A remote intelligent operation and maintenance management method and system for a data center

Abstract

The application provides a remote intelligent operation and maintenance management method and system for a data center, and relates to the technical field of data center operation and maintenance. The method comprises the following steps: acquiring state parameter monitoring data of the data center, calculating energy efficiency indexes and screening out strongly correlated characteristic parameters; randomly generating an initial population based on the strongly correlated characteristic parameters, constructing and training an energy efficiency index prediction model based on a long short-term memory network, and determining an optimal characteristic parameter combination using a genetic algorithm based on model training feedback results to apply to the model; and generating an operation and maintenance management work order based on the energy efficiency index prediction value output by the model. The application fully collects various monitoring data to fully consider the influence of environmental, battery and cold and hot channel conditions on energy consumption, and reduces the model calculation amount while fully considering relevant factors by cooperating and setting the long short-term memory network and the genetic algorithm, thereby improving the accuracy of energy efficiency index prediction and providing a basis for subsequent accurate operation and maintenance management.

Description

Technical Field

[0001] This invention relates to the field of data center operation and maintenance technology, specifically to a remote intelligent operation and maintenance management method and system for data centers. Background Technology

[0002] With the rapid development of the electronics and information industry, countless servers, network devices, and data storage devices support efficient and convenient industry application systems. The scale of communication networks is gradually expanding, and the number of communication network devices is constantly increasing. Early data centers face the challenge of optimization and adaptation to subsequent business development. When rearranging servers and other equipment in existing data center racks or installing new racks, the number of devices increases from one to many, making operation and maintenance increasingly complex. To make data center operation and maintenance management more effective, innovation in operation and maintenance methods is essential. Therefore, scientific, reasonable, and efficient operation and maintenance management has become urgent and necessary.

[0003] In the prior art, a data center visualization operation and maintenance management method and system (classification number G06Q) with authorization announcement number "CN114676862B" includes: acquiring equipment and facility information, pipeline information, and spatial layout information of the data center; acquiring a three-dimensional visualization model of the data center; performing equipment deployment planning and cabling planning through the three-dimensional visualization model; acquiring data center operation and maintenance data information; performing energy consumption analysis based on the operation and maintenance data information; formulating intelligent control schemes based on energy consumption analysis; classifying operation and maintenance data; comparing and analyzing data of the same type at different time periods to detect data center faults; acquiring fault type and location information to generate fault warnings and solutions; and annotating and displaying fault warnings and solutions in the three-dimensional visualization model. This method realizes intelligent operation and maintenance of the data center by displaying the data center operation and maintenance status and operation and maintenance data in three dimensions, reducing human resource allocation and improving operation and maintenance efficiency.

[0004] However, existing technologies still have significant shortcomings. For example, when performing energy consumption analysis, existing technologies only use historical energy consumption data to predict future energy consumption data, failing to fully consider the environment, battery, and hot and cold aisle conditions that affect energy consumption. This can lead to a certain deviation between the predicted energy consumption data and the actual value, resulting in a decrease in the accuracy of subsequent operation and maintenance management work orders generated based on the energy consumption analysis results. This wastes human and material resources and is also detrimental to the safe operation of the data center.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a remote intelligent operation and maintenance management method and system for data centers, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A remote intelligent operation and maintenance management method for data centers includes:

[0009] S1, acquires the status parameter monitoring data of the data center during the monitoring period, including energy consumption monitoring data, environmental monitoring data, battery monitoring data and hot and cold aisle monitoring data;

[0010] S2, based on energy consumption monitoring data, calculates the energy efficiency indicators of the data center during the monitoring period, including power usage efficiency, partial power usage efficiency and water usage efficiency;

[0011] S3, extract features from the state parameter monitoring data to construct an initial feature library including multiple feature parameters. The feature extraction includes descriptive statistics and combined interaction.

[0012] S4. The feature parameters are randomly concatenated to obtain multiple combined feature parameters. The correlation between energy efficiency indicators and each feature parameter and combined feature parameter is evaluated based on the Pearson correlation coefficient. Strongly correlated feature parameters and combined feature parameters are selected as strongly correlated feature parameters to construct the final feature library.

[0013] S5. An initial population is randomly generated based on the strongly correlated feature parameters in the final feature library. Each individual in the initial population corresponds to a feature parameter combination. An energy efficiency index prediction model is constructed and trained based on a long short-term memory network. The input of the model is the energy efficiency index and feature parameter combination for multiple consecutive monitoring time periods, and the output is the energy efficiency index for the next monitoring time period. Based on the model training feedback results, a genetic algorithm is used to optimize the feature parameter combination to obtain the optimal feature parameter combination for application to the model, thus completing the model training.

[0014] S6 obtains the predicted values ​​of energy efficiency indicators for future monitoring periods based on the trained energy efficiency indicator prediction model, and generates operation and maintenance management work orders based on the predicted values ​​of energy efficiency indicators.

[0015] Furthermore, the energy consumption monitoring data includes the data center's water consumption, the power consumption of each functional area of ​​the data center, and the power consumption of IT equipment in each functional area. The functional areas of the data center include server rooms, cooling system areas, power distribution rooms, storage areas, network equipment areas, management and maintenance areas, and IT equipment includes servers, storage devices, network devices, and other IT-related equipment.

[0016] The environmental monitoring data includes time-series data of external ambient temperature, temperature, humidity, air particulate matter concentration, and air velocity within the monitoring period;

[0017] The battery monitoring data includes the time-series data of the working temperature, working voltage, charging current, and discharging current of each battery cell during the monitoring period, as well as the cumulative number of charge-discharge cycles and the maximum available capacity of each battery cell at the beginning and end of the monitoring period.

[0018] The monitoring data for the hot and cold aisles includes time-series temperature data at the outlet of the hot and cold aisles, and time-series coolant flow rate and temperature data at the inlet and outlet of the coolant aisles.

[0019] Furthermore, the formula for calculating the energy efficiency index is as follows:

[0020]

[0021] In the formula, To monitor the power consumption of the data center during a specific time period, To monitor the power consumption of IT equipment in the data center during a specific time period, To monitor the power usage efficiency of the data center during the specified time period;

[0022] In the formula, For the monitoring period, the data center's first Power consumption of each functional area For the monitoring period, the data center's first Power consumption of IT equipment in each functional area For the monitoring period, the data center's first Partial power usage efficiency of each functional area. For indexes of functional areas within the data center, and , The number of functional areas within the data center;

[0023] In the formula, To monitor the data center's water consumption during the specified time period, To monitor the water usage efficiency of the data center during the specified time period.

[0024] Furthermore, the steps for constructing the initial feature library are as follows:

[0025] Step S31: Extract energy consumption monitoring data from the status parameter monitoring data as a primary energy consumption feature group, and process the energy consumption monitoring data to obtain a secondary energy consumption feature group to characterize the energy consumption distribution of each functional area in the data center.

[0026] Step S32: Perform descriptive statistics on the environmental monitoring data to obtain environmental statistical feature groups, and combine interactive temperature, humidity and particulate matter concentration to obtain environmental interactive feature groups.

[0027] Step S33: Perform descriptive statistics on the battery monitoring data to obtain a statistical feature set of battery parameters, and combine the interactive current, voltage, charge-discharge cycle count and capacity to obtain an interactive feature set of battery parameters.

[0028] Step S34: Perform descriptive statistics on the monitoring data of hot and cold aisles to obtain statistical feature groups of hot and cold aisles, and combine temperature and coolant flow rate to obtain interactive feature groups of hot and cold aisles.

[0029] Step S35: Statistically analyze the feature parameters in the primary energy consumption feature group, secondary energy consumption feature group, environmental statistical feature group, environmental interaction feature group, battery parameter statistical feature group, battery parameter interaction feature group, hot and cold channel statistical feature group, and hot and cold channel interaction feature group to form an initial feature library.

[0030] Furthermore, the method for randomly concatenating the feature parameters is as follows: randomly select no less than two feature parameters from the initial feature library for random calculation. The random calculation includes, but is not limited to, using addition, subtraction, multiplication, and division to calculate the selected feature parameters.

[0031] Furthermore, the criterion for determining strong correlation is as follows: if the absolute value of the Pearson correlation coefficient between a feature parameter and any energy efficiency index is greater than the correlation threshold, then the feature parameter is considered to be strongly correlated with the energy efficiency index, and this feature parameter is a strongly correlated feature parameter. The method for determining whether a combined feature parameter is a strongly correlated feature parameter is similar.

[0032] Furthermore, each individual in the initial population consists of K gene loci, each corresponding to a strongly correlated feature parameter. If the numerical label of the k-th gene locus in an individual is 1, it indicates that the feature parameter combination corresponding to that individual includes the strongly correlated feature parameter corresponding to the k-th gene locus; conversely, if the numerical label of the k-th gene locus in an individual is 0, it indicates that the feature parameter combination corresponding to that individual does not include the strongly correlated feature parameter corresponding to the k-th gene locus. Here, k is the index of the gene locus, and... K represents the number of strongly correlated feature parameters in the final feature library.

[0033] Furthermore, the method for obtaining the optimal combination of feature parameters is as follows:

[0034] Step S51: Initialize the mutation probability of each gene locus in the individual, and perform data preprocessing on the energy efficiency indicators and strongly correlated characteristic parameters in each historical monitoring period to form a sample set. Data preprocessing includes removing outliers, filling missing values, and data normalization.

[0035] Step S52: Construct an energy efficiency index prediction model based on a long short-term memory network, and set the batch size, feedback training period and total training period. The feedback training period is set between one-tenth and one-fifth of the total training period. The energy efficiency index prediction model includes an input layer, one or more LSTM layers, a fully connected layer and an output layer.

[0036] Step S53: Based on the combination of feature parameters corresponding to each individual, a training set, a validation set, and a test set are randomly selected from the sample set. The training sets corresponding to each individual are input into different energy efficiency index prediction models for training until the predetermined feedback training period is reached. Then, the validation sets corresponding to each individual are used to validate each energy efficiency index prediction model to obtain the model training feedback results. The model training feedback results include mean squared error, mean absolute error, and coefficient of determination.

[0037] Step S54: Based on the model training feedback results, calculate the fitness value of each individual. The higher the fitness value, the better the combination of feature parameters corresponding to that individual. The calculation formula is as follows:

[0038]

[0039] In the formula, , , These are the mean squared error, mean absolute error, and coefficient of determination, respectively. , , All are preset weights, and , , The specific value is determined by the analytic hierarchy process (AHP). This represents the fitness value of an individual.

[0040] Step S55: Sort the individuals according to their fitness values ​​from largest to smallest, select the top 50% of individuals as parents, and perform crossover and mutation operations on each pair of parents to generate offspring. The mutation probability is determined based on the fitness value of the parents and the frequency of occurrence of the digital tags of each gene locus. Calculate the fitness value of the offspring using the same method, and repeat the selection, crossover, and mutation operations on the new population formed by the offspring and parents to generate new parents and offspring until the predetermined number of iterations is reached. Select the individual with the largest fitness value as the optimal individual, and the combination of feature parameters corresponding to the optimal individual is the optimal combination of feature parameters.

[0041] Furthermore, the method for determining the mutation probability is as follows:

[0042] Step S551: Calculate the frequency of occurrence of digital tags for each gene locus in the parent generation. The calculation formula is as follows:

[0043]

[0044] In the formula, This represents the number of times the numeric label of the k-th gene locus is 1 in all parents during the t-th iteration of optimization. This represents the total number of parent generations during the t-th iteration of optimization. This represents the frequency of the k-th gene position having a numeric label of 1 during the t-th iteration of optimization. This represents the number of times the numeric label of the k-th gene locus is 0 in all parents during the t-th iteration of optimization. This represents the frequency of the k-th gene locus having a numeric label of 0 during the t-th iteration of optimization, where t is the index of the iteration number. , To optimize the total number of iterations;

[0045] Step S552: Based on the fitness value of the parent generation, calculate the initial mutation probability of each gene locus in each round of iterative optimization. The calculation formula is as follows:

[0046]

[0047] In the formula, Let be the initial mutation probability of the k-th gene locus during the t-th iteration of optimization. Let the mean fitness of all parent generations be the value at the time of the t-th iteration of optimization. This represents the maximum fitness value of all parent generations during the optimization process from the 1st iteration to the tth iteration. This represents the maximum fitness value of all parent generations during the t-th iteration of optimization. It is a piecewise function. The fitness evaluation threshold, This refers to the random perturbation introduced during the t-th iteration of optimization.

[0048] Step S553: ​​Based on the initial mutation probability and the frequency of occurrence of digital tags at each gene locus in the parent generation, generate the final mutation probability value. The calculation formula is as follows:

[0049]

[0050] In the formula, This is the final value of the mutation probability when the numeric tag of the k-th gene locus is changed from 0 to 1 during the t-th iteration of optimization. Let be the final value of the mutation probability of changing the numeric tag of the k-th gene locus from 1 to 0 during the t-th iteration of optimization.

[0051] A remote intelligent operation and maintenance management system for data centers, used to execute the aforementioned remote intelligent operation and maintenance management method for data centers, comprising:

[0052] The data monitoring module is used to acquire status parameter monitoring data of the data center during the monitoring period, including energy consumption monitoring data, environmental monitoring data, battery monitoring data, and hot and cold aisle monitoring data.

[0053] The energy efficiency index calculation module calculates the data center's energy efficiency index during the monitoring period based on energy consumption monitoring data, including power usage efficiency, partial power usage efficiency, and water usage efficiency.

[0054] The initial feature library construction module is used to extract features from the state parameter monitoring data to construct an initial feature library including multiple feature parameters. The feature extraction includes descriptive statistics and combined interaction.

[0055] The final feature library construction module is used to randomly concatenate feature parameters to obtain multiple combined feature parameters, and evaluate the correlation between energy efficiency indicators and each feature parameter and combined feature parameter based on Pearson correlation coefficient, and select strongly correlated feature parameters and combined feature parameters as strongly correlated feature parameters to construct the final feature library.

[0056] The prediction model building module randomly generates an initial population based on the strongly correlated feature parameters in the final feature library. Each individual in the initial population corresponds to a feature parameter combination. The energy efficiency index prediction model is built and trained based on a long short-term memory network. The input of the model is the energy efficiency index and feature parameter combination for multiple consecutive monitoring time periods, and the output is the energy efficiency index for the next monitoring time period. Based on the model training feedback results, the feature parameter combination is optimized using a genetic algorithm to obtain the optimal feature parameter combination for application to the model, thus completing the model training.

[0057] The operation and maintenance work order generation module obtains the predicted values ​​of energy efficiency indicators for future monitoring periods based on the trained energy efficiency indicator prediction model, and generates operation and maintenance management work orders based on the predicted values ​​of energy efficiency indicators.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] The present invention provides a remote intelligent operation and maintenance management method and system for data centers. This method comprehensively collects various monitoring data to fully consider the impact of environmental, battery, and hot / cold aisle conditions on energy consumption. Then, based on these monitoring data, it obtains characteristic parameters strongly correlated with energy efficiency indicators. Furthermore, by combining long short-term memory networks and genetic algorithms, it obtains the optimal combination of characteristic parameters. Finally, it uses this optimal combination of characteristic parameters as input to a predictive model for energy efficiency indicators. This approach reduces the computational load of the model while comprehensively considering relevant factors, significantly improving the accuracy of energy efficiency indicator prediction and saving computing power costs. This provides a foundation for subsequent precise operation and maintenance management, ensuring the safe and efficient operation of the data center. Attached Figure Description

[0060] Figure 1 This is a schematic flowchart of the overall method of the present invention;

[0061] Figure 2 This is a modular unit diagram of the overall system of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0063] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0064] Example:

[0065] Please see Figure 1 This invention provides a remote intelligent operation and maintenance management method for data centers, comprising:

[0066] Step S1: Obtain the status parameter monitoring data of the data center during the monitoring period, including energy consumption monitoring data, environmental monitoring data, battery monitoring data, and hot and cold aisle monitoring data.

[0067] As one implementation method, energy consumption monitoring data includes data center water consumption, power consumption of each functional area of ​​the data center, and power consumption of IT equipment in each functional area.

[0068] The water consumption of the data center is obtained through water meters. Since the water consumption of the data center mainly comes from the cooling system, the water consumption in the cooling system area can be used as the water consumption of the data center. The water consumption in the cooling system area is read by water meters installed in the cooling system area. The unit of water consumption of the data center is liters.

[0069] The functional areas of a data center include server rooms, cooling system areas, power distribution rooms, storage areas, network equipment areas, and management and maintenance areas. The general division criteria for functional areas are as follows, but the specific division criteria can be adjusted by staff according to the actual situation.

[0070] Server room: It mainly houses all critical IT equipment and is usually the main source of energy consumption in the data center;

[0071] Cooling system area: includes cooling towers and air conditioning equipment, used to maintain the normal operating temperature of the equipment;

[0072] Power distribution room: The area responsible for power distribution and UPS (uninterruptible power supply) equipment, monitoring power quality and power supply stability;

[0073] Storage area: An area specifically used for storage devices, involving data backup and management;

[0074] Network equipment area: A concentrated area for network switches, routers, and other equipment;

[0075] Management and Operations Area: A work area for operations and maintenance personnel to manage and monitor the data center;

[0076] The power consumption of each of the above functional areas is obtained through the data center's own centralized monitoring system or through relevant energy metering equipment such as electricity meters, and the unit of power consumption is set as kilowatt-hour;

[0077] The IT equipment includes servers, storage devices, network devices, and other IT-related equipment. Specifically, servers include, but are not limited to, blade servers, rack servers, and tower servers. Storage devices include, but are not limited to, network-attached storage (NAS), storage area networks (SAN), disk arrays, and other devices used for storing and managing data. Network devices include, but are not limited to, switches, routers, firewalls, and other devices used for data transmission and network connectivity. Other IT-related equipment includes, but is not limited to, load balancers, backup devices, virtualization servers, and other devices that assist IT operations or support IT infrastructure. The method for obtaining the power consumption of IT equipment is the same as the method for obtaining the power consumption of each functional area mentioned above. This is existing technology and will not be elaborated here.

[0078] As one implementation method, the environmental monitoring data includes time-series data of external ambient temperature, temperature of each functional area, humidity, air particulate matter concentration, and air velocity within the monitoring period.

[0079] Among them, temperature time series data, humidity time series data, air particulate matter concentration time series data, and air velocity time series data can be obtained through temperature sensors (such as digital thermometers and thermocouples), humidity sensors (such as thermo-hygrometers), air quality monitors (such as laser particulate meters), and velocity sensors (such as anemometers and heat flow meters) installed in the corresponding functional areas. Of course, they can also be obtained through the data center's own centralized monitoring system. External environmental temperature time series data can be obtained through weather information released by the meteorological bureau or by installing temperature sensors in the external environment.

[0080] As one implementation method, the battery monitoring data includes the operating temperature time-series data, operating voltage time-series data, charging current time-series data, and discharging current time-series data of each battery cell during the monitoring period, as well as the cumulative number of charge-discharge cycles and the maximum available capacity of each battery cell at the beginning and end of the monitoring period. All of the above battery monitoring data can be directly obtained through the battery management system (BMS).

[0081] As one implementation method, the hot and cold aisle monitoring data includes time-series temperature data at the outlet of the hot and cold aisles, and time-series coolant flow rate and temperature data at the inlet and outlet of the coolant aisles.

[0082] It should be noted that the principle of cooling equipment in the data center is as follows: the coolant inlet receives the coolant after cooling by the cooling system. The coolant flows in the coolant pipes and carries away the heat of the air in the hot and cold aisles, thus cooling the air in the hot and cold aisles. The cooled air then flows through the equipment to be cooled to carry away its heat. The air that has carried away the heat of the equipment to be cooled is finally discharged from the hot and cold aisle outlet. The coolant that has carried away the heat of the air in the hot and cold aisles continues to flow until it re-enters the cooling system through the coolant inlet for further cooling.

[0083] Specifically, the timing data of the outlet temperature of the hot and cold aisles can be obtained by temperature sensors installed at the outlet of the hot and cold aisles. This outlet temperature is the temperature of the air. The timing data of the inlet and outlet temperatures of the coolant aisles are obtained in the same way. This inlet and outlet temperature is the temperature of the coolant. Flow meters are installed at both the inlet and outlet of the coolant aisles to obtain the timing data of the coolant flow rate at the inlet and outlet of the coolant respectively.

[0084] It should be noted that the specific monitoring time period is set by the staff according to the actual situation, such as 30 minutes, 1 hour, 2 hours, 6 hours, etc. There is no restriction here. When collecting the above time series data, different types of parameters can be collected at the same collection frequency to unify the timestamp, or different types of parameters can be collected at different collection frequencies and then unified by interpolation or data aggregation methods to facilitate subsequent processing and analysis.

[0085] Step S2: Based on energy consumption monitoring data, calculate the data center's energy efficiency indicators during the monitoring period, including power usage efficiency, partial power usage efficiency, and water usage efficiency. The calculation formula is as follows:

[0086]

[0087] In the formula, To monitor the power consumption of the data center within a specific time period, specifically the sum of the power consumption of each functional area within that period. To monitor the power consumption of IT equipment in the data center during a specific time period, specifically the sum of the power consumption of IT equipment in each functional area during the monitoring period. To monitor the power usage efficiency of the data center within a specific time period, The ideal value is 1, meaning that all the electricity in the data center is used for IT equipment with no additional energy consumption. However, in reality, it is affected by cooling and other auxiliary systems. It will generally be higher than 1, usually between 1.5 and 2.5, while The higher the value, the lower the efficiency of the data center's power usage, and the more necessary the operation and maintenance adjustments are.

[0088] In the formula, For the monitoring period, the data center's first The power consumption of each functional area is specifically the power consumption of the [number]th functional area within the monitoring period. The sum of the power consumption of each functional area For the monitoring period, the data center's first The power consumption of IT equipment in each functional area, specifically within the monitoring period, is as follows: The sum of the power consumption of IT devices in each functional area For the monitoring period, the data center's first Partial power usage efficiency of each functional area. The ideal value is 1, meaning that the data center's first... All electricity within each functional area is used for the IT equipment in that area, with no additional energy consumption. The higher the value, the better the data center's performance. The lower the efficiency of power use in a functional area, the more necessary operation and maintenance adjustments are. For indexes of functional areas within the data center, and , The number of functional areas within the data center;

[0089] In the formula, To monitor the data center's water consumption during the specified time period, To monitor the water usage efficiency of the data center over a specific period of time, The lower the value, the higher the water resource utilization efficiency of the data center while maintaining performance. Similarly... The higher the value, the lower the water resource utilization efficiency of the data center during the monitoring period, and the more necessary the operation and maintenance adjustments are.

[0090] Step S3 involves extracting features from the state parameter monitoring data to construct an initial feature library containing multiple feature parameters. Feature extraction includes descriptive statistics and combined interactions, as detailed below:

[0091] Step S31: Extract energy consumption monitoring data from the status parameter monitoring data as a primary energy consumption feature group, and process the energy consumption monitoring data to obtain a secondary energy consumption feature group to characterize the energy consumption distribution of each functional area within the data center. The primary and secondary energy consumption feature groups are represented as follows:

[0092]

[0093] In the formula, The first-level energy consumption characteristic group is used to characterize the power consumption of each functional area of ​​the data center, the power consumption of IT equipment in each functional area, and the water consumption of the data center during the monitoring period. The characteristics in the first-level energy consumption characteristic group are an important basis for calculating energy efficiency indicators, and they provide an analytical and predictive basis for the model to predict future energy efficiency indicators in the following text.

[0094]

[0095] In the formula, This is a secondary energy consumption characteristic group, used to characterize the proportion of power consumption in each functional area within the total power consumption of the data center during the monitoring period, as well as the proportion of power consumption of IT equipment in each functional area within the total power consumption of IT equipment in the data center. By using the characteristics in this secondary energy consumption characteristic group, the energy consumption distribution of each functional area within the data center can be characterized, providing an analytical and predictive basis for the subsequent model prediction of future energy efficiency indicators. The larger it is, the more likely it is to be the first The higher the proportion of power consumption in each functional area within the total power consumption of the data center, the better. The larger it is, the more likely it is to be the first The higher the proportion of IT equipment in each functional area in the total power consumption of data center IT equipment;

[0096] Step S32: Descriptive statistics are performed on the environmental monitoring data to obtain environmental statistical feature groups, and the interaction of temperature, humidity, and particulate matter concentration is combined to obtain environmental interaction feature groups. The representations of the environmental statistical feature groups and the environmental interaction feature groups are as follows:

[0097]

[0098] In the formula, These represent the mean, maximum, minimum, and standard deviation of the external ambient temperature during the monitoring period. The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of temperature in each functional area. The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of humidity in each functional area. The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of air particulate matter concentration in each functional area. The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of airflow velocity in each functional area. This is an environmental statistical feature group, used to characterize the environmental conditions affecting the operation of the data center during the monitoring period;

[0099]

[0100] In the formula, Used to characterize the monitoring period, the first The greater the temperature difference between the functional area and the external environment, the more pronounced the temperature difference. Each functional area maintains temperature The more energy required... Used to characterize the effects of temperature and humidity on the first [unclear] during the monitoring period. The impact of temperature and humidity on equipment performance in each functional area is often interdependent, so the product of the two is used as a combined feature to capture this complex relationship. Used to characterize the effects of temperature and air particulate matter concentration on the first [unclear] within the monitoring period. The impact of equipment in each functional area is significant because increased temperature typically leads to an increase in the concentration of particulate matter in the air. Particulate matter affects the scattering and absorption of heat radiation, thus influencing the temperature of the local area. Both factors negatively impact equipment performance. Therefore, the product of these two factors is used as a combined feature to capture this complex relationship. This is the environmental interaction feature group, which is used to characterize the interactive effects of temperature, humidity, and air particulate matter concentration on equipment operation during the monitoring period.

[0101] Step S33: Descriptive statistics are performed on the battery monitoring data to obtain a statistical feature set of battery parameters. This is combined with the interaction current, voltage, charge / discharge cycle count, and capacity to obtain an interaction feature set of battery parameters. The representations of the statistical feature set and the interaction feature set of battery parameters are as follows:

[0102]

[0103] In the formula, The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of the operating temperature of each individual battery cell. The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of the operating voltage of each individual battery cell. The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of the charging current of each individual battery cell. The number of monitoring periods is respectively The mean, maximum, minimum, and standard deviation of the discharge current of each individual battery cell. This is a statistical feature set of battery parameters, used to characterize the operating status of each battery cell during the monitoring period. This is an index for individual battery cells within the data center, and , The number of individual battery cells within the data center;

[0104]

[0105] In the formula, , , , The first The rated operating temperature, rated operating voltage, rated charging current, and rated discharging current of each battery cell. The larger the deviation from 1, the more significant the deviation within the monitoring period. The less suitable the operating temperature of each individual battery cell is, the worse it is. , , The larger the deviation from 1, the more significant the deviation within the monitoring period. The less suitable the working voltage, charging current, and discharging current of a single battery cell are, the more suitable the specific values ​​of the rated operating temperature, rated operating voltage, rated charging current, and rated discharging current can be obtained from the product specification sheet of that battery cell.

[0106] In the formula, Used to characterize the first [item] within the monitoring period. The difference in charging and discharging current among individual battery cells: if the discharging current is significantly lower than the charging current, it indicates an increase in internal impedance or a decrease in capacity, thus indicating a more severe aging of the battery cell. Therefore, by... To quantify the number of monitoring periods within the specified time frame The differences in charging and discharging current of individual battery cells, and The larger the value, the more severe the aging of the j-th battery cell and the shorter its remaining usable lifespan.

[0107] In the formula, The first The maximum usable capacity of each individual battery cell at the start and end of the monitoring period. Used to characterize the The smaller the value of the capacity loss of a single battery cell during the monitoring period, the better the capacity loss of the first battery cell. The more capacity a single battery cell loses during the monitoring period, the more severe the aging process. For the first The rated capacity of each battery cell can be obtained from the product specification sheet for that battery cell. Used to characterize the The percentage of usable capacity of a single battery cell after a monitoring period; the higher the value, the better the capacity of the cell. The smaller the lost capacity of a single battery cell after a monitoring period, the longer the usable lifespan of that battery cell.

[0108] In the formula, The first The cumulative number of charge-discharge cycles for each individual battery cell at the start and end of the monitoring period. Used to characterize the The number of charge-discharge cycles performed by a single battery cell within the monitoring period; the higher the value, the better the number of cycles performed. The higher the charge / discharge intensity of a single battery cell during the monitoring period, the shorter its remaining usable lifespan. For the first The rated charge-discharge cycle life of each battery cell can be obtained from the product specification sheet for that battery cell. Used to characterize the The remaining usable charge-discharge cycles of a single battery cell after a monitoring period; the larger the value, the better the remaining charge-discharge cycle count. The worse the remaining charge-discharge cycle status of a single battery cell after a monitoring period, the shorter its remaining usable lifespan.

[0109] In the formula, Used to characterize the The remaining usable life of each battery cell after a monitoring period is assessed. Since both battery capacity and charge / discharge cycle count can be used to quantify the remaining usable life, combining both provides a comprehensive evaluation of the battery cell, avoiding the limitations of a single assessment. Furthermore, as discussed earlier... The larger the value, the longer the remaining usable lifetime. The smaller the value, the shorter the remaining usable lifetime, therefore adopting... The remaining usable life of a battery is assessed in the form of a value, with a higher value indicating a shorter remaining usable life.

[0110] In the formula, This is a battery parameter interaction feature group, which is used to combine interactive current, voltage, charge-discharge cycle count and capacity to characterize the health status of the battery during operation;

[0111] Step S34: Descriptive statistics are performed on the hot and cold aisle monitoring data to obtain the hot and cold aisle statistical feature group, and temperature and coolant flow rate are combined to obtain the hot and cold aisle interaction feature group. The representations of the hot and cold aisle statistical feature group and the hot and cold aisle interaction feature group are as follows:

[0112]

[0113] In the formula, These represent the mean, maximum, minimum, and standard deviation of the outlet temperature of the hot and cold aisles during the monitoring period. These represent the mean, maximum, minimum, and standard deviation of the coolant channel inlet temperature during the monitoring period. These represent the mean, maximum, minimum, and standard deviation of the coolant channel outlet temperature during the monitoring period. These represent the mean, maximum, minimum, and standard deviation of the coolant flow rate at the coolant channel inlet during the monitoring period. These represent the mean, maximum, minimum, and standard deviation of the coolant flow rate at the coolant channel outlet during the monitoring period. This is a statistical feature group for hot and cold aisles, which is used to characterize the working status of the cooling system during the monitoring period.

[0114]

[0115] In the formula, This is used to characterize the temperature difference between the inlet and outlet of the coolant channel during the monitoring period. Within a reasonable range, a larger temperature difference indicates more thorough heat exchange, but an excessively large temperature difference also indicates insufficient cooling capacity of the cooling system. This value characterizes the evaporation or leakage of coolant in the coolant channels during the monitoring period. A higher value indicates greater coolant evaporation during flow, suggesting insufficient cooling capacity of the cooling system. If the value is excessively high and exceeds the normal range, it indicates a risk of coolant leakage in the channels. This value is used to characterize the heat exchange effect during the monitoring period. The closer the value is to 0, the more thorough the heat exchange and the higher the heat exchange efficiency. This value is used to characterize the thermal contamination at the hot and cold aisle outlets during the monitoring period. The higher the value, the more severe the thermal contamination at the hot and cold aisle outlets, and the greater the need to increase the cooling capacity of the refrigeration system. This is a cold and hot aisle interaction feature group, which is used to combine interaction temperature and coolant flow rate to characterize the working state of the cooling system;

[0116] Step S35: Collect the feature parameters from the primary energy consumption feature group, secondary energy consumption feature group, environmental statistical feature group, environmental interaction feature group, battery parameter statistical feature group, battery parameter interaction feature group, hot and cold channel statistical feature group, and hot and cold channel interaction feature group to construct an initial feature library. The feature parameters are the elements in the feature group, such as the elements in the primary energy consumption feature group. These are the feature parameters. Through the settings in steps S31-S35, multiple feature parameters affecting energy efficiency index values ​​are summarized, laying the foundation for further expansion of the feature library and model training optimization in the following sections.

[0117] S4. The feature parameters are randomly concatenated to obtain multiple combined feature parameters. The correlation between energy efficiency indicators and each feature parameter and combined feature parameter is evaluated based on the Pearson correlation coefficient. Strongly correlated feature parameters and combined feature parameters are selected as strongly correlated feature parameters to construct the final feature library.

[0118] The method for randomly concatenating feature parameters is as follows: at least two feature parameters are randomly selected from the initial feature library for random calculation. Random calculation includes, but is not limited to, using addition, subtraction, multiplication, and division to randomly calculate the selected feature parameters. That is to and A combined feature parameter obtained by random splicing can evaluate the first [unit / item] within the monitoring period. The greater the stability of temperature in a functional area relative to the external ambient temperature, the better the stability within the monitoring period. The temperature in each functional area is more stable than the external environment temperature. By randomly splicing the feature parameters to obtain the combined feature parameter settings, the range of feature selection is expanded, which facilitates the comprehensive analysis of relevant data when modeling and predicting energy efficiency indicators in the later text, and also reduces the workload of staff in manually constructing feature parameters.

[0119] The criteria for determining strong correlation are as follows: if the absolute value of the Pearson correlation coefficient between a feature parameter and any energy efficiency indicator (such as power usage efficiency, partial power usage efficiency, or water usage efficiency) is greater than the correlation threshold, then this feature parameter is considered to be strongly correlated with the energy efficiency indicator. This feature parameter is a strongly correlated feature parameter. The specific value of the correlation threshold is set by the staff according to the actual situation. For example, considering that the energy efficiency indicators of data centers are affected by multiple factors, in order to fully consider the correlation characteristics, the correlation threshold can be set to a lower range of 0.1-0.3. The method for determining whether a combined feature parameter is a strongly correlated feature parameter is the same, and will not be elaborated here.

[0120] It should be noted that step S4 first randomly splices together multiple combined feature parameters, and then selects strongly correlated feature parameters from the feature parameters and combined feature parameters based on the Pearson correlation coefficient. This results in high-quality feature parameters that are strongly correlated with energy efficiency indicators, laying the foundation for determining the model input in the subsequent process.

[0121] Furthermore, to ensure the richness of strongly correlated feature parameters in the final feature library, the total number of strongly correlated feature parameters in the final feature library is generally set to more than 1.5 times the total number of feature parameters in the initial feature library. The specific upper limit can be set by the staff according to the actual situation, but it should generally not exceed 3 times to avoid excessive computation.

[0122] Step S5: Randomly generate an initial population based on the strongly correlated feature parameters in the final feature library. Each individual in the initial population corresponds to a feature parameter combination. Construct and train an energy efficiency index prediction model based on a long short-term memory network. The input of the model is the energy efficiency index and feature parameter combination for multiple consecutive monitoring time periods, and the output is the energy efficiency index for the next monitoring time period. Based on the model training feedback results, use a genetic algorithm to optimize the feature parameter combination to obtain the optimal feature parameter combination for application in the model, thus completing the model training.

[0123] In this initial population, each individual consists of K gene loci, each corresponding to a strongly correlated feature parameter. If the numeric label of the k-th gene locus in an individual is 1, it means that the feature parameter combination corresponding to that individual includes the strongly correlated feature parameter corresponding to the k-th gene locus; conversely, if the numeric label of the k-th gene locus in an individual is 0, it means that the feature parameter combination corresponding to that individual does not include the strongly correlated feature parameter corresponding to the k-th gene locus. Here, k is the index of the gene locus, and... K represents the number of strongly correlated feature parameters in the final feature library;

[0124] In the construction of an energy efficiency index prediction model based on a long short-term memory network, experiments are generally conducted on different window lengths during training. The performance on the validation set is used to determine how many consecutive monitoring time periods to use to predict the data for the next monitoring time period. This is existing technology and will not be elaborated here. Of course, the number of consecutive monitoring time periods to use to predict the data for the next monitoring time period can also be set manually. For example, if 8-12 consecutive monitoring time periods are used to predict the data for the next monitoring time period, taking 10 consecutive monitoring time periods as an example, the input of the energy efficiency index prediction model is the combination of energy efficiency index and feature parameters of each monitoring time period in the 10 consecutive monitoring time periods, and the output is the energy efficiency index of the next monitoring time period after the 10 consecutive monitoring time periods. Of course, the specific number of consecutive monitoring time periods can also be other values, which are not limited here.

[0125] The method for obtaining the optimal combination of feature parameters is as follows:

[0126] Step S51: Initialize the mutation probability of each gene locus in the individual, and perform data preprocessing on the energy efficiency indicators and strongly correlated characteristic parameters in each historical monitoring period to form a sample set. Data preprocessing includes removing outliers, filling missing values, and data normalization.

[0127] The mutation probability of each gene locus after initialization is between 0.01 and 0.1, in order to balance population diversity and avoid excessive randomness in referencing.

[0128] Among them, outliers can be removed using conventional algorithms such as Z-Score, IQR, or box method, and missing values ​​can be filled using conventional algorithms such as mean filling or median filling. However, the optimal choice is to fill with the mean of the previous and next monitoring periods to conform to the time sequence structure logic. Data normalization can be achieved using conventional algorithms such as max-min normalization or Robust normalization, which are existing technologies and will not be elaborated here.

[0129] Step S52: Construct an energy efficiency index prediction model based on a long short-term memory network, and set the batch size, feedback training period and total training period. The feedback training period is set between one-tenth and one-fifth of the total training period. The energy efficiency index prediction model includes an input layer, one or more LSTM layers, a fully connected layer and an output layer.

[0130] The input layer is used to receive energy efficiency indicators and feature parameter combinations for multiple consecutive monitoring time periods. The LSTM layer is used to perform feature processing on the data received by the input layer. The fully connected layer is used to map the output of the LSTM layer to the predicted value. The output layer is used to output the energy efficiency indicators for the next monitoring time period.

[0131] The batch size is the number of samples used in each training iteration, typically between 32 and 256. The total training cycle is the total number of training iterations, typically between 30 and 100. The feedback training cycle is set to determine the time point for obtaining model training feedback results. If this time point is too early, the model training feedback results will lack reference value, while if it is too late, the total model training time will be too long. Therefore, the feedback training cycle is set between one-tenth and one-fifth of the total training cycle to balance the reference value of model training feedback results and the efficiency of model training.

[0132] Step S53: Based on the combination of feature parameters corresponding to each individual, a training set, a validation set, and a test set are randomly selected from the sample set. The training sets corresponding to each individual are input into different energy efficiency index prediction models for training until the predetermined feedback training period is reached. Then, the validation sets corresponding to each individual are used to validate each energy efficiency index prediction model to obtain the model training feedback results. The model training feedback results include mean squared error, mean absolute error, and coefficient of determination.

[0133] It should be noted that the model training process and the calculation of mean squared error, mean absolute error and coefficient of determination are all conventional techniques in this technical field, and will not be elaborated here.

[0134] Step S54: Based on the model training feedback results, calculate the fitness value of each individual. The higher the fitness value, the better the combination of feature parameters corresponding to that individual. The calculation formula is as follows:

[0135]

[0136] In the formula, , , These are mean squared error, mean absolute error, and coefficient of determination, respectively. The smaller the mean squared error and mean absolute error, the better the performance of the energy efficiency index prediction model after the feedback training cycle. The closer the coefficient of determination is to 1, the better the performance of the energy efficiency index prediction model after the feedback training cycle. Therefore, mean squared error, mean absolute error, and coefficient of determination are used to jointly characterize the performance of the energy efficiency index prediction model.

[0137] In the formula, , , All values ​​are preset weights, representing the importance of mean squared error, mean absolute error, and coefficient of determination in the performance evaluation of energy efficiency index prediction models. The larger the value, the higher the importance of the corresponding mean squared error, mean absolute error, or coefficient of determination in the performance evaluation of energy efficiency index prediction models. , , The specific value is determined by the analytic hierarchy process (AHP), as follows:

[0138] The relative importance of three evaluation factors—mean squared error, mean absolute error, and coefficient of determination—to the performance evaluation of the energy efficiency index prediction model was determined pairwise using the nine-scale method, in order to construct a judgment matrix. The indices of mean squared error, mean absolute error, and coefficient of determination were labeled 1, 2, and 3, respectively. The constructed judgment matrix is ​​as follows:

[0139]

[0140] In the formula, Both represent the indexes of the evaluation factors, and This indicates the relative importance of the evaluation factor with index x to the performance evaluation of the energy efficiency index prediction model, relative to the evaluation factor with index y. The specific value was determined by relevant experts using a 1-9 scoring method. This indicates that the evaluation factor with index x is more important to the performance evaluation of the energy efficiency index prediction model than the evaluation factor with index y. This indicates that the evaluation factor with index x has a lower relative importance to the performance evaluation of the energy efficiency index prediction model compared to the evaluation factor with index y.

[0141] Divide each element value in the judgment matrix by the sum of its columns to obtain a normalized judgment matrix. Calculate the mean of the element values ​​in each row of the normalized judgment matrix. Use the mean of the first row, the mean of the second row, and the mean of the third row as the scaling factors for the mean squared error, the mean absolute error, and the coefficient of determination, respectively. With the constraint that the sum of the scaled values ​​equals 1, scale the three scaling factors proportionally and use the scaled values ​​as the weights of the corresponding evaluation factors.

[0142] In the formula, The fitness value of an individual is a comprehensive evaluation factor consisting of three evaluation factors: mean squared error, mean absolute error, and coefficient of determination, used to assess the performance of the energy efficiency index prediction model. The larger the fitness value, the better the combination of characteristic parameters corresponding to that individual, and the more suitable it is as the input parameter for the energy efficiency index prediction model.

[0143] It should be noted that the smaller the mean squared error and mean absolute error, the better the performance of the energy efficiency index prediction model after the feedback training cycle. Similarly, the closer the coefficient of determination is to 1, the better the performance of the energy efficiency index prediction model after the feedback training cycle. Therefore, a weighted summation method is used to... As the main component of the fitness value, it is used to comprehensively evaluate the performance of the energy efficiency index prediction model. The setting is used to quantify the degree of deviation between the coefficient of determination and the ideal value of 1, and because The larger the value, the worse the performance of the energy efficiency index prediction model. Therefore, [the following is implied:] The denominator is 1, and the numerator is 1, which is used to establish the function expression of the fitness value. The addition of 1 to the denominator is to avoid the case where the denominator is 0. This ensures that the final fitness value function expression meets the logical requirement that the smaller the mean square error and mean absolute error, and the closer the coefficient of determination is to 1, the larger the fitness value. Of course, other reasonable function expressions are also acceptable and are not restricted here.

[0144] Step S55: Sort the individuals according to their fitness values ​​from largest to smallest, select the top 50% of individuals as parents, and perform crossover and mutation operations on each pair of parents to generate offspring. The mutation probability is determined based on the fitness value of the parents and the frequency of occurrence of the digital tags of each gene locus. Calculate the fitness value of the offspring using the same method, and repeat the selection, crossover, and mutation operations on the new population formed by the offspring and parents to generate new parents and offspring until the predetermined number of iterations is reached. Select the individual with the largest fitness value as the optimal individual, and the combination of feature parameters corresponding to the optimal individual is the optimal combination of feature parameters.

[0145] The method for determining the mutation probability is as follows:

[0146] Step S551: Calculate the frequency of occurrence of digital tags for each gene locus in the parent generation. The calculation formula is as follows:

[0147]

[0148] In the formula, This represents the number of times the numeric label of the k-th gene locus is 1 in all parents during the t-th iteration of optimization. This represents the total number of parent generations during the t-th iteration of optimization. This represents the frequency of the k-th gene locus having a numeric label of 1 during the t-th iteration of optimization. A larger value indicates a higher proportion of individuals in the parent generation with a numeric label of 1 for the k-th gene locus. Since the parent generation consists of superior individuals in the population, the probability that the feature parameter corresponding to the k-th gene locus is the optimal feature parameter is also higher. This represents the number of times the numeric label of the k-th gene locus is 0 in all parents during the t-th iteration of optimization. This represents the frequency of the k-th gene locus having a numeric label of 0 during the t-th iteration of optimization. A larger value indicates a higher proportion of individuals in the parent generation with a numeric label of 0 for the k-th gene locus. Since the parent generation consists of relatively superior individuals in the population, the probability that the feature parameter corresponding to the k-th gene locus is the optimal feature parameter is lower. t is the index of the iteration number, and... , To optimize the total number of iterations;

[0149] Step S552: Based on the fitness value of the parent generation, calculate the initial mutation probability of each gene locus in each round of iterative optimization. The calculation formula is as follows:

[0150]

[0151] In the formula, This represents the initial mutation probability of the k-th gene locus during the t-th iteration of optimization. A larger value indicates a greater probability that the numeric label of the k-th gene locus will change from 0 to 1 or from 1 to 0 during the t-th iteration. Similarly... Let be the initial mutation probability of the k-th gene locus during the (t-1)-th iteration of optimization, where This represents the initial mutation probability of the k-th gene locus after initialization in step S51.

[0152] In the formula, Let the mean fitness of all parent generations be the value at the time of the t-th iteration of optimization. This represents the maximum fitness value of all parent generations during the optimization process from the 1st iteration to the tth iteration. This represents the maximum fitness value of all parent generations during the t-th iteration of optimization.

[0153] It should be noted that, The larger the value, the closer the selected parent generation is to the historical optimum during the t-th iteration of optimization. This means the quality of the selected parent generation is higher during the t-th iteration. Therefore, a lower mutation probability should be set to avoid introducing too much randomness and ensure the quality of the offspring generated during the t-th iteration. Similarly, if... The smaller the value, the lower the quality of the parent generation selected during the t-th iteration of optimization. A higher mutation probability should be set to introduce higher randomness to ensure the diversity of offspring, thereby searching for the optimal solution in a larger space.

[0154] Based on this, The larger the value, the closer the average fitness value of the selected parent generation is to the maximum value during the t-th iteration of optimization. This indicates that the quality of the selected parent generation is more uniform during the t-th iteration. Conversely, if the value is smaller... The smaller the value, the more uneven the quality of the parent generation selected during the t-th iteration of optimization;

[0155] Therefore in and When the values ​​are all relatively high, it indicates that the selected parent generation is of high and relatively uniform quality during the t-th iteration of optimization. This means that the selected parent generation is basically close to the optimal solution and there is not much room for further exploration. Therefore, by... To set a lower mutation probability to avoid introducing too much invalid randomness, while higher and A lower value indicates that the selected parent generation in the t-th iteration is of high quality but uneven, meaning that while the selected parent generation is good, there is still room for further optimization. Similarly, using... To set a slightly lower mutation probability to introduce an appropriate amount of randomness;

[0156] On the contrary A higher value indicates that the selected parent generation during the t-th iteration of optimization is of poor quality and relatively uniform, meaning that the selected parent generation is far from the optimal solution and lacks diversity, indicating a large space for further exploration. Therefore, by... To introduce a large amount of randomness by setting a high mutation probability, thereby searching for the optimal solution in a larger space, while... and When the values ​​are all low, it indicates that the quality of the selected parent generation is low and uneven during the t-th iteration of optimization. This means that although the selected parent generation is far from the optimal solution, it has high diversity. Therefore, by... To set a slightly higher mutation probability to appropriately enrich the randomness;

[0157] Therefore, a piecewise function is set. and with The mutation probability of the t-th iteration optimization is obtained in the form of , where The fitness evaluation threshold is used to evaluate... The degree of excellence, when When, it indicates that the parent generation selected during the t-th iteration of optimization is of excellent quality, while when When the value is t, it indicates that the parent generation selected during the t-th iteration of optimization is of poor quality. The specific value of the fitness evaluation threshold should be set by the staff according to the actual situation. For example, the fitness evaluation threshold can be set between 0.4 and 0.7. It should be noted that setting the fitness evaluation threshold too high will introduce too much randomness, while setting it too low will easily lead to getting trapped in a local optimum. It is recommended to adjust it dynamically according to the actual situation.

[0158] In the formula, This is a random perturbation introduced during the t-th iteration of optimization. Its value ranges from 0.8 to 1.2, and it is used to randomly change the mutation probability to ensure randomness. Other values ​​close to 1 are also acceptable; no restrictions are imposed here.

[0159] Step S553: ​​Based on the initial mutation probability and the frequency of occurrence of digital tags at each gene locus in the parent generation, generate the final mutation probability value. The calculation formula is as follows:

[0160]

[0161] In the formula, This is the final value of the mutation probability when the numeric tag of the k-th gene locus is changed from 0 to 1 during the t-th iteration of optimization. The larger the value, the greater the probability that the characteristic parameter corresponding to the k-th gene locus is the optimal characteristic parameter. Therefore, to guide the generation of high-quality offspring, use... right Make corrections to The final value of the mutation probability of changing the numeric label of the k-th gene locus from 0 to 1 during the t-th iteration of optimization is calculated in the form of . The larger, The larger the value, the higher the probability that the numerical tag of the k-th gene position will be changed from 0 to 1 during the t-th iteration of optimization, thus achieving the technical effect of guiding the generation of high-quality offspring. To determine the final mutation probability of changing the numeric tag of the k-th gene locus from 1 to 0 during the t-th iteration of optimization, based on... To calculate Similarly, this will not be elaborated upon here;

[0162] It should be noted that the final energy efficiency index prediction model takes energy efficiency indices and optimal feature parameter combinations from multiple consecutive monitoring periods as input and outputs energy efficiency indices from the next monitoring period. The existing technology of randomly selecting training, validation and test sets from the sample set based on the determined optimal feature parameter combinations for model training is not described in detail here.

[0163] Step S6: Obtain the predicted values ​​of energy efficiency indicators for future monitoring periods based on the trained energy efficiency indicator prediction model, and generate operation and maintenance management work orders based on the predicted values ​​of energy efficiency indicators.

[0164] The method for obtaining the predicted energy efficiency index value for future monitoring periods is as follows: input the energy efficiency index and the optimal feature parameter combination of the current monitoring period and several consecutive monitoring periods before it into the trained model to obtain the predicted energy efficiency index value for future monitoring periods.

[0165] It should be noted that generating operation and maintenance management work orders based on predicted energy efficiency indicators can utilize existing technologies. For example, power efficiency is generally between 1.5 and 2.5. When power efficiency exceeds 2.5, it usually means that the data center is overloaded or the cooling system is malfunctioning. Therefore, if the predicted power efficiency exceeds 2.5 in the future monitoring period, operation and maintenance management work orders to check the data center load and cooling system operation will be automatically generated. This will help to detect and address problems in advance and ensure the safe and efficient operation of the data center.

[0166] Example 2:

[0167] Please see Figure 2 This embodiment provides a remote intelligent operation and maintenance management system for data centers, used to execute the remote intelligent operation and maintenance management method for data centers described in Embodiment 1 above, including:

[0168] The data monitoring module is used to acquire status parameter monitoring data of the data center during the monitoring period, including energy consumption monitoring data, environmental monitoring data, battery monitoring data, and hot and cold aisle monitoring data.

[0169] The energy efficiency index calculation module calculates the data center's energy efficiency index during the monitoring period based on energy consumption monitoring data, including power usage efficiency, partial power usage efficiency, and water usage efficiency.

[0170] The initial feature library construction module is used to extract features from the state parameter monitoring data to build an initial feature library that includes multiple feature parameters. Feature extraction includes descriptive statistics and combined interactions.

[0171] The final feature library construction module is used to randomly concatenate feature parameters to obtain multiple combined feature parameters, and evaluate the correlation between energy efficiency indicators and each feature parameter and combined feature parameter based on Pearson correlation coefficient, and select strongly correlated feature parameters and combined feature parameters as strongly correlated feature parameters to construct the final feature library.

[0172] The prediction model building module randomly generates an initial population based on the strongly correlated feature parameters in the final feature library. Each individual in the initial population corresponds to a feature parameter combination. The energy efficiency index prediction model is built and trained based on a long short-term memory network. The input of the model is the energy efficiency index and feature parameter combination for multiple consecutive monitoring time periods, and the output is the energy efficiency index for the next monitoring time period. Based on the model training feedback results, the feature parameter combination is optimized using a genetic algorithm to obtain the optimal feature parameter combination for application to the model, thus completing the model training.

[0173] The operation and maintenance work order generation module obtains the predicted values ​​of energy efficiency indicators for future monitoring periods based on the trained energy efficiency indicator prediction model, and generates operation and maintenance management work orders based on the predicted values ​​of energy efficiency indicators.

[0174] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0175] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0176] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0177] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A remote intelligent operation and maintenance management method for data centers, characterized in that, include: S1, acquires the status parameter monitoring data of the data center during the monitoring period, including energy consumption monitoring data, environmental monitoring data, battery monitoring data and hot and cold aisle monitoring data; S2, based on energy consumption monitoring data, calculates the energy efficiency indicators of the data center during the monitoring period, including power usage efficiency, partial power usage efficiency and water usage efficiency; S3, extract features from the state parameter monitoring data to construct an initial feature library including multiple feature parameters. The feature extraction includes descriptive statistics and combined interaction. S4. The feature parameters are randomly concatenated to obtain multiple combined feature parameters. The correlation between energy efficiency indicators and each feature parameter and combined feature parameter is evaluated based on the Pearson correlation coefficient. Strongly correlated feature parameters and combined feature parameters are selected as strongly correlated feature parameters to construct the final feature library. S5. An initial population is randomly generated based on the strongly correlated feature parameters in the final feature library. Each individual in the initial population corresponds to a feature parameter combination. An energy efficiency index prediction model is constructed and trained based on a long short-term memory network. The input of the model is the energy efficiency index and feature parameter combination for multiple consecutive monitoring time periods, and the output is the energy efficiency index for the next monitoring time period. Based on the model training feedback results, a genetic algorithm is used to optimize the feature parameter combination to obtain the optimal feature parameter combination for application to the model, thus completing the model training. S6. Based on the trained energy efficiency index prediction model, obtain the predicted values ​​of energy efficiency index for the future monitoring period, and generate operation and maintenance management work orders based on the predicted values ​​of energy efficiency index. The method for randomly concatenating the feature parameters is as follows: randomly select no less than two feature parameters from the initial feature library for random calculation. The random calculation includes, but is not limited to, using addition, subtraction, multiplication, and division to calculate the selected feature parameters. Each individual in the initial population consists of K gene loci, each corresponding to a strongly correlated feature parameter. If the numerical label of the k-th gene locus in an individual is 1, it means that the feature parameter combination corresponding to that individual includes the strongly correlated feature parameter corresponding to the k-th gene locus; conversely, if the numerical label of the k-th gene locus in an individual is 0, it means that the feature parameter combination corresponding to that individual does not include the strongly correlated feature parameter corresponding to the k-th gene locus. Here, k is the index of the gene locus, and... K represents the number of strongly correlated feature parameters in the final feature library; The method for obtaining the optimal combination of feature parameters is as follows: Step S51: Initialize the mutation probability of each gene locus in the individual, and perform data preprocessing on the energy efficiency indicators and strongly correlated characteristic parameters in each historical monitoring period to form a sample set. Data preprocessing includes removing outliers, filling missing values, and data normalization. Step S52: Construct an energy efficiency index prediction model based on a long short-term memory network, and set the batch size, feedback training period and total training period. The feedback training period is set between one-tenth and one-fifth of the total training period. The energy efficiency index prediction model includes an input layer, one or more LSTM layers, a fully connected layer and an output layer. The input layer is used to receive energy efficiency indicators and feature parameter combinations for multiple consecutive monitoring time periods. The LSTM layer is used to perform feature processing on the data received by the input layer. The fully connected layer is used to map the output of the LSTM layer to the predicted value. The output layer is used to output the energy efficiency indicators for the next monitoring time period. The batch size is the number of samples used in each training iteration, ranging from 32 to 256. The total training period is the total number of training iterations, ranging from 30 to 100. Step S53: Based on the combination of feature parameters corresponding to each individual, a training set, a validation set, and a test set are randomly selected from the sample set. The training sets corresponding to each individual are input into different energy efficiency index prediction models for training until the predetermined feedback training period is reached. Then, the validation sets corresponding to each individual are used to validate each energy efficiency index prediction model to obtain the model training feedback results. The model training feedback results include mean squared error, mean absolute error, and coefficient of determination. Step S54: Based on the model training feedback results, calculate the fitness value of each individual. The higher the fitness value, the better the combination of feature parameters corresponding to that individual. The calculation formula is as follows: In the formula, , , These are the mean squared error, mean absolute error, and coefficient of determination, respectively. , , All are preset weights, and , , The specific value is determined by the analytic hierarchy process (AHP). This represents the fitness value of an individual. Step S55: Sort the individuals according to their fitness values ​​from largest to smallest, select the top 50% of individuals as parents, and perform crossover and mutation operations on each pair of parents to generate offspring. The mutation probability is determined based on the fitness value of the parents and the frequency of occurrence of the digital tags of each gene locus. Calculate the fitness value of the offspring using the same method, and repeat the selection, crossover, and mutation operations on the new population formed by the offspring and parents to generate new parents and offspring until the predetermined number of iterations is reached. Select the individual with the largest fitness value as the optimal individual, and the combination of feature parameters corresponding to the optimal individual is the optimal combination of feature parameters. The criteria for determining strong correlation are as follows: if the absolute value of the Pearson correlation coefficient between a feature parameter and any energy efficiency index is greater than the correlation threshold, then the feature parameter is considered to be strongly correlated with the energy efficiency index, and this feature parameter is a strongly correlated feature parameter. The method for determining whether a combined feature parameter is a strongly correlated feature parameter is the same.

2. The remote intelligent operation and maintenance management method for data centers according to claim 1, characterized in that: The energy consumption monitoring data includes the data center's water consumption, the power consumption of each functional area of ​​the data center, and the power consumption of IT equipment in each functional area. The functional areas of the data center include server rooms, cooling system areas, power distribution rooms, storage areas, network equipment areas, management and maintenance areas, and IT equipment includes servers, storage devices, network devices, and other IT-related equipment. The environmental monitoring data includes time-series data of external ambient temperature, temperature, humidity, air particulate matter concentration, and air velocity within the monitoring period; The battery monitoring data includes the time-series data of the working temperature, working voltage, charging current, and discharging current of each battery cell during the monitoring period, as well as the cumulative number of charge-discharge cycles and the maximum available capacity of each battery cell at the beginning and end of the monitoring period. The monitoring data for the hot and cold aisles includes time-series temperature data at the outlet of the hot and cold aisles, and time-series coolant flow rate and temperature data at the inlet and outlet of the coolant aisles.

3. The remote intelligent operation and maintenance management method for data centers according to claim 2, characterized in that: The formula for calculating the energy efficiency index is as follows: In the formula, To monitor the power consumption of the data center during a specific time period, To monitor the power consumption of IT equipment in the data center during a specific time period, To monitor the power usage efficiency of the data center during the specified time period; In the formula, For the monitoring period, the data center's first Power consumption of each functional area For the monitoring period, the data center's first Power consumption of IT equipment in each functional area For the monitoring period, the data center's first Partial power usage efficiency of each functional area. For indexes of functional areas within the data center, and , The number of functional areas within the data center; In the formula, To monitor the data center's water consumption during the specified time period, To monitor the water usage efficiency of the data center during a specific time period.

4. The remote intelligent operation and maintenance management method for data centers according to claim 2, characterized in that: The steps for constructing the initial feature library are as follows: Step S31: Extract energy consumption monitoring data from the status parameter monitoring data as a primary energy consumption feature group, and process the energy consumption monitoring data to obtain a secondary energy consumption feature group to characterize the energy consumption distribution of each functional area in the data center. Step S32: Perform descriptive statistics on the environmental monitoring data to obtain environmental statistical feature groups, and combine interactive temperature, humidity and particulate matter concentration to obtain environmental interactive feature groups. Step S33: Perform descriptive statistics on the battery monitoring data to obtain a statistical feature set of battery parameters, and combine the interactive current, voltage, charge-discharge cycle count and capacity to obtain an interactive feature set of battery parameters. Step S34: Perform descriptive statistics on the monitoring data of the hot and cold aisles to obtain the statistical feature group of the hot and cold aisles, and combine temperature and coolant flow rate to obtain the interaction feature group of the hot and cold aisles. Step S35: Statistically analyze the feature parameters in the primary energy consumption feature group, secondary energy consumption feature group, environmental statistical feature group, environmental interaction feature group, battery parameter statistical feature group, battery parameter interaction feature group, hot and cold channel statistical feature group, and hot and cold channel interaction feature group to form an initial feature library.

5. The remote intelligent operation and maintenance management method for data centers according to claim 1, characterized in that: The method for determining the mutation probability is as follows: Step S551: Calculate the frequency of occurrence of digital tags for each gene locus in the parent generation. The calculation formula is as follows: In the formula, This represents the number of times the numeric label of the k-th gene locus is 1 in all parents during the t-th iteration of optimization. This represents the total number of parent generations during the t-th iteration of optimization. This represents the frequency of the k-th gene position having a numeric label of 1 during the t-th iteration of optimization. This represents the number of times the numeric label of the k-th gene locus is 0 in all parents during the t-th iteration of optimization. This represents the frequency of the k-th gene locus having a numeric label of 0 during the t-th iteration of optimization, where t is the index of the iteration number. , To optimize the total number of iterations; Step S552: Based on the fitness value of the parent generation, calculate the initial mutation probability of each gene locus in each round of iterative optimization. The calculation formula is as follows: In the formula, Let be the initial mutation probability of the k-th gene locus during the t-th iteration of optimization. Let the mean fitness of all parent generations be the value at the time of the t-th iteration of optimization. This represents the maximum fitness value of all parent generations during the optimization process from the 1st iteration to the tth iteration. This represents the maximum fitness value of all parent generations during the t-th iteration of optimization. It is a piecewise function. The fitness evaluation threshold, This refers to the random perturbation introduced during the t-th iteration of optimization. Step S553: ​​Based on the initial mutation probability and the frequency of occurrence of digital tags at each gene locus in the parent generation, generate the final mutation probability value. The calculation formula is as follows: In the formula, This is the final value of the mutation probability when the numeric tag of the k-th gene locus is changed from 0 to 1 during the t-th iteration of optimization. Let be the final value of the mutation probability of changing the numeric tag of the k-th gene locus from 1 to 0 during the t-th iteration of optimization.

6. A remote intelligent operation and maintenance management system for data centers, used to execute the remote intelligent operation and maintenance management method for data centers according to any one of claims 1-5, characterized in that... ,include: The data monitoring module is used to acquire status parameter monitoring data of the data center during the monitoring period, including energy consumption monitoring data, environmental monitoring data, battery monitoring data, and hot and cold aisle monitoring data. The energy efficiency index calculation module calculates the data center's energy efficiency index during the monitoring period based on energy consumption monitoring data, including power usage efficiency, partial power usage efficiency, and water usage efficiency. The initial feature library construction module is used to extract features from the state parameter monitoring data to construct an initial feature library including multiple feature parameters. The feature extraction includes descriptive statistics and combined interaction. The final feature library construction module is used to randomly concatenate feature parameters to obtain multiple combined feature parameters, and evaluate the correlation between energy efficiency indicators and each feature parameter and combined feature parameter based on Pearson correlation coefficient, and select strongly correlated feature parameters and combined feature parameters as strongly correlated feature parameters to construct the final feature library. The prediction model building module randomly generates an initial population based on the strongly correlated feature parameters in the final feature library. Each individual in the initial population corresponds to a feature parameter combination. The energy efficiency index prediction model is built and trained based on a long short-term memory network. The input of the model is the energy efficiency index and feature parameter combination for multiple consecutive monitoring time periods, and the output is the energy efficiency index for the next monitoring time period. Based on the model training feedback results, the feature parameter combination is optimized using a genetic algorithm to obtain the optimal feature parameter combination for application to the model, thus completing the model training. The operation and maintenance work order generation module obtains the predicted values ​​of energy efficiency indicators for future monitoring periods based on the trained energy efficiency indicator prediction model, and generates operation and maintenance management work orders based on the predicted values ​​of energy efficiency indicators.

Images