A real-time energy efficiency optimization control method, device and apparatus for an air conditioner used in a data center

Abstract

This invention discloses a real-time energy efficiency optimization control method, equipment, and device for air conditioners used in data centers, relating to the field of HVAC control technology; it includes: Step 1: establishing a neural network-based prediction model for the cooling capacity of a rack-mounted variable frequency air conditioner; Step 2: establishing a prediction model for the energy consumption of a rack-mounted variable frequency air conditioner; Step 3: constructing a Bayesian optimization-based objective function for real-time energy efficiency optimization of the rack-mounted variable frequency air conditioner; This invention, while meeting the load requirements of IT equipment within the rack, ensures that the rack-mounted air conditioning system always operates at its optimal energy efficiency point under different outdoor temperatures and IT equipment loads, effectively improving the energy efficiency and reliability of the rack-mounted variable frequency air conditioner.

Description

Technical Field

[0001] This invention discloses a method, equipment, and device for real-time energy efficiency optimization control of air conditioning for data centers, which relates to the field of HVAC control technology. Background Technology

[0002] In recent years, with the rapid development of the information and communication industry, the number of data center racks and total energy consumption have grown rapidly. Cooling systems account for 30-50% of total energy consumption. Therefore, improving the energy efficiency of cooling systems is crucial for the sustainable development of data centers. Existing rack air conditioners generally use fixed-frequency vapor compression systems, with their compressors and fans operating at maximum power year-round. When the heat generated by the IT equipment inside the rack does not match the cooling capacity of the air conditioner, the only adjustment is usually to start and stop the compressor and fan, leading to significant energy waste. Compared to fixed-frequency air conditioners, variable-frequency air conditioners offer better energy efficiency and adjustability. However, under dynamic operating conditions, due to the complex nonlinear coupling relationships between various operating parameters such as compressor speed, condenser fan speed, evaporator fan speed, and the opening of the electronic expansion valve, most rack variable-frequency air conditioners currently employ control strategies based on fixed temperature setpoints or simple rules, such as maintaining a constant return air temperature. These methods fail to consider changes in the outdoor environment and the coupling relationships between components, and cannot guarantee that the system operates at its optimal energy efficiency point under all conditions. Furthermore, they fail to consider the combined effects of multiple operating parameters and the coordinated control and optimization of these parameters. Summary of the Invention

[0003] This invention addresses the problems of existing technologies by providing a real-time energy efficiency optimization control method, equipment, and device for air conditioning in data centers. Under the premise of meeting the load requirements of IT equipment in the rack, the rack air conditioning system always operates at the optimal energy efficiency point under different outdoor temperatures and IT equipment loads, effectively improving the energy efficiency and reliability of rack inverter air conditioners.

[0004] The specific solution proposed in this invention is as follows:

[0005] This invention provides a real-time energy efficiency optimization control method for inverter air conditioners used in data center racks, comprising:

[0006] Step 1: Establish a neural network-based model for predicting the cooling capacity of rack-mounted inverter air conditioners;

[0007] Step 2: Establish a cabinet inverter air conditioner energy consumption prediction model;

[0008] Step 3: Construct a Bayesian optimization-based objective function for real-time energy efficiency optimization of rack-mounted inverter air conditioners:

[0009] Step 301: Preload the pre-trained cabinet inverter air conditioner cooling capacity prediction model and energy consumption prediction model, and define the control variables and the safe operating boundary of the cabinet inverter air conditioner.

[0010] Controlled variable: Compressor speed N com Condenser fan speed N c Evaporator fan speed N e Electronic expansion valve opening Φ,

[0011] Step 302: Periodically collect the status parameters of the rack-mounted inverter air conditioner in real time. The status parameters include the outdoor ambient temperature. T amb Air conditioning return air set temperature T set Condensing pressure P c,r Evaporation pressure P e,r The target cooling capacity required at present

[0012] Step 303: Construct the Bayesian optimization objective function: In each optimization cycle, based on the collected state parameters, construct the energy efficiency ratio objective function under the current operating conditions:

[0013] EER = f ( N com , N c , N e ,Φ)= Q pred / W total,pred ,

[0014] Combine the current control variables and state parameters ( T amb , T set , N com , N c , N e ,

[0015] The predicted cooling capacity is obtained by inputting the data into the cabinet inverter air conditioner cooling capacity prediction model. Q pred ,

[0016] Combine the current control variables and state parameters ( N com ,N c , N e , P c,r , P e,r The data is input into the cabinet inverter air conditioner energy consumption prediction model to obtain the predicted total energy consumption of the air conditioning system. W total,pred ,

[0017] Step 304: Use the Bayesian optimization algorithm to maximize the Bayesian objective function and obtain the prediction. EER The highest combination of control variables is taken as the optimal solution. N com , N c , N e ,Φ),

[0018] Step 305: The optimal solution is sent to the inverter driver and electronic expansion valve controller of the air conditioning system, causing the air conditioning system to switch to the new parameters.

[0019] Step 306: Enable the air conditioning system to operate in a steady state and enter the periodic data collection and optimization process.

[0020] Furthermore, step 1 of the aforementioned method for real-time energy efficiency optimization control of a data center rack inverter air conditioner specifically includes:

[0021] Step 101: Determine the model's input and output variables: Define the input and output layer nodes of the neural network. The input variable is the outdoor ambient temperature. T amb Air conditioning return air temperature T return Compressor speed N com Condenser fan speed N c Evaporator fan speed N e And the opening degree Φ of the electronic expansion valve; the output variable is the prediction target of the model, which is the cooling capacity of the air conditioning system. Q ;

[0022] Step 102: Collect and preprocess the data of the input variables;

[0023] Step 103: Construct the neural network structure of the model: Construct a multi-layer feedforward neural network, the structure of which includes:

[0024] An input layer must contain at least 6 neurons.

[0025] At least one hidden layer;

[0026] One output layer: contains one neuron, and outputs the predicted cooling capacity value;

[0027] Activation functions: ReLU activation function is used for neurons in the hidden layer, and linear activation function is used for neurons in the output layer;

[0028] Step 104: Based on the preprocessed data, divide the dataset and train the model;

[0029] Step 105: Perform model validation and performance evaluation.

[0030] Furthermore, step 102 of the aforementioned method for real-time energy efficiency optimization control of a data center rack inverter air conditioner includes:

[0031] Step 1021: When collecting data on input variables, the cooling capacity... Q It can be obtained by measuring the enthalpy difference; or by calculating it after measuring with an anemometer and a temperature sensor. When calculating, the formula is used:

[0032] Q = ρvAC e (T e,a,i -T e,a,o )，

[0033] in, ρ air density (kg / m³) 3 ); v The evaporator outlet air velocity (m / s); A Evaporator outlet area (m²) 2 ); C e The specific heat capacity of the air at the evaporator outlet is J / (kg·K); T e,a,i The evaporator inlet air temperature (°C) is the same as the air conditioner return air temperature. T return ; T e,a,o The evaporator outlet temperature (°C);

[0034] v To obtain the average value of multiple wind speed measuring points evenly distributed at the evaporator air outlet; T e,a,i To evenly distribute multiple temperature measuring points at the evaporator air inlet, the average value of the measurements is calculated. T e,a,o To evenly distribute multiple temperature measuring points at the evaporator outlet, the average value of the measurements is calculated.

[0035] Step 1022: Preprocess the collected data, including:

[0036] Data cleaning: Remove abnormal data points caused by sensor malfunctions or unstable operation of the air conditioning system.

[0037] Data normalization: The maximum-minimum normalization method is used to normalize the compressor speed. N com Condenser fan speed N c Evaporator fan speed N e The opening Φ of the electronic expansion valve, scaled to the [0,1] interval, is calculated using the following formula:

[0038] X norm =( X-X min ) / (X max -X min )，

[0039] in, X The original value, X min and X max These are the minimum and maximum values ​​of a certain variable.

[0040] Furthermore, step 2 of the aforementioned method for real-time energy efficiency optimization control of a data center rack inverter air conditioner specifically includes:

[0041] Step 201: Determine the model input and output variables: Input variables include compressor speed. N com Condenser fan speed N c Evaporator fan speed N e Condenser inlet air temperature T c,a,i and evaporator inlet air temperature T e,a,i The output variable is the total power consumption of the air conditioning system. W total ，

[0042] Intermediate physical quantity: Condensation pressure P c,r Evaporation pressure P e,r ;

[0043] Step 202: Collect and preprocess the data of the input variables;

[0044] Step 203: Establish energy consumption models for each component of the rack-mounted inverter air conditioner:

[0045] Establish a compressor power consumption sub-model to predict compressor power consumption. W com ;

[0046] Establish a fan power consumption sub-model to predict the power consumption of the condenser fan. W c Power consumption of evaporator fan W e ,

[0047] Step 204: Perform model integration to obtain the total energy consumption model: W total =W com +W c +W e ，

[0048] Step 205: Perform model validation and performance evaluation.

[0049] Furthermore, step 203 of the aforementioned method for real-time energy efficiency optimization control of a data center rack inverter air conditioner specifically includes:

[0050] Step 2031: Establish a compressor power consumption sub-model, represented by compressor speed. N com and stress correction factor F p Functions:

[0051] W com = f(N com ，F p )，

[0052] Stress correction factor F p From condensation pressure P c,r and evaporation pressure P e,r The ratios reflect the impact of the air conditioning system's pressure ratio on the compressor's power consumption:

[0053] W com = C(P c,r / P e,r ）(k-1) / k ，

[0054] C This is a constant used for model calibration; k The adiabatic index of the refrigerant is obtained by consulting the refrigerant's physical property parameters.

[0055] Step 2032: Establish a sub-model of wind turbine power consumption, simplifying the model into a cubic function of wind turbine speed:

[0056] W c = a 1 N c 3 +b 1 N c 2 +c 1 N c +d 1 ；W e = a 2 N e 3 +b 2 N e 2 +c 2 N e +d 2 ；

[0057] a 1 a 2 、b 1 , b 2 , c 1 、c 2 、d 1 、d 2 The fitting coefficients are denoted as .

[0058] Furthermore, in step 301 of the aforementioned method for real-time energy efficiency optimization control of a data center rack inverter air conditioner, the control variables and the safe operating boundaries of the rack inverter air conditioner are defined, including:

[0059] Define the search space: Define the physically permissible and safe minimum and maximum values ​​for each control variable.

[0060] Define constraints: Define the safety constraints for the operation of the air conditioning system, including: upper and lower limits of evaporation pressure, upper and lower limits of condensation pressure, upper and lower limits of superheat, and upper limit of compressor discharge temperature.

[0061] Furthermore, in step 302 of the aforementioned method for real-time energy efficiency optimization control of inverter air conditioners for data center racks, the optimization of the objective function based on Bayesian optimization is performed under the premise of satisfying the cooling capacity matching constraint, using the formula:

[0062] (| Q pred -Q red | / Q red )≤δ,

[0063] Q red Let δ represent the target cooling capacity and δ be the allowable relative error.

[0064] Furthermore, step 304 of the aforementioned method for real-time energy efficiency optimization control of a data center rack inverter air conditioner specifically includes:

[0065] Step 3041: Construct a surrogate model: Use Gaussian process regression to fit the mapping relationship between the control variables and the objective function EER.

[0066] Step 3042: Using the expected improvement function, when recommending the next sampling point, the expected improvement of the objective function and the probability of satisfying all constraints are considered simultaneously based on the actual situation.

[0067] Step 3043: Solve for the optimal observation point: Within the defined search space, find the next combination of candidate control variables that maximizes the expected improvement function value.

[0068] Step 3044: Evaluation and Model Update: Combine candidate control variables and input them into the objective function to obtain the corresponding... EER The system determines the satisfaction of values ​​and constraints, and updates the surrogate model to make it more closely resemble the actual system response.

[0069] Step 3045: Repeat steps 3042 to 3044 for a preset number of iterations. After the iterations are complete, select the prediction from the surrogate model. EER The highest combination of control variables is taken as the optimal solution for this optimization. N com , N c , N ,Φ).

[0070] The present invention also provides a real-time energy efficiency optimization control device for a rack-mounted variable frequency air conditioner for a data center, comprising: at least one memory and at least one processor;

[0071] The at least one memory is used to store a machine-readable program;

[0072] The at least one processor is used to call the machine-readable program to execute the real-time energy efficiency optimization control method for a data center rack inverter air conditioner.

[0073] This invention also provides a real-time energy efficiency optimization and control device for a data center rack-mounted inverter air conditioner, comprising: a cooling capacity prediction model management module, an energy consumption prediction model management module, and an optimization module.

[0074] The cooling capacity prediction model management module establishes a neural network-based prediction model for the cooling capacity of cabinet inverter air conditioners.

[0075] The energy consumption prediction model management module establishes an energy consumption prediction model for the cabinet inverter air conditioner;

[0076] The optimization module constructs a Bayesian optimization objective function to perform real-time energy efficiency optimization for rack-mounted inverter air conditioners.

[0077] Step 301: Preload the pre-trained cabinet inverter air conditioner cooling capacity prediction model and energy consumption prediction model, and define the control variables and the safe operating boundary of the cabinet inverter air conditioner.

[0078] Controlled variable: Compressor speed N com Condenser fan speed N c Evaporator fan speed N e Electronic expansion valve opening Φ,

[0079] Step 302: Periodically collect the status parameters of the rack-mounted inverter air conditioner in real time. The status parameters include the outdoor ambient temperature. T amb Air conditioning return air set temperature T set Condensing pressure P c,r Evaporation pressure P e,r The target cooling capacity required at present

[0080] Step 303: Construct the Bayesian optimization objective function: In each optimization cycle, based on the collected state parameters, construct the energy efficiency ratio objective function under the current operating conditions:

[0081] EER =f ( N com , N c , N e ,Φ)= Q pred / W total,pred ,

[0082] Combine the current control variables and state parameters ( T amb , T set , N com , N c , N e The input (Φ) is given to the cabinet inverter air conditioner cooling capacity prediction model to obtain the predicted cooling capacity. Q pred ,

[0083] Combine the current control variables and state parameters ( N com , N c , N e , P c,r , P e,r The data is input into the cabinet inverter air conditioner energy consumption prediction model to obtain the predicted total energy consumption of the air conditioning system. W total,pred ,

[0084] Step 304: Use the Bayesian optimization algorithm to maximize the Bayesian objective function and obtain the prediction. EER The highest combination of control variables is taken as the optimal solution. N com , N c , N e ,Φ),

[0085] Step 305: The optimal solution is sent to the inverter driver and electronic expansion valve controller of the air conditioning system, causing the air conditioning system to switch to the new parameters.

[0086] Step 306: Enable the air conditioning system to operate in a steady state and enter the periodic data collection and optimization process.

[0087] The advantages of this invention are:

[0088] 1. Global optimization and adaptation: Bayesian optimization is good at handling black box and nonlinear problems. It can escape local optima, find the global optimum or near the global optimum, and automatically adjust according to changes in operating conditions.

[0089] 2. Safe and reliable: The optimization process takes place in the model's virtual environment, eliminating the need for frequent and dangerous perturbation tests on the real system. By setting safety constraints, it is ensured that the recommended parameters are always within a safe range.

[0090] 3. Strong real-time performance: The optimization calculation is based on a pre-trained high-performance model, which is fast and can complete the optimization in seconds, meeting the requirements of real-time control.

[0091] 4. Significant energy saving effect: By continuously tracking the optimal EER point, the system energy consumption can be reduced to the greatest extent, especially in scenarios where the actual load and outdoor temperature fluctuate frequently, the energy saving effect is particularly outstanding.

[0092] 5. High engineering practicality: This method does not require major modifications to existing hardware systems; the core is the upgrade of software algorithms, making it easy to implement and deploy. Attached Figure Description

[0093] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0094] Figure 1 This is a flowchart for real-time energy efficiency optimization of rack-mounted inverter air conditioners.

[0095] Figure 2 This is a flowchart for establishing a prediction model for the cooling capacity of a cabinet inverter air conditioner.

[0096] Figure 3 This is a schematic diagram of the neural network structure of the cooling capacity prediction model.

[0097] Figure 4 This is a flowchart for establishing a cabinet inverter air conditioner energy consumption prediction model. Detailed Implementation

[0098] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0099] Example 1: This invention provides a real-time energy efficiency optimization control method for inverter air conditioners used in data center racks, comprising:

[0100] Step 1: Establish a neural network-based model for predicting the cooling capacity of rack-mounted inverter air conditioners;

[0101] Step 2: Establish a cabinet inverter air conditioner energy consumption prediction model;

[0102] Step 3: Construct a Bayesian optimization-based objective function for real-time energy efficiency optimization of rack-mounted inverter air conditioners:

[0103] Step 301: Initialize and preload the pre-trained cabinet inverter air conditioner cooling capacity prediction model and energy consumption prediction model, and define the control variables and the safe operating boundary of the cabinet inverter air conditioner:

[0104] Controlled variable: Compressor speed N com Condenser fan speed N c Evaporator fan speed N e Electronic expansion valve opening Φ,

[0105] Define the search space: Define the physically permissible and safe minimum and maximum values ​​for each control variable.

[0106] Define constraints: Define the safety constraints for the operation of the air conditioning system, including: upper and lower limits of evaporation pressure, upper and lower limits of condensation pressure, upper and lower limits of superheat, and upper limit of compressor discharge temperature.

[0107] After normalization, control boundaries are set first:

[0108] N com ∈[0.3,0.9], N c ∈[0.3,0.9], N e ∈[0.3,0.9], Φ∈[0.2,0.8].

[0109] Step 302: Periodically collect the status parameters of the rack-mounted inverter air conditioner in real time. The status parameters include the outdoor ambient temperature. T amb Air conditioning return air set temperature T set Condensing pressure P c,r Evaporation pressure P e,r The current target cooling capacity. Air conditioning return air setpoint temperature. T set This is the evaporator inlet air temperature. T e,a,i Air conditioning return air set temperature Tset The temperature setting can be changed according to actual conditions. Real-time power of IT equipment. P IT This power value is equal to the target cooling capacity currently required by the system. Q req .

[0110] Step 303: Construct the Bayesian optimization objective function: In each optimization cycle, based on the collected state parameters, construct the energy efficiency ratio objective function under the current operating conditions:

[0111] EER = f ( N com , N c , N e ,Φ)= Q pred / W total,pred ,

[0112] Combine the current control variables and state parameters ( T amb , T set , N com , N c , N e The input (Φ) is given to the cabinet inverter air conditioner cooling capacity prediction model to obtain the predicted cooling capacity. Q pred ,

[0113] Combine the current control variables and state parameters ( N com , N c , N e , P c,r , P e,r The data is input into the cabinet inverter air conditioner energy consumption prediction model to obtain the predicted total energy consumption of the air conditioning system. W total,pred .

[0114] To ensure that the optimization of the objective function based on Bayesian optimization is performed while satisfying the cooling capacity matching constraint, the following formula is used:

[0115] (| Q pred -Q red | / Qred )≤δ,

[0116] Q red Let δ represent the target cooling capacity and δ be the allowable relative error.

[0117] Step 304: Use the Bayesian optimization algorithm to maximize the Bayesian objective function and obtain the prediction. EER The highest combination of control variables is taken as the optimal solution. N com , N c , N e ,Φ).

[0118] Specifically, it can include:

[0119] Step 3041: Construct a surrogate model: Use Gaussian process regression to fit the mapping relationship between the control variables and the objective function EER.

[0120] Step 3042: Using the expected improvement function, when recommending the next sampling point, the expected improvement of the objective function and the probability of satisfying all constraints are considered simultaneously based on the actual situation.

[0121] Step 3043: Solve for the optimal observation point: Within the defined search space, find the next combination of candidate control variables that maximizes the expected improvement function value.

[0122] Step 3044: Evaluation and Model Update: Combine candidate control variables and input them into the objective function to obtain the corresponding... EER The system determines the satisfaction of values ​​and constraints, and updates the surrogate model to make it more closely resemble the actual system response.

[0123] Step 3045: Repeat steps 3042 to 3044 for a preset number of iterations, such as 20-50. After the iterations are complete, select the prediction from the surrogate model. EER The highest combination of control variables is taken as the optimal solution for this optimization. N com , N c , N ,Φ).

[0124] Step 305: The optimal solution is sent to the inverter driver and electronic expansion valve controller of the air conditioning system, causing the air conditioning system to switch to the new parameters.

[0125] Step 306: Establish a steady-state operation for the air conditioning system and initiate periodic data acquisition and optimization. For example, the air conditioning system operates stably for a preset time interval, such as 5-15 minutes, under new parameter settings. Once the system reaches a new steady state, return to step S302 to begin the next cycle of data sensing and optimization. This cycle continues, enabling the air conditioning system to adaptively track changes in the external environment and workload, continuously maintaining optimal energy efficiency.

[0126] This invention establishes separate prediction models for the cooling capacity and energy consumption of rack-mounted variable frequency air conditioners. It integrates these models to construct an energy efficiency ratio objective function. Based on real-time data, a Bayesian optimization algorithm is used to find the optimal combination of operating parameters in real time, dynamically adjusting the compressor speed, condenser fan speed, evaporator fan speed, and the opening of the electronic expansion valve. While meeting the load requirements of IT equipment within the rack, the system ensures that the rack-mounted air conditioning system always operates at its optimal energy efficiency point under different outdoor temperatures and IT equipment loads, effectively improving the energy efficiency and reliability of rack-mounted variable frequency systems.

[0127] Example 2: Based on Example 1, step 1 of the method of the present invention may specifically include:

[0128] Step 101: Determine the model's input and output variables: Define the input and output layer nodes of the neural network. The input variable is the outdoor ambient temperature. T amb Air conditioning return air temperature T return Compressor speed N com Condenser fan speed N c Evaporator fan speed N e And the opening degree Φ of the electronic expansion valve; the output variable is the prediction target of the model, which is the cooling capacity of the air conditioning system. Q .

[0129] Step 102: Collect and preprocess the input variable data. There are six input variables, covering their entire expected operating range. After the air conditioning system reaches steady state, the evaporator outlet superheat stabilizes within the range of 5-8℃. The superheat can be adjusted by changing the opening of the electronic expansion valve. Record multiple sets of steady-state input variable data and their corresponding actual cooling capacity values. Specifically:

[0130] Step 1021: When collecting data on input variables, the cooling capacity Q can be obtained by measuring the enthalpy difference method; or by calculating it after measuring with an anemometer and temperature sensor. When calculating, the formula is used:

[0131] Q = ρvAC e(T e,a,i -T e,a,o )，

[0132] in, ρ air density (kg / m³) 3 ); v The evaporator outlet air velocity (m / s); A Evaporator outlet area (m²) 2 ); C e The specific heat capacity of the air at the evaporator outlet is J / (kg·K); T e,a,i The evaporator inlet air temperature (°C) is the same as the air conditioner return air temperature. T return ; T e,a,o The evaporator outlet temperature (°C);

[0133] v To obtain the average value of multiple wind speed measuring points evenly distributed at the evaporator air outlet; T e,a,i To evenly distribute multiple temperature measuring points at the evaporator air inlet, the average value of the measurements is calculated. T e,a,o To evenly distribute multiple temperature measuring points at the evaporator outlet, the average value of the measurements is calculated.

[0134] Step 1022: Preprocess the collected data, including:

[0135] Data cleaning: Remove abnormal data points caused by sensor malfunctions or unstable operation of the air conditioning system.

[0136] Data normalization: The maximum-minimum normalization method is used to normalize the compressor speed. N com Condenser fan speed N c Evaporator fan speed N e The opening Φ of the electronic expansion valve, scaled to the [0,1] interval, is calculated using the following formula:

[0137] X norm =( X-X min ) / (X max -X min )，

[0138] in, XThe original value, X min and X max These are the minimum and maximum values ​​of a certain variable.

[0139] Step 103: Construct the neural network structure of the model: Construct a multi-layer feedforward neural network, the structure of which includes:

[0140] An input layer contains at least 6 neurons, each corresponding to one of the 6 input variables determined in S101.

[0141] At least one hidden layer: The number of hidden layers and their neurons are determined by hyperparameter optimization based on the complexity of the dataset;

[0142] Preferably, two hidden layers are used, with the first hidden layer having 128 neurons and the second hidden layer having 64 neurons;

[0143] One output layer: contains one neuron, and outputs the predicted cooling capacity value;

[0144] Activation functions: ReLU activation function is used for neurons in the hidden layer, and linear activation function is used for neurons in the output layer;

[0145] Step 104: Divide the dataset and train the model based on the preprocessed data: Randomly divide the complete dataset after S102 preprocessing into training set, validation set and test set according to a predetermined ratio;

[0146] The training set is input into the neural network model constructed in step 103, the mean squared error is used as the loss function, the Adam optimization algorithm is used for iterative training, and the validation set is used to monitor the model performance during the training process, and overfitting is prevented by early stopping.

[0147] Step 105: Perform model validation and performance evaluation. This involves using a reserved test set to evaluate the performance of the trained final model and calculating the predicted cooling capacity. Q pred Compared with the true value Q real The root mean square error, mean absolute error, and coefficient of determination are used to quantify the model's prediction accuracy and generalization ability.

[0148] Example 3: Based on Examples 1 and 2, step 2 of the method of the present invention specifically includes:

[0149] Step 201: Determine the model input and output variables: Input variables include compressor speed. N com Condenser fan speed N c Evaporator fan speed Ne Condenser inlet air temperature T c,a,i and evaporator inlet air temperature T e,a,i The output variable is the total power consumption of the air conditioning system. W total ，

[0150] Intermediate physical quantity: Condensation pressure P c,r Evaporation pressure P e,r ;

[0151] Step 202: Collect and preprocess the input variable data:

[0152] Condensation pressure P c,r and evaporation pressure P e,r The pressure is obtained by directly measuring the pressure at the refrigerant-side outlet of the condenser and evaporator; or by measuring the temperature at the refrigerant-side outlet of the condenser and evaporator and then calculating the temperature by referring to the refrigerant's physical property parameter table.

[0153] By changing T c,a,i and T e,a,i Under various operating conditions, the normalized speeds of the compressor, condenser fan, and evaporator fan are varied. After the system reaches steady state, multiple sets of steady-state data are recorded. The collected data includes all input variables, intermediate physical quantities, and compressor power from step 201. W com Condenser fan power W c Evaporator fan power W e and total power W total ,

[0154] Preprocessing of the collected raw dataset includes:

[0155] Data cleaning: Remove abnormal data points caused by sensor malfunctions or unstable operation of the air conditioning system;

[0156] Data normalization: The maximum-minimum normalization method is used to normalize the compressor speed ( N com ), condenser fan speed ( N c ), Evaporator fan speed ( N e) Scale to the [0,1] interval.

[0157] Step 203: Establish energy consumption models for each component of the rack-mounted inverter air conditioner:

[0158] Step 2031: Establish a compressor power consumption sub-model and predict compressor power consumption. W com A compressor power consumption sub-model is established, which is represented by the compressor speed. N com and stress correction factor F p Functions:

[0159] W com = f(N com ，F p )，

[0160] Stress correction factor F p From condensation pressure P c,r and evaporation pressure P e,r The ratios reflect the impact of the air conditioning system's pressure ratio on the compressor's power consumption:

[0161] W com = C(P c,r / P e,r ） (k-1) / k ，

[0162] C This is a constant used for model calibration; k The adiabatic index of the refrigerant is obtained by consulting the refrigerant's physical property parameters.

[0163] Step 2032: Establish a wind turbine power consumption sub-model to predict the wind turbine's power consumption. W c and W e The model establishes a sub-model for wind turbine power consumption, simplifying the model into a cubic function of the wind turbine speed:

[0164] W c = a 1 N c 3 +b 1N c 2 +c 1 N c +d 1 ；W e = a 2 N e 3 +b 2 N e 2 +c 2 N e +d 2 ；

[0165] a 1 a 2 、b 1 , b 2 , c 1 、c 2 、d 1 、d 2 These are the fitting coefficients. Fitting coefficients can be obtained by performing multinomial fitting using Python's NumPy and SciPy libraries, or determined through other conventional methods.

[0166] Step 204: Perform model integration to obtain the total energy consumption model: W total =W com +W c +W e ,

[0167] Step 205: Perform model validation and performance evaluation. Using the reserved test dataset, substitute the input variables of the test set into the integrated total energy consumption model to calculate the total power prediction value. W total,pred ;

[0168] Compare the predicted total power with the actual measured value. W total,real By comparing the results, indicators such as root mean square error and mean absolute percentage error are calculated to quantify the overall prediction accuracy of the model.

[0169] Example 4: The present invention also provides a real-time energy efficiency optimization control device for a data center rack inverter air conditioner, comprising: at least one memory and at least one processor;

[0170] The at least one memory is used to store a machine-readable program;

[0171] The at least one processor is used to call the machine-readable program to execute the real-time energy efficiency optimization control method for a data center rack inverter air conditioner.

[0172] The information interaction and execution of readable programs by the processor within the aforementioned device are based on the same concept as the method embodiments of the present invention, and the specific details can be found in the descriptions in the method embodiments of the present invention, and will not be repeated here.

[0173] Similarly, the advantages of the device of the present invention are:

[0174] 1. Global optimization and adaptation: Bayesian optimization is good at handling black box and nonlinear problems. It can escape local optima, find the global optimum or near the global optimum, and automatically adjust according to changes in operating conditions.

[0175] 2. Safe and reliable: The optimization process takes place in the model's virtual environment, eliminating the need for frequent and dangerous perturbation tests on the real system. By setting safety constraints, it is ensured that the recommended parameters are always within a safe range.

[0176] 3. Strong real-time performance: The optimization calculation is based on a pre-trained high-performance model, which is fast and can complete the optimization in seconds, meeting the requirements of real-time control.

[0177] 4. Significant energy saving effect: By continuously tracking the optimal EER point, the system energy consumption can be reduced to the greatest extent, especially in scenarios where the actual load and outdoor temperature fluctuate frequently, the energy saving effect is particularly outstanding.

[0178] 5. High engineering practicality: This method does not require major modifications to existing hardware systems; the core is the upgrade of software algorithms, making it easy to implement and deploy.

[0179] Example 5: The present invention also provides a real-time energy efficiency optimization and control device for a data center rack-mounted inverter air conditioner, comprising: a cooling capacity prediction model management module, an energy consumption prediction model management module, and an optimization module.

[0180] The cooling capacity prediction model management module establishes a neural network-based prediction model for the cooling capacity of cabinet inverter air conditioners.

[0181] The energy consumption prediction model management module establishes an energy consumption prediction model for the cabinet inverter air conditioner;

[0182] The optimization module constructs a Bayesian optimization objective function to perform real-time energy efficiency optimization for rack-mounted inverter air conditioners.

[0183] Step 301: Preload the pre-trained cabinet inverter air conditioner cooling capacity prediction model and energy consumption prediction model, and define the control variables and the safe operating boundary of the cabinet inverter air conditioner.

[0184] Controlled variable: Compressor speed N com Condenser fan speed N c Evaporator fan speed N e Electronic expansion valve opening Φ,

[0185] Step 302: Periodically collect the status parameters of the rack-mounted inverter air conditioner in real time. The status parameters include the outdoor ambient temperature. T amb Air conditioning return air set temperature T set Condensing pressure P c,r Evaporation pressure P e,r The target cooling capacity required at present

[0186] Step 303: Construct the Bayesian optimization objective function: In each optimization cycle, based on the collected state parameters, construct the energy efficiency ratio objective function under the current operating conditions:

[0187] EER = f ( N com , N c , N e ,Φ)= Q pred / W total,pred ,

[0188] Combine the current control variables and state parameters ( T amb , T set , N com , N c , N e The input (Φ) is given to the cabinet inverter air conditioner cooling capacity prediction model to obtain the predicted cooling capacity. Q pred ,

[0189] Combine the current control variables and state parameters (N com , N c , N e , P c,r , P e,r The data is input into the cabinet inverter air conditioner energy consumption prediction model to obtain the predicted total energy consumption of the air conditioning system. W total,pred ,

[0190] Step 304: Use the Bayesian optimization algorithm to maximize the Bayesian objective function and obtain the prediction. EER The highest combination of control variables is taken as the optimal solution. N com , N c , N e ,Φ),

[0191] Step 305: The optimal solution is sent to the inverter driver and electronic expansion valve controller of the air conditioning system, causing the air conditioning system to switch to the new parameters.

[0192] Step 306: Enable the air conditioning system to operate in a steady state and enter the periodic data collection and optimization process.

[0193] The information interaction and execution process between the modules in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description in the method embodiment of the present invention, and will not be repeated here.

[0194] Similarly, the advantages of the device of the present invention are:

[0195] 1. Global optimization and adaptation: Bayesian optimization is good at handling black box and nonlinear problems. It can escape local optima, find the global optimum or near the global optimum, and automatically adjust according to changes in operating conditions.

[0196] 2. Safe and reliable: The optimization process takes place in the model's virtual environment, eliminating the need for frequent and dangerous perturbation tests on the real system. By setting safety constraints, it is ensured that the recommended parameters are always within a safe range.

[0197] 3. Strong real-time performance: The optimization calculation is based on a pre-trained high-performance model, which is fast and can complete the optimization in seconds, meeting the requirements of real-time control.

[0198] 4. Significant energy saving effect: By continuously tracking the optimal EER point, the system energy consumption can be reduced to the greatest extent, especially in scenarios where the actual load and outdoor temperature fluctuate frequently, the energy saving effect is particularly outstanding.

[0199] 5. High engineering practicality: This method does not require major modifications to existing hardware systems; the core is the upgrade of software algorithms, making it easy to implement and deploy.

[0200] It should be noted that not all steps and modules in the above processes and device structures are mandatory; some steps or modules may be omitted as needed. The execution order of the steps is not fixed and can be adjusted as required. The device structures described in the above embodiments can be physical or logical structures; that is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or they may be jointly implemented by certain components in multiple independent devices.

[0201] The embodiments described above are merely preferred embodiments for fully illustrating the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A real-time energy efficiency optimization control method for air conditioners in data centers, characterized in that include: Step 1: Establish a neural network-based model for predicting the cooling capacity of rack-mounted inverter air conditioners; Step 2: Establish a cabinet inverter air conditioner energy consumption prediction model; Step 3: Construct a Bayesian optimization-based objective function for real-time energy efficiency optimization of rack-mounted inverter air conditioners: Step 301: Preload the pre-trained cooling capacity prediction model and energy consumption prediction model for rack-mounted inverter air conditioners, define control variables and the safe operating boundary of the rack-mounted inverter air conditioner. Controlled variable: Compressor speed N com Condenser fan speed N c Evaporator fan speed N e Electronic expansion valve opening Ф Step 302: Periodically collect the status parameters of the rack-mounted inverter air conditioner in real time. The status parameters include the outdoor ambient temperature. T amb Air conditioning return air set temperature T set Condensing pressure P c,r Evaporation pressure P e,r The target cooling capacity required at present Step 303: Construct the Bayesian optimization objective function: In each optimization cycle, based on the collected state parameters, construct the energy efficiency ratio objective function under the current operating conditions: EER = f ( N com , N c , N e ,Ф)= Q pred / W total,pred , Combine the current control variables and state parameters ( T amb , T set , N com , N c , N e The input (Ф) is given to the cabinet inverter air conditioner cooling capacity prediction model to obtain the predicted cooling capacity. Q pred , Combine the current control variables and state parameters ( N com , N c , N e , P c,r , P e,r The data is input into the cabinet inverter air conditioner energy consumption prediction model to obtain the predicted total energy consumption of the air conditioning system. W total,pred , Step 304: Use the Bayesian optimization algorithm to maximize the Bayesian objective function and obtain the prediction. EER The highest combination of control variables is taken as the optimal solution. N com , N c , N e ,Ф) Step 305: The optimal solution is sent to the inverter driver and electronic expansion valve controller of the air conditioning system, causing the air conditioning system to switch to the new parameters. Step 306: Enable the air conditioning system to operate in a steady state and enter the periodic data collection and optimization process.

2. The real-time energy efficiency optimization control method for air conditioning in data centers according to claim 1, Its characteristic is that step 1 specifically includes: Step 101: Determine the model's input and output variables: Define the input and output layer nodes of the neural network. The input variable is the outdoor ambient temperature. T amb Air conditioning return air temperature T return Compressor speed N com Condenser fan speed N c Evaporator fan speed N e And the opening degree Ф of the electronic expansion valve; the output variable is the model's prediction target, which is the cooling capacity of the air conditioning system. Q ; Step 102: Collect and preprocess the data of the input variables; Step 103: Construct the neural network structure of the model: Construct a multi-layer feedforward neural network, the structure of which includes: An input layer must contain at least 6 neurons. At least one hidden layer; One output layer: contains one neuron, and outputs the predicted cooling capacity value; Activation functions: ReLU activation function is used for neurons in the hidden layer, and linear activation function is used for neurons in the output layer; Step 104: Based on the preprocessed data, divide the dataset and train the model; Step 105: Perform model validation and performance evaluation.

3. The real-time energy efficiency optimization control method for air conditioning in data centers according to claim 2, characterized in that: Step 102 includes: Step 1021: When collecting data on input variables, the cooling capacity... Q It can be obtained by measuring the enthalpy difference; or by calculating it after measuring with an anemometer and a temperature sensor. When calculating, the formula is used: Q=ρvAC e (T e,a,i -T e,a,o )， in, ρ air density kg / m³ 3 ; v The evaporator outlet air velocity is measured in m / s. A The area of ​​the evaporator outlet is m. 2 ; C e The specific heat capacity of the air at the evaporator outlet is J / (kg·K); T e,a,i The evaporator inlet temperature (°C) is the air conditioner return air temperature. T return ; T e,a,o The evaporator outlet temperature is °C. v To obtain the average value of multiple wind speed measuring points evenly distributed at the evaporator air outlet; T e,a,i To evenly distribute multiple temperature measuring points at the evaporator air inlet, the average value of the measurements is calculated. T e,a,o To evenly distribute multiple temperature measuring points at the evaporator outlet, the average value of the measurements is calculated. Step 1022: Preprocess the collected data, including: Data cleaning: Remove abnormal data points caused by sensor malfunctions or unstable operation of the air conditioning system. Data normalization: The maximum-minimum normalization method is used to normalize the compressor speed. N com Condenser fan speed N c Evaporator fan speed N e The opening degree Ф of the electronic expansion valve, scaled to the [0,1] interval, is calculated using the following formula: X norm =( XX min ) / (X max -X min )， in, X The original value, X min and X max These are the minimum and maximum values ​​of a certain variable.

4. The real-time energy efficiency optimization control method for air conditioning in data centers according to claim 1, characterized in that: Step 2 specifically includes: Step 201: Determine the model input and output variables: Input variables include compressor speed. N com Condenser fan speed N c Evaporator fan speed N e Condenser inlet air temperature T c,a,i and evaporator inlet air temperature T e,a,i The output variable is the total power consumption of the air conditioning system. W total ， Intermediate physical quantity: Condensation pressure P c,r Evaporation pressure P e,r ; Step 202: Collect and preprocess the data of the input variables; Step 203: Establish energy consumption models for each component of the rack-mounted inverter air conditioner: Establish a compressor power consumption sub-model to predict compressor power consumption. W com ; Establish a fan power consumption sub-model to predict the power consumption of the condenser fan. W c Power consumption of evaporator fan W e , Step 204: Perform model integration to obtain the total energy consumption model: W total =W com +W c +W e ， Step 205: Perform model validation and performance evaluation.

5. A real-time energy efficiency optimization control method for air conditioning in a data center according to claim 4, characterized in that: Step 203 specifically includes: Step 2031: Establish a compressor power consumption sub-model, represented by compressor speed. N com and stress correction factor F p Functions: W com =f(N com ，F p )， Stress correction factor F p From condensation pressure P c,r and evaporation pressure P e,r The ratios reflect the impact of the air conditioning system's pressure ratio on the compressor's power consumption: W com =C(P) c,r / P e,r ） (k-1) / k ， C This is a constant used for model calibration; k The adiabatic index of the refrigerant is obtained by consulting the refrigerant's physical property parameters. Step 2032: Establish a sub-model of wind turbine power consumption, simplifying the model into a cubic function of wind turbine speed: W c = a 1 N c 3 +b 1 N c 2 +c 1 N c +d 1 ；W e = a 2 N e 3 +b 2 N e 2 +c 2 N e +d 2 ； a 1 a 2 、b 1 , b 2 , c 1 、c 2 、d 1 、d 2 represents the fitting coefficient.

6. The real-time energy efficiency optimization control method for air conditioning in data centers according to claim 1, characterized in that: Step 301 defines the control variables and the safe operating boundaries of the cabinet inverter air conditioner, including: Define the search space: Define the physically permissible and safe minimum and maximum values ​​for each control variable. Define constraints: Define the safety constraints for the operation of the air conditioning system, including: upper and lower limits of evaporation pressure, upper and lower limits of condensation pressure, upper and lower limits of superheat, and upper limit of compressor discharge temperature.

7. The real-time energy efficiency optimization control method for air conditioning in data centers according to claim 1, characterized in that: In step 302, the optimization of the objective function based on Bayesian optimization is carried out under the premise of satisfying the cooling capacity matching constraint, using the formula: (| Q pred -Q red | / Q red )≤δ, Q red Let δ represent the target cooling capacity and δ be the allowable relative error.

8. The real-time energy efficiency optimization control method for air conditioning in data centers according to claim 1, characterized in that: Step 304 specifically includes: Step 3041: Construct a surrogate model: Use Gaussian process regression to fit the mapping relationship between the control variables and the objective function EER. Step 3042: Using the expected improvement function, when recommending the next sampling point, simultaneously consider the expected improvement of the objective function and the probability of satisfying all constraints, based on the actual situation. Step 3043: Solve for the optimal observation point: Within the defined search space, find the next combination of candidate control variables that maximizes the expected improvement function value. Step 3044: Evaluation and Model Update: Combine candidate control variables and input them into the objective function to obtain the corresponding... EER The system determines the satisfaction of values ​​and constraints, and updates the surrogate model to make it more closely resemble the actual system response. Step 3045: Repeat steps 3042 to 3044 for a preset number of iterations. After the iterations are complete, select the prediction from the surrogate model. EER The highest combination of control variables is taken as the optimal solution for this optimization. N com , N c , N ,Ф).

9. A real-time energy efficiency optimization and control device for air conditioning in data centers, characterized in that: include: At least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to invoke the machine-readable program to execute the real-time energy efficiency optimization control method for a data center rack inverter air conditioner according to any one of claims 1 to 8.

10. A real-time energy efficiency optimization control device for air conditioning in data centers, characterized in that: include: The module includes a cooling capacity prediction model management module, an energy consumption prediction model management module, and an optimization module. The cooling capacity prediction model management module establishes a neural network-based prediction model for the cooling capacity of cabinet inverter air conditioners. The energy consumption prediction model management module establishes an energy consumption prediction model for the cabinet inverter air conditioner; The optimization module constructs a Bayesian optimization objective function to perform real-time energy efficiency optimization for rack-mounted inverter air conditioners. Step 301: Preload the pre-trained cabinet inverter air conditioner cooling capacity prediction model and energy consumption prediction model, and define the control variables and the safe operating boundary of the cabinet inverter air conditioner. Controlled variable: Compressor speed N com Condenser fan speed N c Evaporator fan speed N e Electronic expansion valve opening Ф Step 302: Periodically collect the status parameters of the rack-mounted inverter air conditioner in real time. The status parameters include the outdoor ambient temperature. T amb Air conditioning return air set temperature T set Condensing pressure P c,r Evaporation pressure P e,r The target cooling capacity required at present Step 303: Construct the Bayesian optimization objective function: In each optimization cycle, based on the collected state parameters, construct the energy efficiency ratio objective function under the current operating conditions: EER = f ( N com , N c , N e ,Ф)= Q pred / W total,pred , Combine the current control variables and state parameters ( T amb , T set , N com , N c , N e The input (Ф) is given to the cabinet inverter air conditioner cooling capacity prediction model to obtain the predicted cooling capacity. Q pred , Combine the current control variables and state parameters ( N com , N c , N e , P c,r , P e,r The data is input into the cabinet inverter air conditioner energy consumption prediction model to obtain the predicted total energy consumption of the air conditioning system. W total,pred , Step 304: Use the Bayesian optimization algorithm to maximize the Bayesian objective function and obtain the prediction. EER The highest combination of control variables is taken as the optimal solution. N com , N c , N e ,Ф) Step 305: The optimal solution is sent to the inverter driver and electronic expansion valve controller of the air conditioning system, causing the air conditioning system to switch to the new parameters. Step 306: Enable the air conditioning system to operate in a steady state and enter the periodic data collection and optimization process.

Images