooling intensity after quantifying heat distribution characteristics to improve anomaly detection accuracy, reduce data processing costs, optimize energy consumption management, and extend hardware life.

[0115] In one embodiment, the feature extraction module is further configured to: obtain visible light images of each device, and input the visible light images and infrared images into the improved YOLOv8 to extract temperature data;

[0116] Improvements to the YOLOv8 model include:

[0117] An improved YOLOv8 model is obtained by adding an adaptive weight module between the zeroth and first layers of the YOLOv8 model, replacing the C2f modules in the fifteenth and twenty-first layers with spatial attention modules, replacing the C2f modules in the twelfth and eighteenth layers with VoVGSCSP modules, and replacing the Conv modules in the third, fifth, and seventh layers with Axconv modules.

[0118] In one embodiment, the working principle of the adaptive weight module includes:

[0119] The visible light image is input into the 3×3 convolution layer for convolution to obtain the first feature map, and the first feature is input into the 3×3 convolution layer and the 5×5 convolution layer for convolution operations to obtain the first feature map and the second feature map;

[0120] The first feature map and the second feature map are added to obtain a third feature map, the third feature map is subjected to a global average pooling operation to obtain a fourth feature map, the fourth feature map is multiplied by the first feature vector and the second feature map respectively to obtain a fifth feature map and a sixth feature map, and the fifth feature map and the sixth feature map are added to obtain a seventh feature map;

[0121] Determine the maximum temperature T in the area based on the infrared image max and the regional minimum temperature T min , the temperature gradient ΔT is calculated based on the regional maximum temperature and the regional minimum temperature, and the visible light weight and infrared weight are calculated based on the temperature gradient ΔT; the regional maximum temperature and the regional minimum temperature are the average values of the temperature intervals in the infrared image whose temperature area accounts for more than a preset ratio;

[0122] Visible light weight: W Vis =(ΔT max -ΔT) / (ΔT max -ΔT min );

[0123] Infrared weight: W IR =(ΔT-ΔT min ) / (ΔT max -ΔT min );

[0124] Among them, W Vis is the visible light weight, W IR is the infrared weight, ΔT max is the preset maximum temperature difference, ΔT min is the preset minimum temperature difference, ΔT is the temperature gradient;

[0125] The visible light image and the seventh feature map are multiplied according to the visible light weight to obtain a first fused image, the seventh feature map and the infrared image are multiplied according to the infrared weight to obtain a second fused image, and the first fused image and the second fused image are added to obtain a third fused image.

[0126] In one embodiment, the working principle of the spatial attention module includes:

[0127] Input the target image into the residual attention module to obtain a first target image, and perform width global average pooling operation and height global average pooling operation on the first target image to obtain a width enhanced image and a height enhanced image;

[0128] The width-enhanced image and the height-enhanced image are concatenated to obtain a second target image, the second target image is convolved with a 1×1 convolutional layer to obtain a third target image, and the third target image is batch normalized to obtain a fourth target image.

[0129] The fourth target image is input into a 1×1 convolutional layer for convolution operation to obtain a fifth target image, and the fifth target image is activated by Sigmoid to obtain a sixth target image. The target image and the sixth target image are element-wise multiplied to obtain an enhanced target image.

[0130] In one embodiment, the abnormality determination module is configured to: determine a target power consumption and a target resource occupancy rate of the target device per unit time based on the status data; if the temperature change of the target device per unit time is less than a temperature drop threshold, the target power consumption is greater than a power consumption threshold, and the target resource occupancy rate is greater than a resource occupancy threshold, then determine that the target device is an abnormal device;

[0131] The abnormality judgment module also includes: a first judgment module and a second judgment module:

[0132] The first judgment module is used to determine the ratio of abnormal devices in the computer room. If the first preset ratio is less than the abnormal device ratio and less than the second preset ratio, resource computing power scheduling is performed on each device in the computer room.

[0133] The second judgment module is used to cool down the computer room and issue a fault warning if the second preset ratio is less than or equal to the abnormal equipment ratio.

[0134] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for monitoring the status of equipment in a computer room based on multiple parameters, characterized in that: The method comprises: Obtain multi-parameter data of each device in the computer room, and input the multi-parameter data into the device monitoring model for status monitoring to obtain status data; the multi-parameter data includes: environmental parameters, power parameters and device parameters; the status data is the power consumption and resource occupancy rate of the device; Acquire infrared images of each device through an infrared sensor, and perform feature extraction on the infrared image to obtain temperature data; The status of each device in the computer room is evaluated according to the temperature data and the status data. If the abnormal device ratio of each device in the computer room is greater than a preset ratio, it is determined that the device in the computer room is abnormal.

2. The method for monitoring the status of equipment in a computer room based on multiple parameters according to claim 1, characterized in that: Extracting features from the infrared image to obtain temperature data includes: Obtaining visible light images of each device, and inputting the visible light images and the infrared images into the improved YOLOv8 to extract temperature data; The improvements to the improved YOLOv8 model include: An improved YOLOv8 model is obtained by adding an adaptive weight module between the zeroth and first layers of the YOLOv8 model, replacing the C2f modules in the fifteenth and twenty-first layers with spatial attention modules, replacing the C2f modules in the twelfth and eighteenth layers with VoVGSCSP modules, and replacing the Conv modules in the third, fifth, and seventh layers with Axconv modules.

3. The method for monitoring the status of equipment in a computer room based on multiple parameters according to claim 2, characterized in that: The working principle of the adaptive weight module includes: Inputting the visible light image into a 3×3 convolutional layer for convolution to obtain a first feature map, and inputting the first feature into a 3×3 convolutional layer and a 5×5 convolutional layer for convolution operations to obtain a first feature map and a second feature map; Adding the first feature map and the second feature map to obtain a third feature map, performing a global average pooling operation on the third feature map to obtain a fourth feature map, multiplying the fourth feature map by the first eigenvector and the second feature map respectively to obtain a fifth feature map and a sixth feature map, and adding the fifth feature map and the sixth feature map to obtain a seventh feature map; Determine the maximum temperature T in the area according to the infrared image max and the regional minimum temperature T min , a temperature gradient ΔT is calculated based on the maximum temperature and the minimum temperature of the region, and a visible light weight and an infrared weight are calculated based on the temperature gradient ΔT; the maximum temperature and the minimum temperature of the region are respectively the average values of the temperature intervals in which the temperature area in the infrared image exceeds a preset ratio; Visible light weight: W Vis =(ΔT max -ΔT) / (ΔT max -ΔT min ); Infrared weight: W IR =(ΔT-ΔT min ) / (ΔT max -ΔT min ); Among them, W Vis is the visible light weight, W IR is the infrared weight, ΔT max is the preset maximum temperature difference, ΔT min is the preset minimum temperature difference, ΔT is the temperature gradient; The visible light image and the seventh feature map are multiplied according to the visible light weight to obtain a first fused image, the seventh feature map and the infrared image are multiplied according to the infrared weight to obtain a second fused image, and the first fused image and the second fused image are added to obtain a third fused image.

4. The method for monitoring the status of equipment in a computer room based on multiple parameters according to claim 2, characterized in that: The working principle of the spatial attention module includes: Inputting the target image into the residual attention module to obtain a first target image, performing a width global average pooling operation and a height global average pooling operation on the first target image to obtain a width enhanced image and a height enhanced image; splicing the width-enhanced image and the height-enhanced image to obtain a second target image, convolving the second target image through a 1×1 convolutional layer to obtain a third target image, and performing a batch normalization operation on the third target image to obtain a fourth target image; The fourth target image is input into a 1×1 convolutional layer for convolution operation to obtain a fifth target image, and the fifth target image is activated by Sigmoid to obtain a sixth target image. The target image and the sixth target image are element-wise multiplied to obtain an enhanced target image.

5. The method for monitoring the status of equipment in a computer room based on multiple parameters according to claim 1, characterized in that: Performing a status assessment on each device in the computer room based on the temperature data and the status data includes: determining a target power consumption and a target resource occupancy rate of the target device per unit time based on the status data, and determining that the target device is an abnormal device if the temperature change of the target device per unit time is less than a temperature drop threshold, the target power consumption is greater than a power consumption threshold, and the target resource occupancy rate is greater than a resource occupancy rate threshold; If the ratio of abnormal devices in the equipment room is greater than a preset ratio, it is determined that the equipment in the equipment room is abnormal. The preset ratio includes a first preset ratio and a second preset ratio, including: Determine the ratio of abnormal devices in the computer room. If the first preset ratio is less than the abnormal device ratio and less than the second preset ratio, then perform resource computing power scheduling on each device in the computer room. If the second preset ratio is less than or equal to the abnormal equipment ratio, the temperature in the computer room will be lowered and a fault warning will be issued.

6. A multi-parameter-based equipment room status monitoring device, characterized in that: The device includes: a data acquisition module, a feature extraction module and an abnormality judgment module: The data acquisition module is used to obtain multi-parameter data of each device in the computer room and input the multi-parameter data into the device monitoring model for status monitoring to obtain status data; the multi-parameter data includes: environmental parameters, power parameters and device parameters; the status data is the power consumption and resource utilization rate of the device; The feature extraction module is used to obtain infrared images of each device through an infrared sensor, and perform feature extraction on the infrared image to obtain temperature data; The abnormality judgment module is used to evaluate the status of each device in the computer room according to the temperature data and the status data, and determine that the equipment in the computer room is abnormal if the abnormal device ratio of each device in the computer room is greater than a preset ratio.

7. The multi-parameter-based computer room equipment status monitoring device according to claim 6, characterized in that: The feature extraction module is further configured to: obtain visible light images of each device, and input the visible light images and the infrared images into the improved YOLOv8 to extract temperature data; The improvements to the improved YOLOv8 model include: An improved YOLOv8 model is obtained by adding an adaptive weight module between the zeroth and first layers of the YOLOv8 model, replacing the C2f modules in the fifteenth and twenty-first layers with spatial attention modules, replacing the C2f modules in the twelfth and eighteenth layers with VoVGSCSP modules, and replacing the Conv modules in the third, fifth, and seventh layers with Axconv modules.

8. The multi-parameter-based computer room equipment status monitoring device according to claim 7, characterized in that: The working principle of the adaptive weight module includes: Inputting the visible light image into a 3×3 convolutional layer for convolution to obtain a first feature map, and inputting the first feature into a 3×3 convolutional layer and a 5×5 convolutional layer for convolution operations to obtain a first feature map and a second feature map; Adding the first feature map and the second feature map to obtain a third feature map, performing a global average pooling operation on the third feature map to obtain a fourth feature map, multiplying the fourth feature map by the first eigenvector and the second feature map respectively to obtain a fifth feature map and a sixth feature map, and adding the fifth feature map and the sixth feature map to obtain a seventh feature map; Determine the maximum temperature T in the area according to the infrared image max and the regional minimum temperature T min , a temperature gradient ΔT is calculated based on the maximum temperature and the minimum temperature of the region, and a visible light weight and an infrared weight are calculated based on the temperature gradient ΔT; the maximum temperature and the minimum temperature of the region are respectively the average values of the temperature intervals in which the temperature area in the infrared image exceeds a preset ratio; Visible light weight: W Vis =(ΔT max -ΔT) / (ΔT max -ΔT min ); Infrared weight: W IR =(ΔT-ΔT min ) / (ΔT max -ΔT min ); Among them, W Vis is the visible light weight, W IR is the infrared weight, ΔT max is the preset maximum temperature difference, ΔT min is the preset minimum temperature difference, ΔT is the temperature gradient; The visible light image and the seventh feature map are multiplied according to the visible light weight to obtain a first fused image, the seventh feature map and the infrared image are multiplied according to the infrared weight to obtain a second fused image, and the first fused image and the second fused image are added to obtain a third fused image.

9. The multi-parameter-based computer room equipment status monitoring device according to claim 7, characterized in that: The working principle of the spatial attention module includes: Inputting the target image into the residual attention module to obtain a first target image, performing a width global average pooling operation and a height global average pooling operation on the first target image to obtain a width enhanced image and a height enhanced image; splicing the width-enhanced image and the height-enhanced image to obtain a second target image, convolving the second target image through a 1×1 convolutional layer to obtain a third target image, and performing a batch normalization operation on the third target image to obtain a fourth target image; The fourth target image is input into a 1×1 convolutional layer for convolution operation to obtain a fifth target image, and the fifth target image is activated by Sigmoid to obtain a sixth target image. The target image and the sixth target image are element-wise multiplied to obtain an enhanced target image.

10. The multi-parameter-based computer room equipment status monitoring device according to claim 6, characterized in that: The abnormality judgment module is configured to determine a target power consumption and a target resource occupancy rate of the target device within a unit time based on the status data, and determine that the target device is an abnormal device if the temperature change of the target device within a unit time is less than a temperature drop threshold, the target power consumption is greater than a power consumption threshold, and the target resource occupancy rate is greater than a resource occupancy rate threshold; The abnormality judgment module further includes: a first judgment module and a second judgment module: The first judgment module is used to determine the ratio of abnormal devices in the computer room, and if the first preset ratio is less than the abnormal device ratio and less than the second preset ratio, then perform resource computing power scheduling on each device in the computer room; The second judgment module is configured to cool the computer room and issue a fault warning if the second preset ratio is less than or equal to the abnormal equipment ratio.

Images