A data center machine room temperature monitoring method and system based on thermal imaging

Abstract

This invention relates to the field of data center monitoring technology, specifically a method and system for monitoring data center temperature based on thermal imaging. The method includes the following steps: acquiring a thermal imaging image of the data center; determining whether there is an overheating issue based on the binarized data of the thermal imaging image; if so, issuing an alarm; otherwise, proceeding to step S2; step S2: synchronously acquiring the thermal imaging image and basic associated data; preprocessing the thermal imaging image and completing pixel-level partitioning and calibration; and removing invalid pixels. This invention not only achieves real-time monitoring of data center temperature but also identifies potential overheating risks in advance through multi-dimensional data fusion and spatial analysis, preventing equipment downtime or damage due to high temperatures and ensuring business continuity. Simultaneously, by integrating multiple influencing factors, it significantly reduces false alarms and missed alarms, improves the intelligence level of the monitoring system, and provides a reliable guarantee for the stable operation of the data center.

Description

Technical Field

[0001] This invention relates to the field of data center monitoring technology, and in particular to a method and system for monitoring data center temperature based on thermal imaging. Background Technology

[0002] With the rapid development of cloud computing, big data, and artificial intelligence technologies, data center server rooms are expanding in scale and becoming increasingly integrated, leading to a dramatic increase in heat generation per unit area. Excessive temperature is a major cause of server downtime, accelerated component aging, and even fires. Therefore, real-time and accurate temperature monitoring of data center server rooms is crucial for ensuring business continuity and equipment safety.

[0003] Infrared thermal imaging technology is widely used in data center temperature monitoring due to its advantages such as non-contact operation, large area coverage, and intuitive visualization. By deploying thermal imaging cameras, maintenance personnel can obtain images of the temperature distribution on the surfaces of equipment in the data center and promptly identify hotspots. However, most existing thermal imaging monitoring methods remain at the level of "point measurement" or "area display," that is, simply triggering alarms by setting temperature thresholds. This method has a significant lag; the system only alarms when the temperature has already risen to a dangerous level, by which time equipment damage or thermal disasters may have already occurred. Summary of the Invention

[0004] To address the aforementioned shortcomings, the present invention aims to propose a data center computer room temperature monitoring and system based on thermal imaging, thereby solving the problem that existing infrared temperature detection cannot effectively predict the occurrence of overheating.

[0005] To achieve this objective, the present invention adopts the following technical solution: a method for monitoring the temperature of a data center computer room based on thermal imaging, comprising the following steps:

[0006] Step S1: Acquire a thermal imaging image of the computer room. Based on the binarized data of the thermal imaging image, determine whether there is an overheating situation in the current computer room. If there is, issue an alarm. If not, proceed to step S2.

[0007] Step S2: Synchronously acquire thermal imaging images of the computer room and basic correlation data, preprocess the thermal imaging images and complete pixel-level partitioning and calibration, and remove invalid pixels. The basic correlation data includes air conditioning operation data, server load data and computer room distribution map.

[0008] Step S3: Filter out effective temperature change areas based on the data center distribution map, and obtain corrected temperature change values ​​by associating the temperature change areas with the air conditioner outlet temperature;

[0009] Step S4: Introduce server load weighting factor and channel correction factor to calculate the corrected temperature change value and obtain single-point outlier value;

[0010] Step S5: Combine pixel distance and air conditioner distance to optimize the calculation of single-point outliers in the temperature change area and obtain the final outlier value;

[0011] Step S6: Determine whether the final abnormal value is greater than the preset value. If the final abnormal value is greater than the preset value, it is determined that there is a risk of excessively high temperature in the temperature change area.

[0012] Preferably, step S2 includes the following steps:

[0013] Step S21: Perform non-uniformity correction and bad pixel removal on the acquired raw thermal imaging image, convert it into a grayscale image, and unify the resolution through an interpolation algorithm;

[0014] Step S22: Based on the data center distribution map, establish a mapping relationship between pixel coordinates and physical locations, and divide the thermal imaging image into multiple logical regions, wherein the logical regions include at least:

[0015] Cold aisle area, hot aisle area, server rack area, and invalid area;

[0016] Step S23: Based on the partition calibration results, set the temperature value of pixels belonging to invalid areas to 0 or mark them as not participating in subsequent calculations, and retain only the remaining cold aisle area, hot aisle area and cabinet area as valid monitoring areas.

[0017] Preferably, step S3 includes the following steps:

[0018] Step S31: Obtain the thermal imaging image at the current moment and the thermal imaging image at the previous moment, and label them as the first image and the second image respectively. Perform differential calculation on the first image and the second image, and mark the areas with pixel changes as preliminary areas. Select the temperature change area at the current moment based on the preliminary areas and the effective monitoring areas.

[0019] Step S32: Based on the thermal imaging data of the first image and the second image, obtain the temperature change value corresponding to each pixel in each temperature change region;

[0020] Based on the logical region where the temperature change region is located, a corresponding region correction coefficient is assigned. The air outlet temperature of the air conditioner at the current moment is obtained. The temperature change value is then corrected using the air outlet temperature to obtain the corrected temperature change value. The formula for obtaining the corrected temperature change value is as follows:

[0021] ;

[0022] in Let k be the temperature change value of each pixel in the i-th temperature change region t, and k be the region correction coefficient. This refers to the actual air outlet temperature. The set air outlet temperature.

[0023] Preferably, the server load weighting factor in step S4 includes a burst weighting factor and a base load weighting factor;

[0024] The specific steps of step S4 are as follows:

[0025] Step S41: Obtain the average load rate of the servers in the rack corresponding to the temperature change region. and historical average load factor By average load rate and historical average load factor The basic load weighting factor is calculated and obtained, wherein the formula for obtaining the basic load weighting factor is as follows:

[0026] ,in The load sensitivity coefficient has a value ranging from -0.1 to -0.3.

[0027] Step S42: Obtain the load change rate of the servers in the corresponding rack within the temperature change region from the previous moment to the current moment. When the server load change rate is greater than the change threshold, obtain the value of the burst weighting factor by comparing the server load change rate with the change threshold. The formula for obtaining the burst weighting factor is as follows: , This is the mutation weighting coefficient, with a value ranging from 0.2 to 0.5. For server load change rate, The change rate is a threshold value. If the server load change rate is less than the change rate threshold, the burst weighting factor is set to 1.

[0028] Step S43: Based on the logical region where the temperature change region is located, assign a corresponding channel correction factor to each pixel. Among the channel correction factors, the hot aisle region factor is greater than the cabinet region factor, the cabinet region factor is greater than the cold aisle region factor, and the value of the channel correction factor is 0.8~1.5.

[0029] Step S44: Multiply the server load weighting factor, the channel correction factor, and the corrected temperature change value to obtain the temperature impact value;

[0030] Temperature impact values ​​that exceed the impact threshold are selected as the single-point outliers.

[0031] Preferably, step S5 includes the following steps:

[0032] Step S51: Collect the edge point pixels of each logical region to form an edge point set;

[0033] Step S52: For each single-point outlier, the nearest pixel in the edge point set is matched and used as the first point. The edge distance weighted value is obtained through the first point, wherein the formula for obtaining the edge distance weighted value is: ,in Let be the distance between the pixel corresponding to the i-th single-point outlier and the corresponding first point. This is the distance attenuation constant, with a value ranging from 5 to 30 pixels.

[0034] Step S53: Obtain the physical distance from the pixel corresponding to the single-point outlier to the air conditioner vent as the first distance. Calculate the air conditioner distance weighted value using the first distance. The formula for obtaining the air conditioner distance weighted value is as follows: ,in The first distance is the distance to the pixel corresponding to the i-th single-point outlier. This is a constant representing the range of influence of air conditioning. ,in R is a constant coefficient ranging from 0.5 to 3, and R is the proportional relationship between the pixel and the actual size.

[0035] Step S54: Obtain preliminary outlier values ​​by multiplying the edge distance weighted value, the air conditioner distance weighted value, and the single-point outlier value;

[0036] The average value of all preliminary outliers in the temperature change region is obtained by summing them up, and the average value of the preliminary outliers is used as the final outlier.

[0037] A data center computer room temperature monitoring system based on thermal imaging, using the aforementioned data center computer room temperature monitoring method based on thermal imaging, includes:

[0038] The over-temperature preliminary judgment module is used to acquire thermal imaging images of the computer room and determine whether there is an over-temperature situation in the current computer room based on the binarized data of the thermal imaging images. If there is, an alarm is issued; if not, the synchronous acquisition and preprocessing module is activated.

[0039] The synchronous acquisition and preprocessing module is used to synchronously acquire thermal imaging images and basic correlation data of the computer room, preprocess the thermal imaging images and complete pixel-level partitioning and calibration, and remove invalid pixels. The basic correlation data includes air conditioning operation data, server load data and computer room distribution map.

[0040] The temperature change correction module is used to filter effective temperature change areas based on the data center distribution map and obtain corrected temperature change values ​​by associating the temperature change areas with the air conditioner outlet temperature.

[0041] The single-point anomaly calculation module is used to calculate the corrected temperature change value by introducing the server load weighting factor and the channel correction factor, and to obtain the single-point anomaly value.

[0042] The outlier optimization module is used to combine pixel distance and air conditioner distance to optimize the calculation of single-point outliers in the temperature change area and obtain the final outlier value.

[0043] The risk assessment module is used to determine whether the final abnormal value is greater than the preset value. If the final abnormal value is greater than the preset value, it is determined that there is a risk of excessively high temperature in the temperature change area.

[0044] Preferably, the synchronous acquisition and preprocessing module includes:

[0045] The image correction unit is used to perform non-uniformity correction and bad pixel removal on the acquired raw thermal imaging image, convert it into a grayscale image, and unify the resolution through an interpolation algorithm. The partition calibration unit is used to establish the mapping relationship between pixel coordinates and physical locations based on the data center distribution map, and divide the thermal imaging image into multiple logical regions. The logical regions include at least cold aisle region, hot aisle region, rack region, and invalid region. The invalid pixel removal unit is used to set the temperature value of pixels belonging to invalid regions to 0 or mark them as not participating in subsequent calculations according to the partition calibration results, and only retain the cold aisle region, hot aisle region, and rack region as effective monitoring areas.

[0046] Preferably, the temperature change correction module includes:

[0047] The change region extraction unit is used to acquire the thermal imaging image at the current moment and the thermal imaging image at the previous moment, and label them as the first image and the second image, respectively. It performs binarized image difference calculation on the first image and the second image, and marks the regions with pixel changes as preliminary regions. Based on the preliminary regions and the effective monitoring regions, it filters out the temperature change regions at the current moment. The temperature change calculation unit is used to acquire the temperature change value corresponding to each pixel in each temperature change region based on the thermal imaging data of the first image and the second image. The correction calculation unit is used to assign the corresponding region correction coefficient based on the logical region where the temperature change region is located, acquire the air outlet temperature of the air conditioner at the current moment, and correct the temperature change value based on the air outlet temperature to obtain the corrected temperature change value.

[0048] Preferably, the server load weighting factor includes a burst weighting factor and a base load weighting factor; the single-point anomaly calculation module includes:

[0049] The system comprises the following components: a basic load weighting unit, a burst load weighting unit, and a single point anomaly determination unit. The basic load weighting unit acquires the average load rate and historical average load rate of servers within the corresponding rack in the temperature change region. It calculates the basic load weighting factor based on these average and historical load rates. The burst load weighting unit acquires the load change rate of servers within the corresponding rack in the temperature change region from the previous moment to the current moment. When the server load change rate exceeds a threshold, a burst weighting factor is calculated using the server load change rate and the threshold. If the server load change rate is less than the threshold, the burst weighting factor is set to a fixed value. The channel correction allocation unit assigns a corresponding channel correction factor to each pixel based on the logical region where the temperature change region is located. The hot aisle region factor is greater than the rack region factor, and the rack region factor is greater than the cold aisle region factor. The single point anomaly determination unit multiplies the server load weighting factor, the channel correction factor, and the corrected temperature change value to obtain the temperature impact value. It then filters out temperature impact values ​​exceeding the impact value threshold as single point anomalies.

[0050] Preferably, the outlier optimization module includes:

[0051] An edge point collection unit is used to collect edge point pixels of each logical region to form an edge point set; an edge distance weighting unit is used to match the nearest pixel in the edge point set as the first point for each pixel corresponding to a single outlier, and calculate the edge distance weighting value based on the distance between the pixel and the first point; an air conditioner distance weighting unit is used to obtain the physical distance from the pixel corresponding to the single outlier to the air conditioner outlet as the first distance, and calculate the air conditioner distance weighting value based on the first distance; a final outlier calculation unit is used to multiply the edge distance weighting value, the air conditioner distance weighting value and the single outlier to obtain a preliminary outlier value, and accumulate all preliminary outliers in the temperature change region to obtain the average value of the preliminary outlier value, and use the average value of the preliminary outlier value as the final outlier value.

[0052] One of the above technical solutions has the following advantages or beneficial effects: This invention not only realizes real-time monitoring of the computer room temperature, but also identifies potential overheating risks in advance through multi-dimensional data fusion and spatial analysis, avoiding equipment downtime or damage due to high temperature and ensuring business continuity; at the same time, due to the comprehensive consideration of multiple influencing factors, it significantly reduces the false alarm and false alarm rates, improves the intelligence level of the monitoring system, and provides a reliable guarantee for the stable operation of the data center. Attached Figure Description

[0053] Figure 1 This is a flowchart of one embodiment of the method of the present invention.

[0054] Figure 2 This is a schematic diagram of the structure of one embodiment of the system of the present invention.

[0055] Figure 3 This is a distribution diagram of the computer room in one embodiment of the present invention. Detailed Implementation

[0056] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0057] In the description of embodiments of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0058] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0059] like Figure 1 As shown, a method for monitoring the temperature of a data center computer room based on thermal imaging includes the following steps:

[0060] Step S1: Acquire a thermal imaging image of the computer room. Based on the binarized data of the thermal imaging image, determine whether there is an overheating situation in the current computer room. If there is, issue an alarm. If not, proceed to step S2.

[0061] Step S2: Synchronously collect thermal imaging images of the computer room and basic associated data, preprocess the thermal imaging images and complete pixel-level partitioning and calibration, and remove invalid pixels. The basic associated data includes air conditioning operation data, server load data and computer room distribution map. During the collection, thermal imaging images of the computer room need to be collected at fixed time intervals (such as every 30 seconds) to achieve periodic monitoring.

[0062] Step S3: Filter out effective temperature change areas based on the data center distribution map, and obtain corrected temperature change values ​​by associating the temperature change areas with the air conditioner outlet temperature;

[0063] Step S4: Introduce server load weighting factor and channel correction factor to calculate the corrected temperature change value and obtain single-point outlier value;

[0064] Step S5: Combine pixel distance and air conditioner distance to optimize the calculation of single-point outliers in the temperature change area and obtain the final outlier value;

[0065] Step S6: Determine whether the final abnormal value is greater than the preset value. If the final abnormal value is greater than the preset value, it is determined that there is a risk of excessively high temperature in the temperature change area.

[0066] To address the lag in over-temperature alarms during infrared thermal imaging monitoring, this invention correlates the raw thermal image with baseline data when no over-temperature conditions are detected in the server room, establishing a dynamic monitoring foundation. Specifically, the thermal image is preprocessed and pixel-level partitioned, eliminating invalid pixels to lay the groundwork for subsequent accurate analysis. Then, effective temperature change areas are selected based on the server room distribution map, and corrected temperature change values ​​are obtained by combining this with the air conditioning outlet temperature. This process eliminates the interference of air conditioning airflow, making the corrected temperature change values ​​more reflective of the equipment's own heating trends. Of course, the heat dissipation of different servers in the rack varies under different load conditions. While the corrected temperature change value can reflect normal temperature changes, it cannot determine whether the changes are reasonable. Therefore, it is necessary to calculate the corrected temperature change value by combining server load weighting factors and channel correction factors, and then filter out single-point outliers with abnormal temperature data. A single outlier only indicates an anomaly in a single data point. However, in infrared thermal imaging monitoring, a single outlier may be due to other interference and cannot indicate that there is a risk of overheating in the current computer room. Therefore, it is necessary to combine the pixel distance and the air conditioner distance to optimize the calculation of single outliers in the temperature change area and obtain the final outlier. In this process, multiple single outliers in the temperature change area will be combined to obtain a more accurate final outlier. Finally, the final outlier is compared with a preset threshold. If it is greater than the preset value, it is judged that there is a risk of overheating, so that an early warning is issued before the temperature reaches the danger threshold.

[0067] This invention not only enables real-time monitoring of the computer room temperature, but also identifies potential overheating risks in advance through multi-dimensional data fusion and spatial analysis, preventing equipment from crashing or being damaged due to high temperatures and ensuring business continuity. At the same time, by integrating multiple influencing factors, it significantly reduces the false alarm and false alarm rates, improves the intelligence level of the monitoring system, and provides a reliable guarantee for the stable operation of the data center.

[0068] Preferably, step S2 includes the following steps:

[0069] Step S21: Perform non-uniformity correction and bad pixel removal on the acquired raw thermal imaging image, convert it into a grayscale image, and unify the resolution through an interpolation algorithm;

[0070] Step S22: Based on the data center distribution map, establish a mapping relationship between pixel coordinates and physical locations, and divide the thermal imaging image into multiple logical regions, wherein the logical regions include at least:

[0071] Cold aisle area (air intake side of the rack), hot aisle area (air exhaust side of the rack), rack area, and ineffective area (such as walls, pillars, and blank racks without servers).

[0072] Step S23: Based on the partition calibration results, set the temperature value of pixels belonging to invalid areas to 0 or mark them as not participating in subsequent calculations, and retain only the remaining cold aisle area, hot aisle area and cabinet area as valid monitoring areas.

[0073] In traditional methods, raw thermal imaging images often suffer from non-uniform noise due to uneven sensor response and dead pixels caused by pixel failure. Direct use of these images for analysis can easily lead to misjudgments of temperature. Furthermore, due to differences in lens focal length and shooting angle, the resolution of images acquired at different times may vary, resulting in a lack of comparability in subsequent time-series temperature change analysis. Therefore, this invention eliminates systematic errors inherent in the equipment by performing non-uniformity correction and dead pixel removal on the raw thermal imaging images. The images are then converted to grayscale, and an interpolation algorithm is used to unify the resolution of all acquired images, thus constructing a clean and standardized thermal imaging image dataset. Spatial dimension information is then introduced. Based on the data center distribution map, a precise mapping relationship between pixel coordinates and physical locations is established, and the thermal imaging images are divided into cold aisle areas, hot aisle areas, rack areas, and invalid areas. This division is crucial because it gives the originally abstract pixel temperature values ​​clear physical meaning—for example, it can distinguish which pixels belong to the cold aisle on the server intake side, which belong to the hot aisle on the exhaust side, and which pixels correspond to walls or empty racks without servers. Finally, based on the above partitioning and calibration results, the temperature values ​​of pixels belonging to invalid areas were directly set to zero or marked as not participating in subsequent calculations, thereby eliminating interfering data and retaining only the cold aisle, hot aisle, and rack areas as effective monitoring areas. This reduced unnecessary data calculations.

[0074] Specifically, in one embodiment, the layout of the computer room is as follows: Figure 3As shown, the first and second rows of server racks are arranged face-to-face, with an aisle between them. An air conditioner is installed in front of this aisle. In this case, the aisle between the first and second rows of server racks is the cold aisle. The second and third rows of server racks are arranged back-to-back, also with an aisle between them; this aisle is the hot aisle. The corresponding logical areas can be divided according to the data center layout diagram. For example, the red dashed box represents the cold aisle area, the black dashed box represents the server rack area, and the blue dashed box represents the hot aisle area. Of course... Figure 3 It is just a data center distribution map. When used, the mapping relationship between pixel coordinates and physical locations in the thermal imaging image can be constructed based on the data center distribution map, so as to divide the corresponding logical areas in the thermal imaging image.

[0075] Preferably, step S3 includes the following steps:

[0076] Step S31: Obtain the thermal imaging image at the current moment and the thermal imaging image at the previous moment, and label them as the first image and the second image respectively. Perform differential calculation on the first image and the second image, and mark the areas with pixel changes as preliminary areas. Select the temperature change area at the current moment based on the preliminary areas and the effective monitoring areas.

[0077] Step S32: Based on the thermal imaging data of the first image and the second image, obtain the temperature change value corresponding to each pixel in each temperature change region;

[0078] Based on the logical region where the temperature change region is located, a corresponding region correction coefficient is assigned. The air outlet temperature of the air conditioner at the current moment is obtained. The temperature change value is then corrected using the air outlet temperature to obtain the corrected temperature change value. The formula for obtaining the corrected temperature change value is as follows:

[0079] ;

[0080] in Let k be the temperature change value of each pixel in the i-th temperature change region t, and k be the region correction coefficient. This refers to the actual air outlet temperature. The set air outlet temperature.

[0081] Since air conditioning is used to regulate the temperature in the server room, when considering the risk of overheating by obtaining temperature change values ​​from images at two adjacent times, it is necessary to take into account the influence of air conditioning vents and the different positions of the server racks. Therefore, in this invention, in step S31, binarized difference calculation is first performed on the first and second images to obtain a preliminary image with pixel changes. During thermal imaging, the binarized pixel values ​​can be used to determine if there is a temperature change. Difference processing compares the differences between two images / pixels to determine temperature changes, diffusion, addition / disappearance, thus obtaining the corresponding preliminary areas with temperature changes. Then, areas overlapping with the effective monitoring area are selected from the preliminary areas to obtain the temperature change area at the current moment. This avoids interference from invalid areas and reduces the amount of data to be calculated. The data obtained through binarized difference calculation corresponds to each temperature value. Based on the thermal imaging data of the first and second images, the data from the binarized difference calculation is used to obtain the temperature change value corresponding to each temperature change area.

[0082] The cooling effect of air conditioning varies depending on the location of the server rack. For example, the side of the rack facing the cold aisle experiences significant cooling, while the side facing the hot aisle experiences less cooling. Therefore, staff can assign corresponding correction coefficients to the logical regions where temperature changes occur using correction coefficients measured in a pre-tested environment. Specifically, the cold aisle region is significantly affected, and its correction coefficient is greater than that of the server rack and hot aisle regions. The correction coefficient for the server rack region is also correspondingly greater than that for the hot aisle region. By using the correction coefficients, the air conditioner's outlet temperature, and the actual measured outlet temperature, the actual temperature change (i.e., the corrected temperature change value) for each temperature change region can be calculated. This lays a solid foundation for subsequent judgments.

[0083] Preferably, the server load weighting factor in step S4 includes a burst weighting factor and a base load weighting factor;

[0084] The specific steps of step S4 are as follows:

[0085] Step S41: Obtain the average load rate of the servers in the rack corresponding to the temperature change region. and historical average load factor By average load rate and historical average load factor The basic load weighting factor is calculated and obtained, wherein the formula for obtaining the basic load weighting factor is as follows:

[0086] ,in The load sensitivity coefficient has a value ranging from -0.1 to -0.3.

[0087] Step S42: Obtain the load change rate of the servers in the corresponding rack within the temperature change region from the previous moment to the current moment. When the server load change rate is greater than the change threshold, obtain the value of the burst weighting factor by comparing the server load change rate with the change threshold. The formula for obtaining the burst weighting factor is as follows: , This is the mutation weighting coefficient, with a value ranging from 0.2 to 0.5. For server load change rate, The change rate is a threshold value. If the server load change rate is less than the change rate threshold, the burst weighting factor is set to 1.

[0088] Step S43: Based on the logical region where the temperature change region is located, assign a corresponding channel correction factor to each pixel. Among the channel correction factors, the hot aisle region factor is greater than the cabinet region factor, the cabinet region factor is greater than the cold aisle region factor, and the value of the channel correction factor is 0.8~1.5.

[0089] Step S44: Multiply the server load weighting factor, the channel correction factor, and the corrected temperature change value to obtain the temperature impact value;

[0090] Temperature impact values ​​that exceed the impact threshold are selected as the single-point outliers.

[0091] Traditional monitoring methods often simply attribute temperature increases to heat dissipation failures, neglecting the core internal factor of server load changes. When server load increases, its heat generation inevitably rises, and a moderate temperature increase is normal. Indiscriminately triggering alarms in this case easily leads to frequent false alarms. Conversely, if the load suddenly drops without a corresponding temperature decrease, it may indicate a problem with the cooling system, but traditional methods struggle to detect such issues. Therefore, in step S4 of this invention, the average load rate and its historical average of the servers within the rack corresponding to the temperature change area are first obtained. A basic load weighting factor is calculated using a formula, where the load sensitivity coefficient ranges from -0.1 to -0.3. This design ensures that the factor is negatively correlated with the load level. When the current server load is higher than the historical average, the basic load weighting factor decreases, thus assigning a lower abnormal weight to temperature increases and avoiding misjudgments caused by normal load fluctuations. Conversely, when the load is lower than the historical level, the value of the basic load weighting factor increases, assigning a higher risk weight to the same temperature change, thereby revealing potential problems such as decreased heat dissipation efficiency.

[0092] Further focus is placed on the instantaneous change trend of the load. By calculating the server load change rate and comparing it with the change threshold, a burst weighting factor is introduced when the load changes abruptly. When the load change rate exceeds the threshold, the burst weighting factor is calculated by formula, where the burst weighting coefficient is between 0.2 and 0.5, so that the factor is positively correlated with the change rate. This amplifies the abnormal temperature value when the load increases sharply, and provides early warning of the overheating risk that may be caused by it.

[0093] Then, based on the logical region where the temperature change area is located (cold aisle, hot aisle, server rack), a channel correction factor is assigned to each pixel. This differentiated assignment fully considers the physical characteristics of the airflow organization in the data center. The hot aisle, which accumulates the heat expelled by the server, allows for higher temperatures, so a larger factor is assigned to avoid oversensitivity; the cold aisle is the air intake side, and even slight temperature anomalies may indicate cooling failure, so a smaller factor is assigned to enhance sensitivity; the server rack area is in the middle. Through this correction, the anomaly assessment is highly consistent with the actual environment of the equipment, avoiding misjudgments caused by a "one-size-fits-all" threshold.

[0094] Finally, the obtained server load weighting factor, channel correction factor, and corrected temperature change value are multiplied together to obtain the temperature impact value. Outliers greater than the impact value threshold are filtered out. These outliers indicate that there may be a problem with the temperature, so they need to be marked to prepare for subsequent calculations.

[0095] This invention integrates historical load trends, sudden load events, and spatial location characteristics, transforming the calculation of single-point anomalies from isolated temperature reading comparisons to an intelligent assessment that comprehensively considers equipment operating status and the physical environment. This significantly improves the accuracy and predictability of anomaly detection. A basic load weighting factor ensures system stability under normal high-load conditions, while a burst weighting factor enables the system to respond quickly to load fluctuations. A channel correction factor makes the assessment more closely aligned with the airflow characteristics of the data center. The synergistic effect of these three factors effectively reduces false alarm rates and allows for timely detection of anomalies in their early stages.

[0096] Preferably, step S5 includes the following steps:

[0097] Step S51: Collect the edge point pixels of each logical region to form an edge point set;

[0098] Step S52: For each single-point outlier, the nearest pixel in the edge point set is matched and used as the first point. The edge distance weighted value is obtained through the first point, wherein the formula for obtaining the edge distance weighted value is: ,in Let be the distance between the pixel corresponding to the i-th single-point outlier and the corresponding first point. This is the distance attenuation constant, with a value ranging from 5 to 30 pixels.

[0099] Step S53: Obtain the physical distance from the pixel corresponding to the single-point outlier to the air conditioner vent as the first distance. Calculate the air conditioner distance weighted value using the first distance. The formula for obtaining the air conditioner distance weighted value is as follows: ,in The first distance is the distance to the pixel corresponding to the i-th single-point outlier. This is a constant representing the range of influence of air conditioning. ,in R is a constant coefficient ranging from 0.5 to 3, and R is the proportional relationship between the pixel and the actual size.

[0100] Step S54: Obtain preliminary outlier values ​​by multiplying the edge distance weighted value, the air conditioner distance weighted value, and the single-point outlier value;

[0101] The average value of all preliminary outliers in the temperature change region is obtained by summing them up, and the average value of the preliminary outliers is used as the final outlier.

[0102] When a server experiences a thermal failure, such as a fan stopping or a sudden increase in load, the heat doesn't remain stationary. It spreads to the surrounding area through heat conduction (solid-to-solid heat transfer) and heat convection (airflow). The edge points in this diffusion process serve as early warnings of the expanding high-temperature region. Therefore, this invention collects edge point pixels from each logical region to form an edge point set. Then, for each single-point outlier, the nearest pixel in the edge point set is selected as the first point. An edge distance weighting value is calculated using a formula, where is the distance between the pixel corresponding to the i-th single-point outlier and the first point, and is a distance decay constant. In the edge distance weighting value calculation, the closer the outlier (the pixel corresponding to the single-point outlier) is to the edge of the region, the smaller the distance, and the larger the edge distance weighting value (approaching 1). Conversely, the deeper the outlier is into the region, the greater the distance, and the edge distance weighting value decays exponentially. This design fully reflects the physical laws of heat diffusion. Temperature anomalies near the edge are more likely to reflect an imbalance in heat exchange at the regional boundary and pose a higher risk, while temperature anomalies inside the region may be caused by local equipment problems and are relatively isolated risks. This allows for differentiated risk weighting of anomaly points in different spatial locations.

[0103] Further, after obtaining the edge distance weighted value, it is also necessary to obtain the physical distance from the pixel corresponding to the single-point outlier to the air conditioner vent as the first distance, and calculate the air conditioner distance weighted value using a formula. Areas close to the air conditioner should be adequately cooled; if temperature anomalies occur in these areas, it is highly likely that there is a problem with the air conditioning system itself or the airflow organization, and the air conditioner distance weighted value should be higher. Conversely, slightly higher temperatures in areas far from the air conditioner are normal, and even if anomalies occur, they should be assigned lower air conditioner distance weighted values ​​to avoid misjudgment.

[0104] Finally, all the preliminary outliers in the temperature change region are summed to obtain the average value, resulting in the final outlier. This summation and averaging operation aggregates the outliers of isolated points into regional risk indicators for the corresponding temperature change region, thereby enabling an overall assessment of the thermal risk level of the temperature change region.

[0105] A data center computer room temperature monitoring system based on thermal imaging, using the aforementioned data center computer room temperature monitoring method based on thermal imaging, includes:

[0106] The over-temperature preliminary judgment module is used to acquire thermal imaging images of the computer room and determine whether there is an over-temperature situation in the current computer room based on the binarized data of the thermal imaging images. If there is, an alarm is issued; if not, the synchronous acquisition and preprocessing module is activated.

[0107] The synchronous acquisition and preprocessing module is used to synchronously acquire thermal imaging images and basic correlation data of the computer room, preprocess the thermal imaging images and complete pixel-level partitioning and calibration, and remove invalid pixels. The basic correlation data includes air conditioning operation data, server load data and computer room distribution map.

[0108] The temperature change correction module is used to filter effective temperature change areas based on the data center distribution map and obtain corrected temperature change values ​​by associating the temperature change areas with the air conditioner outlet temperature.

[0109] The single-point anomaly calculation module is used to calculate the corrected temperature change value by introducing the server load weighting factor and the channel correction factor, and to obtain the single-point anomaly value.

[0110] The outlier optimization module is used to combine pixel distance and air conditioner distance to optimize the calculation of single-point outliers in the temperature change area and obtain the final outlier value.

[0111] The risk assessment module is used to determine whether the final abnormal value is greater than the preset value. If the final abnormal value is greater than the preset value, it is determined that there is a risk of excessively high temperature in the temperature change area.

[0112] Preferably, the synchronous acquisition and preprocessing module includes:

[0113] The image correction unit is used to perform non-uniformity correction and bad pixel removal on the acquired raw thermal imaging image, convert it into a grayscale image, and unify the resolution through an interpolation algorithm. The partition calibration unit is used to establish the mapping relationship between pixel coordinates and physical locations based on the data center distribution map, and divide the thermal imaging image into multiple logical regions. The logical regions include at least cold aisle region, hot aisle region, rack region, and invalid region. The invalid pixel removal unit is used to set the temperature value of pixels belonging to invalid regions to 0 or mark them as not participating in subsequent calculations according to the partition calibration results, and only retain the cold aisle region, hot aisle region, and rack region as effective monitoring areas.

[0114] Preferably, the temperature change correction module includes:

[0115] The change region extraction unit is used to acquire the thermal imaging image at the current moment and the thermal imaging image at the previous moment, and label them as the first image and the second image, respectively. It performs binarized image difference calculation on the first image and the second image, and marks the regions with pixel changes as preliminary regions. Based on the preliminary regions and the effective monitoring regions, it filters out the temperature change regions at the current moment. The temperature change calculation unit is used to acquire the temperature change value corresponding to each pixel in each temperature change region based on the thermal imaging data of the first image and the second image. The correction calculation unit is used to assign the corresponding region correction coefficient based on the logical region where the temperature change region is located, acquire the air outlet temperature of the air conditioner at the current moment, and correct the temperature change value based on the air outlet temperature to obtain the corrected temperature change value.

[0116] Preferably, the server load weighting factor includes a burst weighting factor and a base load weighting factor; the single-point anomaly calculation module includes:

[0117] The system comprises the following components: a basic load weighting unit, a burst load weighting unit, and a single point anomaly determination unit. The basic load weighting unit acquires the average load rate and historical average load rate of servers within the corresponding rack in the temperature change region. It calculates the basic load weighting factor based on these average and historical load rates. The burst load weighting unit acquires the load change rate of servers within the corresponding rack in the temperature change region from the previous moment to the current moment. When the server load change rate exceeds a threshold, a burst weighting factor is calculated using the server load change rate and the threshold. If the server load change rate is less than the threshold, the burst weighting factor is set to a fixed value. The channel correction allocation unit assigns a corresponding channel correction factor to each pixel based on the logical region where the temperature change region is located. The hot aisle region factor is greater than the rack region factor, and the rack region factor is greater than the cold aisle region factor. The single point anomaly determination unit multiplies the server load weighting factor, the channel correction factor, and the corrected temperature change value to obtain the temperature impact value. It then filters out temperature impact values ​​exceeding the impact value threshold as single point anomalies.

[0118] Preferably, the outlier optimization module includes:

[0119] An edge point collection unit is used to collect edge point pixels of each logical region to form an edge point set; an edge distance weighting unit is used to match the nearest pixel in the edge point set as the first point for each pixel corresponding to a single outlier, and calculate the edge distance weighting value based on the distance between the pixel and the first point; an air conditioner distance weighting unit is used to obtain the physical distance from the pixel corresponding to the single outlier to the air conditioner outlet as the first distance, and calculate the air conditioner distance weighting value based on the first distance; a final outlier calculation unit is used to multiply the edge distance weighting value, the air conditioner distance weighting value and the single outlier to obtain a preliminary outlier value, and accumulate all preliminary outliers in the temperature change region to obtain the average value of the preliminary outlier value, and use the average value of the preliminary outlier value as the final outlier value.

[0120] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0121] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A thermal imaging based data center room temperature monitoring method, characterized in that, Includes the following steps: Step S1: Acquire a thermal imaging image of the computer room. Based on the binarized data of the thermal imaging image, determine whether there is an overheating situation in the current computer room. If there is, issue an alarm. If not, proceed to step S2. Step S2: Synchronously acquire thermal imaging images of the computer room and basic correlation data, preprocess the thermal imaging images and complete pixel-level partitioning and calibration, and remove invalid pixels. The basic correlation data includes air conditioning operation data, server load data and computer room distribution map. Step S3: Filter out effective temperature change areas based on the data center distribution map, and obtain corrected temperature change values ​​by associating the temperature change areas with the air conditioner outlet temperature; Step S4: Introduce server load weighting factor and channel correction factor to calculate the corrected temperature change value and obtain single-point outlier value; Step S5: Combine pixel distance and air conditioner distance to optimize the calculation of single-point outliers in the temperature change area and obtain the final outlier value; Step S6: Determine whether the final outlier is greater than the preset value. If the final outlier is greater than the preset value, then determine that there is a risk of excessively high temperature in the temperature change area. The server load weighting factor in step S4 includes a burst weighting factor and a base load weighting factor; The specific steps of step S4 are as follows: Step S41: Obtain the average load rate of the servers in the corresponding rack within the temperature change region. and historical average load factor By average load rate and historical average load factor The basic load weighting factor is calculated and obtained, wherein the formula for obtaining the basic load weighting factor is as follows: ,in The load sensitivity coefficient has a value ranging from -0.1 to -0.

3. Step S42: Obtain the load change rate of the servers in the corresponding rack within the temperature change region from the previous moment to the current moment. When the server load change rate is greater than the change threshold, obtain the value of the burst weighting factor by comparing the server load change rate with the change threshold. The formula for obtaining the burst weighting factor is as follows: , This is the mutation weighting coefficient, with a value ranging from 0.2 to 0.

5. For server load change rate, The change rate is a threshold value. If the server load change rate is less than the change rate threshold, the burst weighting factor is set to 1. Step S43: Based on the logical region where the temperature change region is located, assign a corresponding channel correction factor to each pixel. Among the channel correction factors, the hot aisle region factor is greater than the cabinet region factor, the cabinet region factor is greater than the cold aisle region factor, and the value of the channel correction factor is 0.8~1.

5. Step S44: Multiply the server load weighting factor, the channel correction factor, and the corrected temperature change value to obtain the temperature impact value; Temperature impact values ​​that exceed the impact threshold are selected as the single-point outliers.

2. The data center computer room temperature monitoring method based on thermal imaging according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Perform non-uniformity correction and bad pixel removal on the acquired raw thermal imaging image, convert it into a grayscale image, and unify the resolution through an interpolation algorithm; Step S22: Based on the data center distribution map, establish a mapping relationship between pixel coordinates and physical locations, and divide the thermal imaging image into multiple logical regions, wherein the logical regions include at least: Cold aisle area, hot aisle area, server rack area, and invalid area; Step S23: Based on the partition calibration results, set the temperature value of pixels belonging to invalid areas to 0 or mark them as not participating in subsequent calculations, and retain only the remaining cold aisle area, hot aisle area and cabinet area as valid monitoring areas.

3. The data center computer room temperature monitoring method based on thermal imaging according to claim 2, characterized in that, Step S3 includes the following steps: Step S31: Obtain the thermal imaging image at the current moment and the thermal imaging image at the previous moment, and label them as the first image and the second image respectively. Perform differential calculation on the first image and the second image, and mark the areas with pixel changes as preliminary areas. Select the temperature change area at the current moment based on the preliminary areas and the effective monitoring areas. Step S32: Based on the thermal imaging data of the first image and the second image, obtain the temperature change value corresponding to each pixel in each temperature change region; Based on the logical region where the temperature change region is located, a corresponding region correction coefficient is assigned. The air outlet temperature of the air conditioner at the current moment is obtained. The temperature change value is then corrected using the air outlet temperature to obtain the corrected temperature change value. The formula for obtaining the corrected temperature change value is as follows: ； in Let k be the temperature change value of each pixel in the i-th temperature change region t, and k be the region correction coefficient. This refers to the actual air outlet temperature. The set air outlet temperature.

4. The data center computer room temperature monitoring method based on thermal imaging according to claim 3, characterized in that, Step S5 includes the following steps: Step S51: Collect the edge point pixels of each logical region to form an edge point set; Step S52: For each single-point outlier, the nearest pixel in the edge point set is matched and used as the first point. The edge distance weighted value is obtained through the first point, wherein the formula for obtaining the edge distance weighted value is: ,in Let be the distance between the pixel corresponding to the i-th single-point outlier and the corresponding first point. This is the distance attenuation constant, with a value ranging from 5 to 30 pixels. Step S53: Obtain the physical distance from the pixel corresponding to the single-point outlier to the air conditioner vent as the first distance. Calculate the air conditioner distance weighted value using the first distance. The formula for obtaining the air conditioner distance weighted value is as follows: ,in The first distance is the pixel corresponding to the i-th single-point outlier. This is a constant representing the range of influence of air conditioning. ,in R is a constant coefficient ranging from 0.5 to 3, and R is the proportional relationship between the pixel and the actual size. Step S54: Obtain preliminary outlier values ​​by multiplying the edge distance weighted value, the air conditioner distance weighted value, and the single-point outlier value; The average value of all preliminary outliers in the temperature change region is obtained by summing them up, and the average value of the preliminary outliers is used as the final outlier.

5. A data center computer room temperature monitoring system based on thermal imaging, using the data center computer room temperature monitoring method based on thermal imaging as described in any one of claims 1 to 4, characterized in that, include: The over-temperature preliminary judgment module is used to acquire thermal imaging images of the computer room and determine whether there is an over-temperature situation in the current computer room based on the binarized data of the thermal imaging images. If there is, an alarm is issued; if not, the synchronous acquisition and preprocessing module is activated. The synchronous acquisition and preprocessing module is used to synchronously acquire thermal imaging images and basic correlation data of the computer room, preprocess the thermal imaging images and complete pixel-level partitioning and calibration, and remove invalid pixels. The basic correlation data includes air conditioning operation data, server load data and computer room distribution map. The temperature change correction module is used to filter effective temperature change areas based on the data center distribution map and obtain corrected temperature change values ​​by associating the temperature change areas with the air conditioner outlet temperature. The single-point anomaly calculation module is used to calculate the corrected temperature change value by introducing the server load weighting factor and the channel correction factor, and to obtain the single-point anomaly value. The outlier optimization module is used to combine pixel distance and air conditioner distance to optimize the calculation of single-point outliers in the temperature change area and obtain the final outlier value. The risk assessment module is used to determine whether the final abnormal value is greater than the preset value. If the final abnormal value is greater than the preset value, it is determined that there is a risk of excessively high temperature in the temperature change area.

6. The data center computer room temperature monitoring system based on thermal imaging according to claim 5, characterized in that, The synchronous acquisition and preprocessing module includes: The image correction unit is used to perform non-uniformity correction and bad pixel removal on the acquired raw thermal imaging image, convert it into a grayscale image, and unify the resolution through an interpolation algorithm. The partition calibration unit is used to establish the mapping relationship between pixel coordinates and physical locations based on the data center distribution map, and divide the thermal imaging image into multiple logical regions. The logical regions include at least cold aisle region, hot aisle region, rack region, and invalid region. The invalid pixel removal unit is used to set the temperature value of pixels belonging to invalid regions to 0 or mark them as not participating in subsequent calculations according to the partition calibration results, and only retain the cold aisle region, hot aisle region, and rack region as effective monitoring areas.

7. The data center computer room temperature monitoring system based on thermal imaging according to claim 6, characterized in that, The temperature change correction module includes: The change region extraction unit is used to acquire the thermal imaging image at the current moment and the thermal imaging image at the previous moment, and label them as the first image and the second image, respectively. It performs binarized image difference calculation on the first image and the second image, and marks the regions with pixel changes as preliminary regions. Based on the preliminary regions and the effective monitoring regions, it filters out the temperature change regions at the current moment. The temperature change calculation unit is used to acquire the temperature change value corresponding to each pixel in each temperature change region based on the thermal imaging data of the first image and the second image. The correction calculation unit is used to assign the corresponding region correction coefficient based on the logical region where the temperature change region is located, acquire the air outlet temperature of the air conditioner at the current moment, and correct the temperature change value based on the air outlet temperature to obtain the corrected temperature change value.

8. The data center computer room temperature monitoring system based on thermal imaging according to claim 7, characterized in that, The server load weighting factor includes a burst weighting factor and a base load weighting factor; The single-point anomaly calculation module includes: The system comprises the following components: a basic load weighting unit, a burst load weighting unit, and a single point anomaly determination unit. The basic load weighting unit acquires the average load rate and historical average load rate of servers within the corresponding rack in the temperature change region. It calculates the basic load weighting factor based on these average and historical load rates. The burst load weighting unit acquires the load change rate of servers within the corresponding rack in the temperature change region from the previous moment to the current moment. When the server load change rate exceeds a threshold, a burst weighting factor is calculated using the server load change rate and the threshold. If the server load change rate is less than the threshold, the burst weighting factor is set to a fixed value. The channel correction allocation unit assigns a corresponding channel correction factor to each pixel based on the logical region where the temperature change region is located. The hot aisle region factor is greater than the rack region factor, and the rack region factor is greater than the cold aisle region factor. The single point anomaly determination unit multiplies the server load weighting factor, the channel correction factor, and the corrected temperature change value to obtain the temperature impact value. It then filters out temperature impact values ​​exceeding the impact value threshold as single point anomalies.

9. The data center computer room temperature monitoring system based on thermal imaging according to claim 8, characterized in that, The outlier optimization module includes: An edge point collection unit is used to collect edge point pixels of each logical region to form an edge point set; an edge distance weighting unit is used to match the nearest pixel in the edge point set as the first point for each pixel corresponding to a single outlier, and calculate the edge distance weighting value based on the distance between the pixel and the first point; an air conditioner distance weighting unit is used to obtain the physical distance from the pixel corresponding to the single outlier to the air conditioner outlet as the first distance, and calculate the air conditioner distance weighting value based on the first distance; a final outlier calculation unit is used to multiply the edge distance weighting value, the air conditioner distance weighting value and the single outlier to obtain a preliminary outlier value, and accumulate all preliminary outliers in the temperature change region to obtain the average value of the preliminary outlier value, and use the average value of the preliminary outlier value as the final outlier value.

Images