An operation data management system

Abstract

The present invention provides an operation data management system, comprising: a monitoring module, an acquisition module, a processing module, a construction module, a grouping module, a rating module, an optimization module, an output module, and an application module. The system of the present invention can effectively and automatically optimize the optimal combination of refrigeration parameters under different external conditions, as well as the optimal operation mode of the refrigeration system, reduce the degree of manual participation, improve the accuracy of the refrigeration system, and ensure the safe operation of the server.

Description

Technical Field

[0001] The invention belongs to the technical field of data management, and in particular relates to an operation data management system. Background Art

[0002] Data centers are facilities that consume a lot of energy and have high requirements for environmental variables. For example, if the temperature is too high, the life of components will be reduced, and if the temperature is too low, the discharge capacity of the battery will also decrease. Earlier, physicists tried to solve this problem by analyzing the structure of the refrigeration system and combining the formula of fluid mechanics, but it is easy to ignore some minor factors in the complex and changeable actual situation, and the modeling process requires sufficient professional knowledge and a huge amount of engineering work.

[0003] At present, the research on methods of controlling temperature and humidity in data centers mainly focuses on balancing loads or changing the topological positions of equipment and air conditioners through CFD simulation. However, actual research has found that the cooling parameters of many super-large computer rooms are not as simple as imagined. For example, indoor air conditioners are usually directly controlled by refrigeration units, and there is no one-to-one correspondence between refrigeration units and air conditioners. This makes CFD simulation difficult - it is impossible to directly set the required parameters and it is difficult to correspond the relationships separately.

[0004] In recent years, deep neural networks have been widely used in various fields. Especially in the case of dealing with systems with huge amounts of data: neural networks have advantages in modeling complex systems because they do not require users to preset the interactive features of the model, but instead look for patterns and interactions in the features, and then automatically generate the best matching model. After obtaining an accurate simulation model, it is necessary to obtain an optimal combination of parameters to make the temperature and humidity most consistent with the standard. At present, it is usually more convenient and accurate to use machine learning algorithms to calculate the optimal solution. The advantage of genetic algorithms is that they can handle constraints well, can jump out of local optimality well, and finally obtain the global optimal solution. They have strong global search capabilities, but conventional algorithms can also lead to inaccurate results. Summary of the invention

[0005] The system of the present invention can effectively and automatically recommend the optimal combination of refrigeration parameters under different external conditions, reduce the degree of manual participation, improve the accuracy of the refrigeration system, and ensure the safe operation of the server.

[0006] The present invention provides an operation data management system for refrigeration management in a data center, the system comprising:

[0007] A monitoring module, used to monitor the real-time data of the data center;

[0008] An acquisition module is used to obtain data center cooling parameters and external environment parameters;

[0009] A processing module, used for preprocessing and formatting the refrigeration parameters and external environment parameters;

[0010] A building module is used to build a cabinet temperature and humidity model based on the cooling parameters of the neural network;

[0011] A grouping module, used to group cabinets in a data center cooling system according to location;

[0012] Rating module, used to build a hierarchical and segmented fitness mechanism to rate the cooling conditions;

[0013] An optimization module, used to iteratively find the best setting parameter combination of the data center refrigeration according to the genetic algorithm, and determine the best operation mode of the refrigeration system;

[0014] An output module, used for combining the optimal operation mode and setting parameters to output data;

[0015] An application module is used to apply the optimal operation mode and setting parameter combination to the data center cooling system.

[0016] The system of the present invention can be applied to the automatic setting optimization of the refrigeration system of the data center, and can effectively and automatically accurately recommend the optimal combination of refrigeration parameters under different external conditions, and is applicable to various types and quantities of refrigeration parameter combinations. While reducing manual participation and solving the time-consuming and labor-intensive problems of traditional manual temperature control, this invention also exceeds the original manual and greedy algorithms in terms of both effect and speed. It can not only reasonably adjust the various parameters of the refrigeration system, but also greatly improve its ability to make its refrigeration conditions meet the standard values, and can be well applied to the actual data center refrigeration system. Moreover, for the setting of the fitness function, in order to better take into account the values ​​that are more different from the standard value, a hierarchical and segmented evaluation function is set on the basis of the neural network model, so that the relationship between distance and rating score can be considered more completely. In addition, the innovation of the calculation formula for inputting the similarity value can also obtain the similarity value more quickly and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below with reference to the accompanying drawings and in combination with examples.

[0018] Figure 1 It is a system framework diagram of the present invention;

[0019] Figure 2 A schematic diagram of formatted input data according to an embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of the effect of a neural network simulation part of an embodiment of the present invention;

[0021] Figure 4 A simplified schematic diagram comparing the optimization effects of the automated setting recommendation of an embodiment of the present invention and the traditional method. DETAILED DESCRIPTION

[0022] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0023] like Figure 1 As shown, an operation data management system of the present invention is used for refrigeration management in a data center, and the system includes:

[0024] A monitoring module, used to monitor the real-time data of the data center;

[0025] An acquisition module is used to obtain data center cooling parameters and external environment parameters;

[0026] A processing module, used for preprocessing and formatting the refrigeration parameters and external environment parameters;

[0027] A building module is used to build a cabinet temperature and humidity model based on the cooling parameters of the neural network;

[0028] A grouping module, used to group cabinets in a data center cooling system according to location;

[0029] Rating module, used to build a hierarchical and segmented fitness mechanism to rate the cooling conditions;

[0030] An optimization module, used to iteratively find the best setting parameter combination of the data center refrigeration according to the genetic algorithm, and determine the best operation mode of the refrigeration system;

[0031] An output module, used for combining the optimal operation mode and setting parameters to output data;

[0032] An application module is used to apply the optimal operation mode and setting parameter combination to the data center cooling system.

[0033] Preferably, the acquisition module is used to acquire the cooling parameters of the data center and the external environment parameters, specifically including:

[0034] A reading unit, used to read the cooling parameters of the data center and the external environment parameters using a Restful interface;

[0035] The feature grouping unit is used to group the data center cooling parameters and external environment parameters according to the actual topological location and operation method and select features and labels.

[0036] Preferably, the processing module is used to pre-process and format the refrigeration parameters and external environment parameters, specifically including:

[0037] A preprocessing unit is used to preprocess the refrigeration parameters and external environment parameters, and extract their main feature information through principal component analysis according to the features and labels.

[0038] A formatting unit, used to format and normalize the main feature information and output a required parameter data set;

[0039] Wherein, the pre-processing unit specifically includes:

[0040] An encoding unit, used for one-hot encoding the non-numerical data items in the obtained refrigeration parameters and external environment parameters;

[0041] The screening unit is used to screen and remove outliers using IForest detection;

[0042] The principal component unit is used for cross-validation using xgboost. The importance of each feature is obtained based on principal component analysis, and the feature data items with low importance are eliminated to retain the main feature information.

[0043] The processing module pre-processes and formats the cooling parameters and external environment parameters. Figure 2 shown.

[0044] Preferably, the construction module is used to construct a cabinet temperature and humidity model based on the refrigeration parameters of the neural network, and specifically includes:

[0045] A classification unit, used to divide the data set into a training set and a test set in a ratio of 5:1;

[0046] The input unit is used to randomly shuffle the order of the training set and input it into the neural network in batches;

[0047] The network model unit is used to build a neural network model consisting of a fully connected layer, a batch_norm layer, an activation layer, and a loss layer;

[0048] Output unit, used to iteratively reduce the loss layer output using Adam optimized gradient descent;

[0049] A testing unit, used to perform model testing using the test set, and if the test fails, adjust the neural network parameters, repeat the operations of the input unit, the model unit, and the output unit, and save the neural network model until the test conditions are met;

[0050] A temperature and humidity model unit, used to construct a cabinet temperature and humidity model based on the cooling parameters of the current neural network;

[0051] Wherein, the network model unit specifically includes:

[0052] Normalization unit, used to input data through the batch_norm layer, where the variance and mean are obtained by directly calculating the data of each batch, and the scaling factor and offset factor are learned by the network;

[0053] A first execution unit is used to input the execution result of the normalization unit into a unit structure of alternating fully connected layers and PReLU activation layers, and repeat N times;

[0054] A second execution unit, used for inputting the execution result of the first execution unit into the fully connected layer and tanh activation layer unit structure;

[0055] A third execution unit, used for inputting the execution result of the second execution unit into a loss layer, wherein the loss function is MSE;

[0056] Wherein, the test unit specifically includes:

[0057] A calculation unit is used to calculate the input similarity value and the output difference value of the data input-output pair of the training set and the data input-output pair of the test set, wherein the calculation of the input similarity value includes:

[0058]

[0059] Among them, m i represents the mean of the data input of the i-th batch of training set, m represents the mean of the data input of the test set, σ i represents the standard deviation of the data input of the i-th batch of training set, and σ represents the standard deviation of the data input of the test set;

[0060] a determination unit, configured to determine whether the following test conditions are met: a. whether the input similarity value is greater than or equal to a similarity threshold, and b. whether the output difference value is less than or equal to a difference threshold;

[0061] The adjustment unit is used to adjust the neural network parameters if there are data batches that do not meet the above conditions, repeatedly execute the operations of the input unit, the network model unit, and the output unit, and save the neural network model until the test conditions are met.

[0062] In the embodiment of the present invention, a cabinet temperature and humidity model is constructed by building a module. The comparison result between the simulated value output by the model and the true value is as follows: Figure 3 As shown, the ordinate represents relative humidity and the abscissa represents humidity points.

[0063] Preferably, the rating module is used to construct a hierarchical and segmented fitness mechanism to rate the refrigeration conditions, specifically including:

[0064] A division unit is used to calculate the absolute difference between the simulated value and the standard value, and divide the threshold intervals into different levels according to the actual situation;

[0065] A division unit, used for calculating a rating score according to the absolute difference and the threshold interval in which the simulation value is located, wherein the rating score decreases faster when the threshold interval is farther from the standard value;

[0066] The weight unit is used to count the rating scores of each cabinet in each group calculated by the division unit, and to assign different weights to calculate the final rating scores of the group and the entire data center cooling system. The weight allocation method is the inverse of the proportion after l2 regularization.

[0067] Preferably, the optimization module is used to iteratively find the optimal setting parameter combination of the data center refrigeration according to the genetic algorithm to determine the optimal operation mode of the refrigeration system, specifically including:

[0068] An initialization unit, used for initializing a population, wherein each dimension in the population represents an adjustable refrigeration parameter, wherein an initialization criterion is based on a mean value of each parameter;

[0069] A generation unit, used for generating the next generation population through selection, crossover and mutation steps of the parent population, wherein the probability of each step depends on the rating score obtained in the rating module;

[0070] An adjustment unit, used for adaptively adjusting the probabilities of the selection, crossover and mutation, wherein the individual retention probability increases linearly with the increase of fitness, i.e., rating score;

[0071] An iterative unit, used for repeatedly executing the generating unit and adjusting unit until an optimization standard or an iteration number is reached;

[0072] The recommendation unit is used to obtain the best recommended setting parameter combination of the data center refrigeration system and determine the best operation mode of the refrigeration system.

[0073] Preferably, wherein N=4.

[0074] like Figure 4 As shown in the figure, the optimization module uses genetic algorithm to iteratively solve the problem, which is significantly better than the traditional algorithm.

[0075] The system of the present invention can be applied to the automatic setting optimization of the refrigeration system of the data center, and can effectively and automatically accurately recommend the optimal combination of refrigeration parameters under different external conditions, and is applicable to various types and quantities of refrigeration parameter combinations. While reducing manual participation and solving the time-consuming and labor-intensive problems of traditional manual temperature control, this invention also exceeds the original manual and greedy algorithms in terms of both effect and speed. It can not only reasonably adjust the various parameters of the refrigeration system, but also greatly improve its ability to make its refrigeration conditions meet the standard values, and can be well applied to the actual data center refrigeration system. Moreover, for the setting of the fitness function, in order to better take into account the values ​​that are more different from the standard value, a hierarchical and segmented evaluation function is set on the basis of the neural network model, so that the relationship between distance and rating score can be considered more completely. In addition, the innovation of the calculation formula for inputting the similarity value can also obtain the similarity value more quickly and accurately.

[0076] The description of the present invention is given for the purpose of illustration and description, and is not intended to be exhaustive or to limit the invention to the disclosed form. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the invention and to enable those of ordinary skill in the art to understand the invention and thereby design various embodiments with various modifications suitable for particular uses.

Claims

1. An operation data management system, characterized in that: The system is used for refrigeration management in a data center, and the system comprises: A monitoring module, used to monitor the real-time data of the data center; An acquisition module is used to obtain data center cooling parameters and external environment parameters; A processing module, used for preprocessing and formatting the refrigeration parameters and external environment parameters; A building module is used to build a cabinet temperature and humidity model based on the cooling parameters of the neural network; A grouping module, used to group cabinets in a data center cooling system according to location; Rating module, used to build a hierarchical and segmented fitness mechanism to rate the cooling conditions; An optimization module, used to iteratively find the best setting parameter combination of the data center refrigeration according to the genetic algorithm, and determine the best operation mode of the refrigeration system; An output module, used for combining the optimal operation mode and setting parameters to output data; An application module for applying the optimal operating mode and setting parameter combination to the data center cooling system; The construction module is used to construct a cabinet temperature and humidity model based on the refrigeration parameters of the neural network, and specifically includes: The classification unit is used to divide the data set into a training set and a test set in a ratio of 5:1; The input unit is used to randomly shuffle the order of the training set and input it into the neural network in batches; The network model unit is used to build a neural network model consisting of a fully connected layer, a batch_norm layer, an activation layer, and a loss layer; Output unit, used to iteratively reduce the loss layer output using Adam optimized gradient descent; A testing unit, used to perform model testing using the test set, and if the test fails, adjust the neural network parameters, repeat the operations of the input unit, the model unit, and the output unit, and save the neural network model until the test conditions are met; The temperature and humidity model unit is used to build a cabinet temperature and humidity model based on the cooling parameters of the current neural network; The rating module is used to construct a hierarchical and segmented fitness mechanism to rate the refrigeration conditions, specifically including: A division unit is used to calculate the absolute difference between the simulated value and the standard value, and divide the threshold intervals into different levels according to the actual situation; A division unit, used for calculating a rating score according to the absolute difference and the threshold interval in which the simulation value is located, wherein the rating score decreases faster when the threshold interval is farther from the standard value; The weight unit is used to count the rating scores of each cabinet in each group calculated by the division unit, and to assign different weights to calculate the final rating scores of the group and the entire data center cooling system. The weight allocation method is the inverse of the proportion after l2 regularization.

2. The system of claim 1, wherein: The acquisition module is used to acquire the cooling parameters of the data center and the external environment parameters, specifically including: A reading unit, used to read the cooling parameters of the data center and the external environment parameters using a Restful interface; The feature grouping unit is used to group the data center cooling parameters and external environment parameters according to the actual topological location and operation method and select features and labels.

3. The system of claim 1, wherein: The processing module is used to pre-process and format the refrigeration parameters and external environment parameters, specifically including: A preprocessing unit is used to preprocess the refrigeration parameters and external environment parameters, and extract their main feature information through principal component analysis according to the features and labels. A formatting unit, used to format and normalize the main feature information and output a required parameter data set; Wherein, the pre-processing unit specifically includes: An encoding unit, used for one-hot encoding the non-numerical data items in the obtained refrigeration parameters and external environment parameters; The screening unit is used to screen and remove outliers using IForest detection; The principal component unit is used for cross-validation using xgboost. The importance of each feature is obtained based on principal component analysis, and the feature data items with low importance are eliminated to retain the main feature information.

4. The system according to claim 1, in, The network model unit specifically includes: Normalization unit, used to input data through the batch_norm layer, where the variance and mean are obtained by directly calculating the data of each batch, and the scaling factor and offset factor are learned by the network; A first execution unit is used to input the execution result of the normalization unit into a unit structure of alternating fully connected layers and PReLU activation layers, and repeat N times; A second execution unit, used for inputting the execution result of the first execution unit into the fully connected layer and tanh activation layer unit structure; A third execution unit, used for inputting the execution result of the second execution unit into a loss layer, wherein the loss function is MSE; Wherein, the test unit specifically includes: A calculation unit is used to calculate the input similarity value and the output difference value of the data input-output pair of the training set and the data input-output pair of the test set, wherein the calculation of the input similarity value includes: Among them, m i represents the mean of the data input of the i-th batch of training set, and m represents the mean of the data input of the test set. Represents the standard deviation of the data input of the i-th batch of training set, Represents the standard deviation of the test set data input; A determination unit, configured to determine whether the following test conditions are met: a. whether the input similarity value is greater than or equal to a similarity threshold, and b. whether the output difference value is less than or equal to a difference threshold; The adjustment unit is used to adjust the neural network parameters if there are data batches that do not meet the above conditions, repeatedly execute the operations of the input unit, the network model unit, and the output unit, and save the neural network model until the test conditions are met.

5. The system of claim 1, wherein: The optimization module is used to iteratively find the optimal setting parameter combination of the data center refrigeration according to the genetic algorithm, and determine the optimal operation mode of the refrigeration system, specifically including: An initialization unit, used for initializing a population, wherein each dimension in the population represents an adjustable refrigeration parameter, wherein an initialization criterion is based on a mean value of each parameter; A generation unit, used for generating the next generation population through selection, crossover and mutation steps of the parent population, wherein the probability of each step depends on the rating score obtained in the rating module; An adjustment unit, used for adaptively adjusting the probabilities of the selection, crossover, and mutation, wherein the individual retention probability increases linearly with the increase of fitness, i.e., rating score; An iterative unit, used for repeatedly executing the generating unit and the adjusting unit until an optimization standard or an iteration number is reached; The recommendation unit is used to obtain the best recommended setting parameter combination of the data center refrigeration system and determine the best operation mode of the refrigeration system.

6. The system of claim 4, wherein: Said N=4.

Images