A Dynamic Control Method for Data Center Cooling Systems Adapted to Equipment Aging

Abstract

The present invention relates to the technical field of data center energy control, and discloses a dynamic control method for a data center cooling system adapted to equipment aging. First, based on equipment parameters and the principle of energy conservation, an energy consumption model for controlling the data center cooling system is constructed. Aiming at the problem of performance degradation that may occur during the long-term operation of equipment, the method monitors the operating parameters of key equipment such as cooling towers, heat exchangers, chillers, and air-conditioning terminals in real time. When it is detected that the predicted value of the energy consumption model for control deviates from the actual operating value by a certain threshold, the long short-term memory network model in deep learning is used to iteratively update the equipment energy consumption model in a rolling manner, and then the energy consumption model for control is dynamically updated to improve its prediction accuracy, and finally efficient model predictive control is achieved. This method is not only applicable to the design and construction of new data centers, but also applicable to the renovation and upgrade of existing data centers, and has broad application prospects and practical value.

Description

Technical Field

[0001] The present invention relates to the technical field of data center energy control, and particularly to a dynamic control method for a data center cooling system adapted to equipment aging. Background Art

[0002] As the core of modern social information processing, the data center undertakes the tasks of storing, processing, and transmitting massive amounts of data. With the rapid development of technologies such as cloud computing, big data, and the Internet of Things, the scale and complexity of the data center have been continuously increasing, and the demand for energy has also been growing day by day. Traditional data center energy management methods often adopt static or extensive control strategies, which are difficult to adapt to the dynamic changes of equipment and energy demand in the data center, resulting in low energy utilization efficiency and high operating costs.

[0003] During the operation of the data center, the energy consumption of key components such as servers, storage devices, network devices, and cooling systems accounts for the vast majority of the total energy consumption. The energy consumption of these devices is not only related to the performance of the devices themselves but also affected by various factors such as operating load, ambient temperature, and humidity. In addition, the energy demand of the data center has obvious temporal and spatial fluctuations, and the energy demand in different time periods and different regions may vary significantly.

[0004] In order to improve the energy utilization efficiency of the data center, reduce operating costs, and achieve green and sustainable development, there is an urgent need for a method that can perform adaptive dynamic control according to the equipment aging state and the real-time energy demand of the data center. This control strategy should be able to monitor the operating state of the data center in real time, accurately predict the change trends of equipment and energy demand, and dynamically adjust the equipment operating parameters and energy supply strategies to achieve the optimal allocation and utilization of energy.

[0005] Based on the above requirements, the present invention proposes a dynamic control method for a data center cooling system adapted to equipment aging. By constructing a prediction model of equipment aging state and energy demand and combining advanced control algorithms, this method realizes the refined management of various devices in the data center and the dynamic optimization of energy supply. Through the application of this method, the energy utilization efficiency of the data center can be significantly improved, energy consumption can be reduced, operating costs can be decreased, and at the same time, the operating stability and reliability of the data center can be enhanced. Summary of the Invention

[0006] The present invention aims to solve the problems of inaccurate prediction and high energy consumption caused by equipment aging in traditional cooling systems of data centers, and proposes a dynamic control method for data center cooling systems adapted to equipment aging. By combining a control energy consumption model with a long short-term memory network model, this method uses the historical operation data of cooling towers, water pumps, heat exchangers, and chillers to train the model and predict the aging degree of these devices. The prediction results are fed back to the energy consumption system, and the equipment parameters are adjusted through control equations for adaptive control. At the same time, the uncertain factors in the system operation are considered to achieve precise equipment control, thereby improving energy use efficiency.

[0007] In specific implementation, first, a control energy consumption model of the data center cooling system is constructed based on equipment parameters and the principle of energy conservation. This model is used to describe the energy consumption of the system under different working conditions and provides a basis for subsequent control strategies. Aiming at the possible performance degradation problem during long-term operation of equipment, the method monitors the operation parameters of key equipment such as cooling towers, heat exchangers, chillers, and air-conditioning terminals in real time. When it is detected that the predicted value of the control energy consumption model deviates from the actual operation value by a certain threshold, the long short-term memory network model in deep learning is used to iteratively update the equipment energy consumption model in a rolling manner. The long short-term memory network model can capture the long-term dependence relationships in time series data, and improve the prediction accuracy of equipment aging trends through learning and analysis of operation data.

[0008] The present invention is not only applicable to the design and construction of new data centers, but also applicable to the renovation and upgrade of existing data centers. For new data centers, the control method of the present invention can be incorporated in the design stage to optimize the layout and configuration of the cooling system and improve the overall energy efficiency; for existing data centers, the energy-saving renovation can be carried out by upgrading the control system software and hardware facilities and applying the method of the present invention. Whether it is a new construction or renovation project, the present invention can significantly improve the energy utilization rate of the data center, reduce the operation cost, and has broad application prospects and practical value.

[0009] To achieve the technical purpose, the present invention adopts the following technical solutions for implementation.

[0010] The dynamic control method for data center cooling systems adapted to equipment aging proposed by the present invention includes the following steps:

[0011] S1. Construct a control energy consumption model according to the data of the data center

[0012] A control energy consumption model is constructed according to the control equations of the data center. This model includes four parts: cooling load calculation, air-conditioning terminal module calculation, cold source module calculation, and energy consumption module calculation;

[0013] Among them, the cooling load calculation includes the hourly heat transfer through the building envelope of the computer room, the heat dissipation of the servers, the heat dissipation of the PDU and UPS, and the heat dissipation of the lighting and dehumidification equipment, so as to calculate the total cooling load; the air-conditioning terminal module calculates the refrigerating capacity of the air-conditioning terminal using the ε-NTU method; the cold source module calculates the refrigerating capacity of the cooling tower and the cold water unit and the COP of the cold water unit, and judges the operation mode of the equipment; the energy consumption module calculates the number of equipment units to be turned on based on the equipment parameters and operation parameters;

[0014] S2. Obtain the key parameter information of the equipment

[0015] Real-time collect the key operation data of the equipment, the cooling water outlet temperature T of the cooling tower cw_out 、the cooling tower return water temperature T cw_in 、the air volume m of the cooling tower cta and the water flow rate m of the cooling tower ctw ; the cold water supply temperature T ch_in 、the cold water return water temperature T ch_out 、the air supply temperature of the air-conditioning terminal, the air return temperature of the air-conditioning terminal; collect the equipment operation data once every hour, and collect the data for the set duration for the training and testing of the deep learning model;

[0016] S3: Equipment performance evaluation

[0017] Adopt the cooling efficiency as the evaluation index of the performance of the cooling tower. The calculation of the cooling efficiency E is carried out according to the following formula, where represents the wet bulb temperature at that time:

[0018] ,

[0019] The efficiency calculation formulas of the cold water pump and the cooling water pump are as follows:

[0020] ,

[0021] where ρ is the density of water, with the unit of kg / m 3 , g is the acceleration of gravity with the unit of m / s 2 , Q is the water flow rate m 3 / s, H is the pump head, with the unit of m, and P represents the shaft power, with the unit of kW;

[0022] The heat exchanger efficiency calculation formula is as follows:

[0023] ,

[0024] where refers to the water flow rate flowing through the heat exchanger in real time, with the unit of kg / s, refers to the maximum water flow rate of the heat exchanger, with the unit of kg / s, refers to the specific heat capacity of water, with the unit of J / (kg•℃), refers to theoretically the maximum temperature difference passing through the heat exchanger, with the unit of ℃;

[0025] The performance of the chiller is judged by COP, and the COP calculation formula is as follows:

[0026] ,

[0027] wherein refers to the refrigerating capacity of the chiller, with the unit of kW, refers to the electric energy consumed by the chiller, with the unit of kW.

[0028] Preferably, the device key parameter information obtained in step S2 is used as the data set of the long short-term memory network model for model training and testing, and specifically includes the following steps:

[0029] S2.1. Data preprocessing

[0030] The collected data is preprocessed. First, the abnormal data caused by external force majeure factors (such as equipment maintenance, equipment damage, etc.) is removed, and then the data is normalized by the maximum-minimum normalization method. The return water temperature and the outlet water temperature are selected as the important features of the cooling tower for normalization, and the inlet water temperature and the outlet water temperature of the heat exchanger are also normalized. The standardized data is converted into a time series format suitable for the input of the long short-term memory network model; the time step needs to be defined in this step, that is, how many time points of data are included in each input sample; the maximum-minimum normalization formula is as follows:

[0031] ,

[0032] wherein is the original value of the i-th data point, and are the minimum and maximum values of the data set;

[0033] S2.2. Model training

[0034] Since the device needs to be dynamically and adaptively adjusted according to the operation data, the long short-term memory network in deep learning can be used to achieve prediction. The specific long short-term memory network unit formula is shown in Formulas 6-11; the historical operation parameter data set is divided into a training set and a validation set according to a ratio of 7:3; the mean square error is imported from the sklearn.metrics library in Python as the evaluation index for the quality of the model;

[0035] ,

[0036] ,

[0037] ,

[0038] ,

[0039] ,

[0040] ,

[0041] where is the sigmoid activation function, is the input, is the forget gate, is the input gate, is the candidate state, is the cell state, is the output gate;

[0042] According to step S3, it can be known that: the key parameters of the cooling tower operation are obtained: the return water temperature T cw_in of the cooling tower, the cooling water flow rate m ctw and the outlet water temperature T cw_out of the cooling tower,

[0043] The model contains two long short-term memory network layers and a fully connected layer. To prevent overfitting, a Dropout layer is added after the long short-term memory network layer; the Dropout layer will randomly discard some neurons during the training process, thereby enhancing the generalization ability of the model; after the model structure is built, the model needs to be compiled; when compiling, the loss function, optimizer, and evaluation metrics need to be specified. The loss function is selected as the mean squared error, and the formula is as follows:

[0044] ,

[0045] where MSE is the mean squared error, is the true value, is the predicted value, N is the number of samples, and the Adam optimizer is used. The Adam optimizer combines the advantages of the gradient descent method and the momentum method, and its update rule is shown as follows:

[0046] ,

[0047] wherein is the current parameter, is the learning rate, which can be set according to requirements, is the first - order moment estimate of the gradient, is the second - order moment estimate of the gradient, is a very small number set to prevent the denominator from being zero; the evaluation index adopts the common mean square error as the index to evaluate the performance.

[0048] Preferably, the device parameter prediction and model update strategy are as follows:

[0049] Perform real - time prediction on the device according to the constructed long - short - term memory network model; use the cooling water outlet temperature T cw_out of the cooling tower in the past n hours, the cooling water return temperature T cw_in of the cooling tower, the chilled water supply temperature T ch_in and the chilled water return temperature T ch_out to perform data pre - processing. After normalizing and converting the time series of the data, input it into the constructed long - short - term memory network model to predict the efficiency of the device;

[0050] By real - time monitoring of the key operating parameters such as the outlet temperature and return temperature of the cooling tower, the supply temperature and outlet temperature of the chilled water, the pump head, and the power consumption of the chiller, combined with the prediction model, the system can dynamically calculate the instantaneous efficiency of the device; this process not only promotes the continuous optimization of the model and control strategy, but also ensures that the device can operate in the best state; when the difference between the actual measured value and the predicted value exceeds the preset threshold, the system will automatically adjust the relevant device parameters until it returns to the standard operating condition requirements; and adopt an advanced rolling update mechanism to continuously improve the driving device model; the update formula is as follows:

[0051] ,

[0052] ,

[0053] ,

[0054] wherein is the predicted energy consumption of the chiller, is the temperature difference of the liquid passing through the chiller, in ℃, is the mass of the liquid flowing into the chiller, in kg / s, and are the weights trained by the model, bias, is the energy consumption prediction of the fan, where , and are the weight parameters obtained through model training b is the deviation; is the heat transfer efficiency of the heat exchanger, and its and are the convective heat transfer coefficients of the fluids on both sides, δ is the plate thickness in m, λ is the thermal conductivity of the plate material, and usually refers to the convective heat transfer area of the fluids on both sides in m 2 , and are the weight parameters and deviation obtained through model training;

[0055] Based on the real-time operating parameters of the equipment collected, this mechanism retrains the original model structure using deep learning algorithms and adjusts according to the law of conservation of energy; this not only improves the accuracy of the model's prediction of equipment energy consumption, but also provides more accurate data support for comprehensive energy consumption analysis; the entire process forms a closed-loop control system, which not only ensures high energy efficiency, but also greatly improves the overall operation and management level.

[0056] Preferably, the system is adaptively adjusted and controlled according to the results of model prediction, and the steps are as follows:

[0057] When the running time is less than the set time and the equipment model has not triggered an update, continue with the adaptive dynamic control; otherwise, update the aging parameters of the cooling tower, water pump and heat exchanger, and collect the running data of equipment such as the cooling tower again to retrain the model.

[0058] Compared with the related technologies, the dynamic control method of the data center cooling system for adapting to equipment aging proposed by the present invention has the following beneficial effects:

[0059] Based on the data set obtained from the energy consumption model of the data center, the present invention combines the long short-term memory network method in deep learning to predict the performance of the equipment and update the aging factor. By real-time monitoring the running state of the equipment, the present invention can collect data feedback in a timely manner and adjust the equipment parameters according to these data. This real-time adjustment ensures that the equipment always operates under the best working conditions, thereby optimizing the overall energy consumption and energy use efficiency of the data center. By combining the data center energy consumption model with the long short-term memory network technology, the present invention not only improves the accuracy of equipment performance prediction, but also realizes the optimization of equipment aging management, providing an innovative solution for the energy consumption optimization and equipment health management of the data center. Specific embodiments

[0060] To enable those of ordinary skill in the relevant art to more easily master and implement the present invention, we will provide a detailed explanation of the method steps of the present invention. Please note that these examples are only for explaining the present invention and do not limit the scope of protection of the present invention. In addition, it should be understood that after reading and understanding the detailed content of the present invention, those skilled in the art can make various modifications or adjustments based on the present invention, and these equivalent modifications or adjustments also fall within the scope of protection required by this application.

[0061] Embodiment

[0062] The dynamic control method for a data center cooling system adapted to equipment aging proposed by the present invention includes the following steps:

[0063] S1. Construct an energy consumption model for control based on the data in the data center

[0064] Construct an energy consumption model for control according to the control equations of the data center. This model includes four parts: cooling load calculation, air conditioning terminal module calculation, cold source module calculation, and energy consumption module calculation;

[0065] Among them, the cooling load calculation includes the hourly heat transfer through the enclosure structure of the computer room, the heat dissipation of servers, the heat dissipation of PDUs and UPSs, and the heat dissipation of lighting and dehumidification equipment, so as to calculate the total cooling load; the air conditioning terminal module calculation uses the ε-NTU method to calculate the refrigerating capacity of the air conditioning terminal; the cold source module calculation calculates the refrigerating capacity of the cooling tower, the cooling capacity of the chiller and the COP of the chiller, and determines the operating mode of the equipment; the energy consumption module calculation determines the number of equipment units to be turned on based on the equipment parameters and operating parameters;

[0066] S2. Obtain key parameter information of the equipment

[0067] Real-time collect the key operating data of the equipment, the cooling water outlet temperature T of the cooling tower cw_out , the cooling tower return water temperature T cw_in , the air volume m of the cooling tower cta and the water flow rate m of the cooling tower ctw ; the chilled water supply temperature T ch_in , the chilled water return temperature T ch_out , the air supply temperature of the air conditioning terminal, and the air return temperature of the air conditioning terminal; Collect the equipment operation data once an hour, and collect the data for a set duration for the training and testing of the deep learning model;

[0068] S3: Equipment performance evaluation

[0069] Use the cooling efficiency as the evaluation index for the performance of the cooling tower. The cooling efficiency E is calculated using the following formula, where represents the wet bulb temperature at that time:

[0070] ,

[0071] The efficiency calculation formulas for the cold water pump and the cooling water pump are as follows:

[0072] ,

[0073] where ρ is the density of water, with the unit of kg / m 3 , g is the acceleration due to gravity with the unit of m / s 2 , Q is the water flow rate in m 3 / s, H is the head of the pump, with the unit of m, and P represents the shaft power, with the unit of kW;

[0074] The efficiency calculation formula for the heat exchanger is as follows:

[0075] ,

[0076] where refers to the water flow rate flowing through the heat exchanger in real time, with the unit of kg / s, refers to the maximum water flow rate of the heat exchanger, with the unit of kg / s, refers to the specific heat capacity of water, with the unit of J / (kg•°C), refers to theoretically the maximum temperature difference passing through the heat exchanger, with the unit of °C;

[0077] The performance of the chiller is judged using COP, and the COP calculation formula is as follows:

[0078] ,

[0079] where refers to the refrigerating capacity of the chiller, with the unit of kW, refers to the electric energy consumed by the chiller, with the unit of kW.

[0080] Furthermore, the device key parameter information obtained in step S2 is used as the data set of the long short-term memory network model for model training and testing, and specifically includes the following steps:

[0081] S2.1. Data preprocessing

[0082] Preprocess the collected data. First, remove the abnormal data caused by external force majeure factors, including equipment maintenance and equipment damage. Then, use the maximum-minimum normalization method to normalize the data. Select the return water temperature and the outlet water temperature as the important features of the cooling tower for normalization. The inlet water temperature and the outlet water temperature of the heat exchanger are also normalized. Convert the standardized data into a time series format suitable for input into the long short-term memory network model. This step requires defining the time step, that is, how many time points of data are included in each input sample. The maximum-minimum normalization formula is as follows:

[0083] ,

[0084] where is the original value of the i-th data point, and are the minimum and maximum values of the data set;

[0085] S2.2. Training of the model

[0086] Since the equipment needs to be dynamically and adaptively adjusted according to the operation data, the long short-term memory network in deep learning can be used to achieve prediction. The specific long short-term memory network unit formula is shown in Formulas 6-11. Divide the historical operation parameter data set into a training set and a validation set according to a ratio of 7:3. Import the mean squared error from the sklearn.metrics library in Python as the evaluation index for the quality of the model.

[0087] ,

[0088] ,

[0089] ,

[0090] ,

[0091] ,

[0092] ,

[0093] where is the sigmoid activation function, is the input, is the forget gate, is the input gate, is the candidate state, is the unit status, is the output gate;

[0094] According to step S3, the key parameters of the cooling tower operation are obtained: the return water temperature T of the cooling tower cw_in , the cooling water flow rate m ctw and the outlet water temperature T of the cooling tower cw_out . The model contains two layers of long short-term memory network layers and a fully connected layer. To prevent overfitting, a Dropout layer is added after the long short-term memory network layer; the Dropout layer will randomly discard a part of neurons during the training process, thereby enhancing the generalization ability of the model; after constructing the model structure, the model needs to be compiled; when compiling, the loss function, optimizer and evaluation metrics need to be specified. The loss function selects the mean squared error (MSE), and the formula is as follows:

[0095] ,

[0096] where MSE is the mean squared error, is the true value, is the predicted value, N is the number of samples, and the Adam optimizer is used. The Adam optimizer combines the advantages of the gradient descent method and the momentum method, and its update rule is as follows:

[0097] ,

[0098] where is the current parameter, is the learning rate, which can be set according to requirements, is the first moment estimate of the gradient, is the second moment estimate of the gradient, is a very small number set to prevent the denominator from being zero; the evaluation metric adopts the common mean squared error as the index to evaluate the performance.

[0099] Furthermore, the equipment parameter prediction and model update strategy are as follows:

[0100] Perform real-time prediction on the equipment according to the constructed long short-term memory network model; use the cooling water outlet temperature T of the cooling tower in the past n hours cw_out , the cooling water return temperature T of the cooling tower cw_in , the supply water temperature T of the chilled water ch_in , the return water temperature T of the chilled water ch_out for data preprocessing. After normalizing and performing time series conversion on the data, input it into the constructed long short-term memory network model to predict the efficiency of the equipment;

[0101] By real-time monitoring of key operating parameters such as the outlet water temperature and return water temperature of the cooling tower, the chilled water supply temperature and outlet water temperature, the pump head, and the power consumption of the chiller, and combining with the prediction model, the system can dynamically calculate the instant efficiency of the equipment; this process not only promotes the continuous optimization of the model and control strategy, but also ensures that the equipment can operate in the best state; when the difference between the actual measured value and the predicted value exceeds the preset threshold, the system will automatically adjust the relevant equipment parameters until it returns to the standard operating condition requirements; and an advanced rolling update mechanism is adopted to continuously improve the drive equipment model; the update formula is as follows:

[0102] ,

[0103] ,

[0104] ,

[0105] where is the predicted energy consumption of the chiller, is the temperature difference of the liquid passing through the chiller, in °C, is the mass of the liquid flowing into the chiller, in kg / s, and are the weights trained by the model, bias, is the energy consumption prediction of the fan, where , and are the weight parameters obtained through model training b is the deviation; is the heat transfer efficiency of the heat exchanger, where and are the convective heat transfer coefficients of the fluids on both sides, δ is the thickness of the plate, in m, λ is the thermal conductivity of the plate material, and usually refer to the convective heat transfer area of the fluids on both sides, in m 2 , and are the weight parameters and deviation obtained through model training;

[0106] This mechanism is based on the real-time operating parameters of the equipment collected, retrains the original model structure using deep learning algorithms, and adjusts according to the law of conservation of energy; this can not only improve the accuracy of the model's prediction of equipment energy consumption, but also provide more accurate data support for comprehensive energy consumption analysis; the entire process forms a closed-loop control system, which not only ensures high energy efficiency, but also greatly improves the overall operation and management level.

[0107] Furthermore, the system is adaptively adjusted and controlled according to the results predicted by the model, and the steps are as follows:

[0108] When the running time is less than the set time and the device model has not triggered an update, continue with the adaptive dynamic control; otherwise, update the aging parameters of the cooling tower, water pump, and heat exchanger, and re-collect the running data of devices such as the cooling tower to retrain the model.

[0109] Compared with the related technologies, the dynamic control method for the data center cooling system that adapts to equipment aging proposed by the present invention has the following beneficial effects:

[0110] Based on the data set obtained from the energy consumption model of the data center, the present invention combines the long short-term memory network method in deep learning to predict the performance of the device and update the aging factor. By real-time monitoring the running state of the device, the present invention can timely collect data feedback and adjust the device parameters according to these data. This real-time adjustment ensures that the device always operates under the best working conditions, thereby optimizing the overall energy consumption and energy use efficiency of the data center. By combining the data center energy consumption model with the long short-term memory network technology, the present invention not only improves the accuracy of device performance prediction but also realizes the optimization of device aging management, providing an innovative solution for the energy consumption optimization and device health management of the data center.

[0111] The present invention is not only applicable to the design and construction of new data centers but also to the renovation and upgrade of existing data centers. For new data centers, the control method of the present invention can be incorporated in the design stage to optimize the layout and configuration of the cooling system and improve the overall energy efficiency; for existing data centers, the energy-saving renovation can be carried out by upgrading the control system software and hardware facilities and applying the method of the present invention. Whether it is a new construction or renovation project, the present invention can significantly improve the energy utilization rate of the data center, reduce the operating cost, and has broad application prospects and practical value.

[0112] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A dynamic control method for a data center cooling system adapted to equipment aging, characterized in that, It includes the following steps: S1. Construct an energy consumption model for control based on the data in the data center. Construct an energy consumption model for control according to the control equations of the data center. This model includes four parts: cooling load calculation, air-conditioning terminal module calculation, cold source module calculation, and energy consumption module calculation. Among them, the cooling load calculation includes the hourly heat transfer of the computer room enclosure structure, the heat dissipation of servers, the heat dissipation of PDUs and UPSs, and the heat dissipation of lighting and dehumidification equipment, so as to calculate the total cooling load; the air-conditioning terminal module calculates the refrigerating capacity of the air-conditioning terminal using the ε-NTU method; the cold source module calculates the refrigerating capacity of the cooling tower and the cold capacity of the chiller and the COP of the chiller, and judges the operating mode of the equipment; the energy consumption module calculates the number of equipment units to be turned on based on the equipment parameters and operating parameters; S2. Obtain the key parameter information of the equipment. Collect key operating data of the real-time acquisition device, the cooling water outlet temperature T of the cooling tower cw_out , the cooling tower return water temperature T cw_in , the air volume m of the cooling tower cta and the water flow rate m of the cooling tower ctw ; the chilled water supply temperature T ch_in , the chilled water return temperature T ch_out , the air supply temperature at the air-conditioning terminal, the air return temperature at the air-conditioning terminal; collect the device operation data once every hour, and collect the data for the set duration for the training and testing of the deep learning model; S3: Equipment performance evaluation. The cooling efficiency is adopted as the evaluation index for the performance of the cooling tower. The cooling efficiency E is calculated using the following formula, where represents the wet bulb temperature at that time: ， The efficiency calculation formulas for the chilled water pump and the cooling water pump are as follows: ， where ρ is the density of water in kg / m 3 , g is the acceleration due to gravity in m / s 2 , Q is the water flow rate in m 3 / s, H is the pump head in m, P represents the shaft power in kW; The efficiency calculation formula for the heat exchanger is as follows: ， wherein refers to the water flow rate flowing through the heat exchanger in real time, with the unit of kg / s, refers to the maximum water flow rate of the heat exchanger, with the unit of kg / s, refers to the specific heat capacity of water, with the unit of J / (kg•°C), refers to theoretically the maximum temperature difference passing through the heat exchanger, with the unit of °C; Use COP to judge the performance of the chiller. The COP calculation formula is as follows: ， wherein refers to the refrigerating capacity of the chiller, with the unit of kW, refers to the electric energy consumed by the chiller, with the unit of kW; S4. The obtained key parameter information of the equipment is used as the data set of the long short-term memory network model for model training and testing. The specific steps are as follows: S4.

1. Data preprocessing. Preprocess the collected data. First, remove the abnormal data caused by external force majeure factors including equipment maintenance and equipment damage. Then, use the maximum-minimum normalization method to normalize the data. Select the return water temperature and the outlet water temperature as the important features of the cooling tower for normalization. The inlet water temperature and the outlet water temperature of the heat exchanger are also normalized. Convert the standardized data into a time series format suitable for input into the long short-term memory network model. This step requires defining the time step, that is, how many time points of data are included in each input sample. The maximum-minimum normalization formula is as follows: ， where is the original value of the i-th data point, and are the minimum and maximum values of the data set; S4.

2. Model training. Since the equipment needs to be dynamically and adaptively adjusted according to the operation data, the long short-term memory network in deep learning can be used to achieve prediction. The specific long short-term memory unit formula is shown in Formulas 6-11. Divide the historical operation parameter data set into a training set and a validation set according to a ratio of 7:

3. Import the mean squared error from the sklearn.metrics library in Python as the evaluation index for the quality of the model. ， ， ， ， ， ， Among them is the sigmoid activation function, is the input, is the forget gate, is the input gate, is the candidate state, is the cell state, is the output gate; According to step S3, the key parameters of the cooling tower operation are obtained: the return water temperature T of the cooling tower cw_in , the cooling water flow rate m ctw and the outlet water temperature T of the cooling tower cw_out , The model includes two long short-term memory network layers and a fully connected layer. To prevent overfitting, add a Dropout layer after the long short-term memory network layer. The Dropout layer will randomly discard some neurons during the training process, thereby enhancing the generalization ability of the model. After constructing the model structure, the model needs to be compiled. When compiling, it is necessary to specify the loss function, optimizer, and evaluation index. The loss function selects the mean squared error. The formula is as follows: ， where MSE is the mean squared error, is the true value, is the predicted value, N is the number of samples, and the optimizer used is the Adam optimizer. The Adam optimizer combines the advantages of the gradient descent method and the momentum method, and its update rule is shown as follows: ， Among them is the current parameter is the learning rate, which can be set according to requirements is the first-order moment estimate of the gradient is the second-order moment estimate of the gradient is a very small number set to prevent the denominator from being zero; the evaluation index adopts the commonly used mean square error as the index to evaluate the performance S5. Equipment parameter prediction and model update strategy. The specific steps are as follows: Perform real-time prediction on the device according to the constructed long short-term memory network model; use the cooling water outlet temperature T of the cooling tower in the past n hours cw_out , the cooling water return temperature T of the cooling tower cw_in , the supply temperature T of the chilled water ch_in , the return temperature T of the chilled water ch_out Perform data preprocessing, input the data into the constructed long short-term memory network model after standardization and time series conversion, and predict the efficiency of the device; By real-time monitoring of the key operation parameters such as the outlet water temperature and return water temperature of the cooling tower, the chilled water supply temperature and outlet water temperature, the pump head, and the power consumption of the chiller, combined with the prediction model, the system can dynamically calculate the instantaneous efficiency of the equipment. This process not only promotes the continuous optimization of the model and the control strategy but also ensures that the equipment can operate in the best state. When the difference between the actual measurement value and the predicted value exceeds the preset threshold, the system will automatically adjust the relevant equipment parameters until the standard working condition requirements are restored. And adopt an advanced rolling update mechanism to continuously improve the driving equipment model. The update formula is as follows: ， ， ， Among them is the predicted energy consumption of the chiller, is the temperature difference of the liquid passing through the chiller, in °C, is the mass of the liquid flowing into the chiller, in kg / s, and are the weights trained by the model, bias, is the energy consumption prediction of the fan, among which , and are the weight parameters obtained through model training b is the deviation; is the heat transfer efficiency of the heat exchanger, where and are the convective heat transfer coefficients of the fluids on both sides, δ is the thickness of the plate, in m, λ is the thermal conductivity of the plate material, and usually refers to the convective heat transfer area of the fluids on both sides, in m 2 , and are the weight parameters and deviation obtained through model training; This mechanism is based on the real-time operation parameters of the equipment collected, retrains the original model structure using deep learning algorithms, and adjusts according to the law of conservation of energy.

2. The dynamic control method for a data center cooling system adapting to equipment aging according to claim 1, characterized in that, Adaptive adjustment control of the system is carried out according to the results predicted by the model, and the steps are as follows: When the running time is less than the set time and the device model has not triggered an update, continue with adaptive dynamic control; otherwise, update the aging parameters of the cooling tower, water pump, and heat exchanger, and collect the running data of the cooling tower equipment again to retrain the model.

Images