Method for evaluating influence of precise air conditioner on temperature and humidity sensor of computer IT machine room

Abstract

The invention discloses a method for evaluating the influence of a precise air conditioner on temperature and humidity sensors of a computer IT machine room, and the basic thought of the method comprises the steps: changing the temperature setting of a certain step length of each air conditioner in turn, then obtaining the temperature changes of all sensors in the machine room in corresponding time periods, and carrying out the vectorization of the temperature change data, according to the method, the K-MEANS clustering algorithm is used for carrying out preliminary grouping clustering to obtain the element density of clustering sub-clusters, the density serves as the clustering radius of the DBSCAN clustering algorithm to carry out clustering operation grouping again, and therefore the influence association degree between each air conditioner and each sensor is obtained. And the task can be completed only through a data mining technology, so that the workload is greatly saved.

Description

Technical Field

[0001] The present invention belongs to the technical field of HVAC energy-saving intelligent control, and particularly relates to the AI intelligent group control algorithm technology for air conditioners in IT computer rooms. Background Art

[0002] The cabinets in the computer rooms of telecommunication operators are generally arranged in rows and columns, and most of the terminal air conditioners are arranged against the wall. The air supply methods of the air conditioners include under-floor air supply, upper-air supply, duct air supply, etc. The power consumption of the precision air conditioners in the computer room accounts for about 40% of the total energy consumption of the computer room. The primary task of computer room energy saving is to reduce the energy consumption of the air conditioners in the computer room. Due to the uneven distribution of IT loads in the computer room caused by the rack arrangement, in order to achieve air conditioner energy saving, multiple temperature and humidity sensors need to be arranged on each row of cabinets to detect the temperature field in the room in real time, and the AI algorithm is used to adjust the supply air temperature or return air temperature of each air conditioner, so as to achieve a basic balance of hot and cold in the temperature field in the computer room and achieve the effect of reducing the air conditioner energy consumption.

[0003] According to actual needs, a number of temperature and humidity sensors are arranged on each row of cabinets according to the cabinet layout, which can sense the temperature field in the computer room in real time, and adjust the supply air temperature of each air conditioner to adjust the air flow temperature in the channels between the rows and columns of cabinets. The set value of the supply air temperature of the air conditioner needs to be based on the temperature values reported by each temperature and humidity sensor within the influence range of the air conditioner. Due to the different air flow organizations and air supply orientations in the computer room, for each computer room, it is necessary to calculate the correlation degree between the air conditioner and the temperature and humidity sensors in order to correctly adjust the temperature field expressed by the sensors and create a basically balanced and energy-efficient temperature field. Therefore, there is an urgent need for an AI algorithm that can evaluate the influence of precision air conditioners on temperature and humidity sensors in computer IT rooms. Summary of the Invention

[0004] Object of the Invention: Aiming at the above existing problems and deficiencies, the object of the present invention is to provide a method for evaluating the influence of precision air conditioners on temperature and humidity sensors in computer IT rooms. This algorithm does not require manual on-site investigation and data annotation, and can complete the task only through data mining technology, greatly saving the workload.

[0005] Technical Solution: To achieve the above object of the invention, the technical solution adopted by the present invention is: A method for evaluating the influence of precision air conditioners on temperature and humidity sensors in computer IT rooms, comprising the following steps:

[0006] S1, adjust the return air temperature of the air conditioners in the computer room one by one, and record the temperature change values of all temperature and humidity sensors after the adjustment; perform multiple rounds of adjustment on the return air temperature of the air conditioners, so as to construct a two-dimensional point data sample set of the sensor temperature change values and the return air temperature adjustment values;

[0007] S2. Based on the K-KEANS clustering algorithm and the empirical K value, using the temperature change value of the sensor as the distance, perform clustering operations on the two-dimensional point data samples in step S1 to obtain K sub-clusters; the points in each sub-cluster are sorted according to the temperature change value, and the distances between adjacent points are calculated, and the average distance L between adjacent points within all sub-clusters is calculated, L = {L1, L2, …, L p , …, L K}; and then obtain the average point density L mean of the two-dimensional point data sample set and the point density L min of the sub-cluster with the highest density,

[0008]

[0009] L min = Min{L1, L2, … L p , … L k};

[0010] S3. Based on the DBSCAN clustering algorithm, using the point density L min of the sub-cluster with the highest density as the clustering radius parameter, and setting the minimum number of points MinPts, perform clustering operations to obtain N sub-clusters; if clustering cannot be successfully performed with L min as the clustering radius parameter, then use the average point density L mean and as the clustering radius, and perform clustering again; through the following formula, calculate the correlation degree R m of the points within the sub-cluster after clustering, R = MTn / t

[0011] m m / t

[0012] In the formula, MT m represents the average value of the temperature change values within each sub-cluster after clustering, t represents the adjustment value of the return air temperature of the air-conditioning unit during each round of temperature adjustment; R m represents the correlation degree of the points within the sub-cluster;

[0013] S4. Calculation of the correlation between the sensor and the air conditioner: Calculate the minimum value MT m of the average temperature change value of the sensor numbered i corresponding to the air conditioner numbered j through the following formula,

[0014]

[0015] MT m = min(mT ij )

[0016] In the formula, mT ij represents the average temperature change value of the sensor numbered i and the air conditioner numbered j, MT m represents the minimum value of the average temperature change value of the sensor; ΔT ij,aDenote the temperature change value of sensor i when adjusting the air conditioner numbered j in the a-th round of adjustment.

[0017] Finally, take the R corresponding to MT m as the evaluation of the influence correlation degree between sensor i and air conditioner j. m As the evaluation of the influence correlation degree between sensor i and air conditioner j.

[0018] Furthermore, the return air temperature of the air conditioner described in step S1 is adjusted through at least 3 rounds, and the adjustment value t of the return air temperature of the air conditioner unit during each round of temperature adjustment is reduced by 3°C.

[0019] Furthermore, the cabinets deployed with temperature and humidity sensors in the computer room are arranged in rows and columns, and a separate air supply and refrigeration method with a hot air duct and a cold air duct is adopted.

[0020] Furthermore, the value range of the empirical K value in step S2 is K ∈ {3, 4, 5}.

[0021] Furthermore, the value of the empirical K value in step S2 is 3, and the value range of the minimum number of included points MinPts in step S3 is MinPts ∈ {3, 4, 5, 6, 7, 8}

[0022] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0023] (1) Use the KMEANS clustering algorithm for rough clustering. By aggregating points with similar temperature changes together for grouping, estimate the approximate density distribution of the correlation degree between sensors and air conditioners according to the grouping density.

[0024] (2) Use the grouping density values Lmin and Lmean obtained by the KMEANS clustering algorithm as the threshold parameters of the clustering radius of the DBSCAN density clustering algorithm.

[0025] (3) Use the DBSCAN density clustering algorithm to cluster the temperature correlation change data set, and apply the aforementioned parameters to group all the temperature correlation change value data sets.

[0026] (4) Use the average correlation degree within the DBSCAN density grouping as the correlation degree between the sensor points and the air conditioner within the grouping, taking into account the spatial similarity of the sensor points arranged in a determinant on the cabinet. Brief Description of the Drawings

[0027] Figure 1 It is a flowchart showing the method for evaluating the influence of precision air conditioners on temperature and humidity sensors in the computer IT room described in the present invention. Detailed Embodiments

[0028] The present invention is further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, modifications of various equivalent forms of the present invention made by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0029] The technical problem addressed by this invention is how to accurately and reasonably calculate the correlation and degree of correlation between each terminal air conditioner and the temperature changes of each temperature and humidity sensor in the computer room. This technical solution eliminates the need for manual on-site surveys and data annotation; the task can be accomplished solely through data mining, significantly reducing workload.

[0030] The basic idea of this invention is to change the temperature setting of each air conditioner in turn by a certain step length, and then obtain the temperature changes of all sensors in the room during the corresponding period. After vectorizing the temperature change data, the K-MEANS clustering algorithm is used to perform preliminary grouping and clustering to obtain the element density of the cluster sub-cluster. This density is used as the clustering radius of the DBSCAN clustering algorithm to perform clustering operations and grouping again, thereby obtaining the influence correlation between each air conditioner and each sensor, thereby realizing the correlation evaluation between temperature and humidity sensors and air conditioning units. Figure 1 Shown are the specific steps of the present invention:

[0031] S1 computer room temperature adjustment

[0032] (1) First, ensure that the temperature and humidity sensors in the computer room and the sensors on the air conditioners are working properly so that the system can continue to receive sensor data normally.

[0033] Then, lower the return air temperature or supply air temperature of all air conditioners in the computer room by 3 degrees. After lowering the return air temperature of each air conditioner, maintain the temperature for 20 minutes before lowering the return air temperature of another air conditioner. Repeat this cycle for all air conditioners. After completing one cycle, multiple cycles may be required to improve calculation accuracy. However, before each cycle, the air conditioner's initial temperature must be returned to its initial setting. Next, check whether the temperature sensor data and the air conditioner temperature setting IOT data have been received properly. If the data is missing or abnormal, continue with the next cycle.

[0034] (2) Sensor temperature data collation

[0035] The subscript numbers for the temperature sensor in the computer room are i: 1→m, and the subscript numbers for the air conditioner in the computer room are j: 1→n;

[0036] Querying and counting from the sensor database can yield:

[0037] The average temperature of the temperature and humidity sensor within a 20 - minute period before a certain air conditioner adjusts the return air temperature is TP ij The average temperature of the temperature and humidity sensor within a 20 - minute period after a certain air conditioner adjusts the return air temperature is TN ij For temperature and humidity sensor i, the average temperature change ΔT within a 20 - minute period before and after a certain air conditioner j adjusts the return air temperature ij = TP ij - TN ij Using the temperature change values of all sensors to construct two - dimensional points in the form of (Δt, 3) respectively with the return air temperature adjustment values of the air conditioner, all two - dimensional point data sample set S, S = {(ΔT ij , 3)...}.

[0038] S2 correlation distribution density detection

[0039] Because the sensors in the computer room are arranged according to the rows and columns of the cabinets, the degree of influence of the sensors in the same row and column by the cold air of the air conditioner has a certain similarity. Therefore, the above - mentioned two - dimensional point set will show a certain characteristic of grouped close proximity in spatial distribution. Using the K - MEANS clustering algorithm and the empirical K value, after clustering operation, all elements of each group in the K groups can be roughly found.

[0040] The specific steps of the K - MEANS clustering algorithm are as follows:

[0041] STEP1 Randomly select K objects from N sample data as the initial clustering centers; STEP2 Calculate the distance from each sample point to each clustering center respectively, and assign them to the cluster with the closest distance one by one;

[0042] STEP3 After all objects are assigned, update the positions of the K class centers. The class center is defined as the mean value of all objects in the cluster in each dimension;

[0043] STEP4 Compare with the K clustering centers obtained in the previous calculation. If the clustering centers change, go to step 2, otherwise go to step 5;

[0044] STEP5 When the class centers no longer change, stop and output the clustering result, and then organize the information we need, such as the class to which each sample belongs, etc., for subsequent statistics and analysis.

[0045] Specifically, in the present invention, based on the K-MEANS clustering algorithm, the K-MEANS clustering algorithm is used on the two-dimensional point set S, and the number K of clustering clusters is set to 3. The formula form is model = K-MEANS(K = 3).fit(S). After clustering, 3 sub-clusters can be obtained, namely s1, s2, and s3. Since the second-dimensional values of all points in these three sub-clusters are 3, all points can be sorted according to the first dimension Δt, arranged in ascending order, and then the distances between two adjacent points are calculated in ascending order. Then, the average distances L1, L2, and L3 of all the nearest adjacent points within each sub-cluster are statistically calculated.

[0046] L1, L2, and L3 are the representative values of the distribution density of the two-dimensional point set in the three sub-clusters respectively. Lmean = (L1 + L2 + L3) / 3, then L is the average density of the two-dimensional point set S, and this density can measure the tightness of the two-dimensional points in S. Lmin = Min(L1, L2, L3), and Lmin is the element density of the sub-cluster with the highest density among the three sub-clusters.

[0047] S3 correlation density clustering

[0048] Due to the distribution of the rows and columns of the sensors in the computer room, the degree of influence of the sensors by the cold air of the air conditioner will show a hierarchical phenomenon. Using the density clustering algorithm and the grouping density obtained in the above steps for re-clustering, sub-clusters with hierarchical influence degrees can be obtained.

[0049] The specific steps of the DBSCAN clustering algorithm are as follows:

[0050] STEP1 Select core points: If the number of points within the ∈-neighborhood of a point exceeds minPTS, mark it as a core point;

[0051] STEP2 Construct neighborhood chains: For each core point, connect all the points (including other core points) within its eps-neighborhood to form a cluster;

[0052] STEP3 Attribution of border points: Assign border points to the cluster of the core point to which they are connected;

[0053] STEP4 Mark noise: Finally, the points that are not assigned to any cluster are marked as noise.

[0054] Specifically, in the present invention, the DBSCAN density clustering algorithm is used on the two-dimensional point set S, the clustering radius parameter ∈ is set to L min , and the minimum number of points per cluster parameter minPTS is set to 3. The formula form is model = DBSCAN(∈ = Lmin, minPTS = 3).fit(S), and then the clustering algorithm is executed.

[0055] After the algorithm is executed, N sub-clusters set can be obtained m (m: 1 → N). For each sub-cluster, the Δt values of its internal elements are relatively close, while the Δt differences between different sub-clusters are large. If using L min as the clustering radius parameter fails to successfully cluster (due to the temperature drop change and the small change in different rounds caused by the indirect air supply of the air duct), then use L mean as the radius parameter for density clustering and perform clustering again.

[0056] Since most of the cabinets in the computer room are arranged in rows and columns, the temperature and humidity sensors installed on each row of cabinets have a certain degree of similarity in the correlation with the temperature adjustment of the same air conditioner. Therefore, calculate the average value MT of Δt within each sub-cluster m , R m = MT m / 3, and use R m as the correlation degree of each point within this sub-cluster.

[0057] Calculation of the correlation degree between the S4 sensor points and the air conditioner

[0058] Finally, for sensor i and air conditioner j, there are multiple ΔT after multiple rounds of temperature adjustment of the air conditioner return air temperature ij,a , and take the average value as mT ij , then it is easy to select the MT ij with the smallest distance from mT m , and the corresponding R m is the influence degree correlation between the sensor i and the air conditioner j.

[0059] The present invention aims at the problem of how to accurately and reasonably calculate the correlation relationship and degree between the temperature changes of each terminal air conditioner in the computer room and each temperature and humidity sensor. Based on the AI algorithm of the present invention, adjusting the supply air temperature or return air temperature of each air conditioner can achieve a basic balance of cold and heat in the temperature field of the computer room, achieving the effect of reducing air conditioner energy consumption. It does not require on-site investigation and data annotation by humans, and can complete the task only through data mining technology, greatly saving the workload.

Claims

1. An evaluation method for the influence of temperature and humidity sensors in a computer IT computer room on precision air conditioners, characterized in that Including the following steps: S1. Adjust the return air temperature of the air conditioners in the computer room one by one, and record the temperature change values of all temperature and humidity sensors after adjustment; perform multiple rounds of adjustment on the return air temperature of the air conditioners, so as to construct a two-dimensional point data sample set of the sensor temperature change values and the return air temperature adjustment values; S2. Based on the K-KEANS clustering algorithm and the empirical K value, using the temperature change value of the sensor as the distance, perform clustering operations on the two-dimensional point data samples in step S1 to obtain K sub-clusters; the points in each sub-cluster are sorted according to the temperature change value, and the distance between adjacent points is calculated, and the average distance L between adjacent points in all sub-clusters is calculated, L = {L1, L2, …, L p , …, L K}; and then obtain the average point density L mean of the two-dimensional point data sample set and the point density L min of the sub-cluster with the highest density. L min = Min{L1, L2, … L p , … L k}; S3. Based on the DBSCAN clustering algorithm, using the point density L of the highest density sub-cluster min as the clustering radius parameter, and setting the minimum number of points MinPts, perform clustering operations to obtain N sub-clusters; if clustering cannot be successfully performed with L min as the clustering radius parameter, then use the average point density L mean and as the clustering radius, and perform clustering again; through the following formula, calculate the association degree R of the points within the sub-cluster after clustering m , R = MTm / t m m / t where, MT m represents the average value of the temperature change within each sub-cluster after clustering, and t represents the adjustment value of the return air temperature of the air conditioner unit during each round of temperature adjustment; R m represents the correlation degree of the points within the sub-cluster; S4, Calculation of the correlation degree between the sensor and the air conditioner: Calculate the minimum value MT of the average temperature change of the sensor numbered i corresponding to the air conditioner numbered j by the following formula m , MT m = min(mT ij ) where, mT ij represents the average temperature change of the sensor with number i and the air conditioner with number j, MT m represents the minimum value of the average temperature change of the sensor; ΔT ij,a represents the temperature change value of the sensor with number i during the adjustment of the air conditioner with number j in the a-th round of adjustment; Finally, take the R m corresponding to MT m as the impact correlation degree between the evaluation sensor i and the air conditioner j.

2. The method for evaluating the influence of precision air conditioners on temperature and humidity sensors in a computer IT computer room according to claim 1, wherein: The return air temperature of the air conditioner described in step S1 is adjusted at least 3 rounds, and the adjustment value t of the return air temperature of the air conditioner unit during each round of temperature adjustment is reduced by 3°C.

3. The method for evaluating the influence of the precision air conditioner on the temperature and humidity sensors in the computer IT machine room according to claim 1, characterized in that: The cabinets with temperature and humidity sensors deployed in the computer room are deployed in rows and columns, and a separate air supply and refrigeration method of hot air channels and cold air channels is adopted.

4. The method for evaluating the influence of the temperature and humidity sensor in the computer IT machine room on the precision air conditioner according to claim 1, wherein: In step S2, the value range of the empirical K value is K ∈ {3, 4, 5}.

5. The method for evaluating the influence of the precision air conditioner on the temperature and humidity sensors in the computer IT room according to claim 1, characterized in that: In step S2, the value of the empirical K value is 3, and in step S3, the value range of the minimum inclusion points MinPts is MinPts ∈ {3, 4, 5, 6, 7, 8}.

Images