A multi-mode energy-saving control method, system and device for a computer room air conditioner

Abstract

The present application relates to a kind of multi-modal machine room air conditioning energy-saving control method, system and device, its method includes: offline CFD airflow simulation, obtains the reference value of air supply floor opening rate and air valve opening degree;Online LSTM multiscale equipment power prediction, obtain the equipment power prediction value of future period, based on air supply temperature optimization model, dynamically calculate the target value of air supply floor opening rate and air valve opening degree;According to equipment power prediction value, the target value of air supply floor opening rate and air valve opening degree, the reference value of air supply floor opening rate and air valve opening degree carries out fuzzy PID control, obtains energy-saving control instruction.The present application adopts the scheme that offline CFD simulation is combined with online dynamic real-time control, solves the problem that simulation and actual of traditional scheme are disjointed;LSTM prediction provides feedforward compensation for fuzzy PID control, solves the pain point of traditional air conditioning control lagging adjustment, excessive refrigeration, realizes predictive accurate temperature control.

Description

Technical Field

[0001] This invention relates to the field of energy-saving control, specifically to a multimodal energy-saving control method, system, and device for computer room air conditioning. Background Technology

[0002] The current status of data center air conditioning energy consumption and energy-saving needs have the following characteristics: (1) Prominent energy consumption ratio: The data center air conditioning system is the main source of energy consumption in the data center, accounting for 35%-45% of the total energy consumption; as the power density of data center equipment increases, the air conditioning energy consumption problem becomes more and more serious; for example, in the annual power consumption of a certain super-large data center, the air conditioning system accounts for as much as 42%, and it is on the rise with the expansion of equipment. (2) Annual cooling demand: The data center needs to maintain stable temperature and humidity throughout the year (such as temperature 23±2℃, humidity 55±10%), and cooling is required even in winter; traditional air conditioning systems rely on mechanical cooling, resulting in energy waste; for example, in the northern region, the outdoor temperature in winter is as low as -10℃, but the data center still needs to turn on the compressor for cooling. (3) Redundancy design exacerbates energy consumption: In order to ensure data security, the data center air conditioning adopts N+N redundancy design, and the equipment operates at partial load for a long time, resulting in low efficiency; for example, the actual measurement of a certain financial data center shows that the average load rate of the air conditioning system is only 45%, but the energy consumption reaches 70% of the design value.

[0003] However, the existing energy-saving technologies for air conditioning in computer rooms have the following bottlenecks: (1) Insufficient utilization of natural cold sources: Although free cooling technologies (such as fresh air systems and heat pipe heat exchange) can reduce energy consumption by more than 30%, only 15% of large data centers in China have effectively applied them; the reasons include the difficulty of system integration, the complexity of switching and maintenance, and concerns about affecting equipment stability. (2) Limitations of variable frequency technology: Variable frequency air conditioners achieve energy saving by adjusting the compressor speed, but the single unit capacity is small (suitable for small computer rooms), and harmonic interference may affect communication equipment; a test of a telecommunications computer room showed that the energy saving rate of variable frequency air conditioners under partial load was only 18%, which did not meet expectations. (3) Defects in airflow organization: The mixing of hot and cold air in the computer room leads to local overheating, and the manager is forced to lower the temperature setpoint, resulting in excessive cooling; a test of a cloud computing center showed that the temperature difference at the rack inlet reached 8℃, resulting in an increase of 22% in air conditioning energy consumption. (4) Coarse control strategy: Traditional air conditioning systems rely on thermostats and cannot dynamically adapt to load changes; statistics from a bank data center showed that manual adjustment caused temperature fluctuations of ±3℃, and energy consumption fluctuations of up to 15%. Summary of the Invention

[0004] This invention provides a multimodal energy-saving control method, system, and device for computer room air conditioning to solve at least one of the above-mentioned technical problems.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A multimodal data center air conditioning energy-saving control method, comprising: Acquire basic parameters of the data center, as well as comprehensive environmental data, equipment power data, and equipment operating condition data; wherein, the environmental data includes temperature data, relative humidity data, and differential pressure data; Edge preprocessing is performed on the environmental data and the equipment power data to obtain a standardized feature matrix, and the real-time air enthalpy values ​​of the cold aisle inlet and server exhaust vent in the computer room are obtained. Offline CFD airflow simulation is performed based on the basic parameters of the computer room, the environmental data, the equipment power data, and the predicted heat generation of the equipment. Combined with the real-time air enthalpy value, the cooling demand and the baseline values ​​of the floor opening ratio and the opening degree of the air valve are calculated. Online LSTM multi-scale equipment power prediction is performed using historical equipment power data and the standardized feature matrix to obtain the predicted equipment power value for future periods, and the predicted equipment heat generation value is calculated based on the power-heat generation mapping model. Based on the supply air temperature optimization model, combined with the predicted equipment power and the cooling demand, the target values ​​of the supply air floor opening ratio and the air valve opening are dynamically calculated. Fuzzy PID control is performed based on the predicted power of the equipment, the target values ​​of the air supply floor opening ratio and the air valve opening, and the reference values ​​of the air supply floor opening ratio and the air valve opening to obtain energy-saving control commands.

[0006] Based on the above technical solution, the present invention can be further improved as follows.

[0007] Furthermore, edge preprocessing is performed on the environmental data and the device power data, specifically including: Data cleaning is performed on the device power data, temperature data, and relative humidity data. The time data and the power data, temperature data and relative humidity data of the cleaned equipment are subjected to feature engineering and standardization to obtain standardized single-dimensional features. A feature matrix is ​​constructed using the standardized single-dimensional features, resulting in a standardized feature matrix. The standardized feature matrix is ​​represented as follows: ; In the formula, express The standardized feature matrix at time step; express Device power data at any given time express Device power data at any given time; express Temperature data at any given time express Temperature data at any given time; express Relative humidity data at any given time, express Relative humidity data at any given time; Indicates hours, Indicates the day of the week. Indicates the month, and , and All data are time data encoded using sine / cosine encoding.

[0008] Furthermore, the formula for calculating the real-time air enthalpy value is as follows: ; In the formula, This indicates the real-time air enthalpy value. This represents real-time temperature data. This represents real-time relative humidity data.

[0009] Furthermore, offline CFD airflow simulation is performed based on the basic parameters of the computer room, the environmental data, the equipment power data, and the predicted heat generation of the equipment. Combined with the real-time air enthalpy, the baseline values ​​for cooling demand, floor perforation ratio, and damper opening are calculated, specifically including: Offline CFD airflow simulation is performed based on the basic parameters of the computer room, the environmental data, the equipment power data, and the predicted heat generation of the equipment to simulate the airflow distribution before and after the thermal aisle is closed in the computer room, thereby obtaining the air mass flow rate. Calculate the cooling requirements before and after the hot aisle is closed based on the air mass flow rate and the real-time air enthalpy. Based on the cooling demand before and after the hot aisle is closed, the baseline values ​​for the floor opening ratio and damper opening are calculated.

[0010] Furthermore, the formula for calculating the cooling demand is as follows: ; In the formula, This indicates the cooling demand. This indicates the air mass flow rate. This indicates the real-time air enthalpy value at the server's exhaust vent. This indicates the real-time air enthalpy value at the cold aisle inlet.

[0011] Furthermore, the power-heat generation mapping model is expressed as: ; In the formula, This represents the predicted heat output of the device. This represents the predicted power value of the device. This represents the model parameters of the power-calorific value mapping model, and The value range is [0.8, 1.0].

[0012] Furthermore, based on the supply air temperature optimization model, and combined with the predicted equipment power and the cooling demand, the target values ​​for the supply air floor opening ratio and the damper opening are dynamically calculated, specifically including: Based on the supply air temperature optimization model, the optimal supply air temperature setpoint is calculated according to the predicted power of the equipment, real-time temperature data and air mass flow rate. Based on the optimal supply air temperature setpoint and the cooling capacity requirement, the target values ​​for the air supply floor opening ratio and the air valve opening degree are dynamically calculated.

[0013] Furthermore, the air supply temperature optimization model is expressed as follows: ; In the formula, This represents the optimal supply air temperature setpoint. This represents the predicted power value of the device. This indicates the air mass flow rate. This indicates the specific heat capacity of air at constant pressure. This represents real-time temperature data; The model parameters of the air supply temperature optimization model are represented, and their values ​​range from [0.5, 1.2].

[0014] Furthermore, based on the predicted power of the equipment, the target values ​​of the air supply floor opening ratio and the damper opening, and the reference values ​​of the air supply floor opening ratio and the damper opening, fuzzy PID control is performed to obtain energy-saving control commands, specifically including: The pressure difference data before and after the cabinet, the temperature difference data of the hot and cold aisles, and the real-time equipment power data are used as real-time feedback quantities. The predicted power value of the equipment is used as the feedforward compensation amount; The target values ​​for the floor perforation rate and damper opening are used as target quantities. Use the baseline values ​​of the air supply floor opening ratio and the air valve opening degree as initial quantities; Fuzzy PID control is performed based on the real-time feedback, the feedforward compensation, the target, and the initial values ​​to obtain energy-saving control commands; The energy-saving control commands include control commands for air conditioning frequency adjustment, air guide plate angle, electric air valve opening degree, and air supply floor opening ratio.

[0015] Furthermore, it also includes: Calculate the rate of decrease in cooling demand based on the cooling demand before and after the hot aisle is closed. Obtain the air conditioning energy consumption before and after the equipment is operated, and calculate the air conditioning energy saving rate accordingly; The power usage efficiency after the equipment is activated is obtained, and the overall energy efficiency is calculated by combining the cooling demand reduction rate and the air conditioning energy saving rate. Determine whether the overall energy efficiency is greater than or equal to the preset energy efficiency threshold. If not, return to start re-running the energy-saving control of the computer room air conditioning. If yes, end the energy-saving control of the computer room air conditioning.

[0016] Furthermore, the formula for calculating the rate of decrease in cooling demand is as follows: ; In the formula, This indicates the rate of decrease in cooling demand. This indicates the cooling demand before the hot aisle is closed. This indicates the cooling demand after the hot aisle is closed; The formula for calculating the air conditioner's energy-saving rate is as follows: ; In the formula, This indicates the energy-saving rate of the air conditioner. This indicates the air conditioning energy consumption after the equipment has been activated. This indicates the air conditioning energy consumption before the equipment was activated; The formula for calculating comprehensive energy efficiency is as follows: ; In the formula, This indicates the overall energy efficiency. This indicates the power efficiency after the device is activated.

[0017] Furthermore, after the equipment is activated, it also includes: The model parameters of the power-heat generation mapping model and the air supply temperature optimization model are refitted based on real-time equipment operating data. Acquire actual airflow data, and then use the actual airflow data to reverse-correct the boundary conditions and turbulence parameters of the CFD model.

[0018] Furthermore, the environmental data also includes infrared thermal imaging data; The multimodal data center air conditioning energy-saving control method also includes: The air supply direction is dynamically adjusted based on the infrared thermal imaging data.

[0019] Based on the above-mentioned multimodal energy-saving control method for computer room air conditioning, the present invention also provides a multimodal energy-saving control system for computer room air conditioning.

[0020] A multimodal data center air conditioning energy-saving control system, applied to the multimodal data center air conditioning energy-saving control method described above, includes: The data acquisition module is used to acquire basic parameters of the computer room, as well as environmental data, equipment power data, and equipment operating condition data of the computer room in all dimensions; wherein, the environmental data includes temperature data, relative humidity data, and pressure difference data; An edge preprocessing module is used to perform edge preprocessing on the environmental data and the equipment power data to obtain a standardized feature matrix, and to obtain the real-time air enthalpy values ​​of the cold aisle inlet and server exhaust vent in the computer room. The offline CFD airflow simulation module is used to perform offline CFD airflow simulation based on the basic parameters of the computer room, the environmental data, the equipment power data and the predicted value of equipment heat generation, and to calculate the cooling demand and the benchmark values ​​of the floor opening ratio and the opening degree of the air valve in combination with the real-time air enthalpy value. The online LSTM multi-scale prediction module is used to perform online LSTM multi-scale equipment power prediction using historical equipment power data and the standardized feature matrix to obtain the equipment power prediction value for future periods, and calculate the equipment heat generation prediction value according to the power-heat generation mapping model. The air supply temperature optimization module is used to dynamically calculate the target values ​​of the air supply floor opening ratio and the air valve opening degree based on the air supply temperature optimization model, combined with the predicted equipment power value and the cooling demand. The fuzzy PID control module is used to perform fuzzy PID control based on the predicted power value of the equipment, the target values ​​of the air supply floor opening ratio and the air valve opening, and the reference values ​​of the air supply floor opening ratio and the air valve opening, to obtain energy-saving control commands.

[0021] Based on the above-mentioned multimodal data center air conditioning energy-saving control method, the present invention also provides a multimodal data center air conditioning energy-saving control device.

[0022] A multimodal data center air conditioning energy-saving control device includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the multimodal data center air conditioning energy-saving control method as described above.

[0023] The beneficial effects of this invention are as follows: This invention provides a multimodal energy-saving control method, system, and device for computer room air conditioning. It forms a complete digital twin closed loop through physical sensing, digital modeling, decision generation, physical execution, effect verification, and model iteration, with no data breakpoints. At the same time, it adopts a scheme that combines offline CFD simulation with online dynamic real-time control. Offline CFD simulation provides a reference boundary for online dynamic real-time control, and real-time running data corrects the CFD simulation model in reverse, solving the problem of simulation being disconnected from reality in traditional schemes. In addition, LSTM prediction provides feedforward compensation for fuzzy PID control, solving the pain points of lag adjustment and overcooling in traditional air conditioning control, and achieving predictive and precise temperature control. Attached Figure Description

[0024] Figure 1 This is a flowchart of a multimodal energy-saving control method for computer room air conditioning according to the present invention; Figure 2 This is a structural block diagram of a multimodal energy-saving control system for computer room air conditioning according to the present invention; Figure 3 This is a structural block diagram of a multimodal energy-saving control device for computer room air conditioning according to the present invention. Detailed Implementation

[0025] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0026] Example 1: like Figure 1 As shown, a multimodal energy-saving control method for computer room air conditioning includes: S1, acquire basic parameters of the data center, as well as comprehensive environmental data, equipment power data, and equipment operating condition data of the data center; wherein, the environmental data includes temperature data, relative humidity data, and differential pressure data; S2, perform edge preprocessing on the environmental data and the equipment power data to obtain a standardized feature matrix, and obtain the real-time air enthalpy values ​​of the cold aisle entrance and server exhaust vent in the computer room; S3. Based on the basic parameters of the computer room, the environmental data, the equipment power data and the predicted value of the equipment heat generation, perform offline CFD airflow simulation, and combine the real-time air enthalpy value to calculate the cooling demand and the benchmark values ​​of the floor opening ratio and the opening degree of the air valve. S4. Using historical equipment power data and the standardized feature matrix, perform online LSTM multi-scale equipment power prediction to obtain the equipment power prediction value for future periods, and calculate the equipment heat generation prediction value according to the power-heat generation mapping model. S5. Based on the supply air temperature optimization model, combined with the predicted equipment power value and the cooling demand, dynamically calculate the target values ​​of the supply air floor opening ratio and the air valve opening degree. S6. Based on the predicted power value of the equipment, the target values ​​of the air supply floor opening ratio and the air valve opening, and the reference values ​​of the air supply floor opening ratio and the air valve opening, fuzzy PID control is performed to obtain energy-saving control instructions.

[0027] The steps of the method of the present invention will be illustrated below with specific examples.

[0028] In the computer room of this invention, the server racks are arranged face-to-face and back-to-back to form cold and hot aisles. Transparent acrylic panels are installed on the top of the cold and hot aisles, and the hot aisles are enclosed with fireproof curtains. The opening rate of the air supply floor can be dynamically adjusted.

[0029] In S1 of this embodiment: S1 is a real-time, synchronous acquisition step of all physical quantities in the computer room. It can collect data through distributed sensing terminals to provide raw data input for the entire process, covering three core data sources: environment, equipment power, and equipment operating conditions. Table 1 below is the data acquisition table for environment, equipment power, and equipment operating conditions.

[0030] Table 1: Data Collection Table for Environment, Equipment Power, and Equipment Operating Conditions In addition, basic parameters of the data center include data such as data center size and rack layout.

[0031] The data acquired in S1 is synchronously transmitted to the edge computing gateway via the communication module, providing a data source for subsequent edge preprocessing. The communication module includes a wired transmission module and a wireless transmission module. In wired transmission, a Modbus TCP / IP protocol is used via an industrial Ethernet switch to transmit high-frequency, high-reliability data such as temperature data, relative humidity data, equipment power data, and equipment operating condition data from fixed locations. In wireless transmission, a LoRa wireless gateway is used to transmit data from dispersed and temporary locations such as differential pressure data and infrared thermal imaging data, with a transmission latency of ≤100ms. All data is ultimately converged at the edge computing gateway for unified preprocessing.

[0032] In S2 of this embodiment: Edge preprocessing is performed on the environmental data and the device power data, specifically including: Data cleaning is performed on the device power data, temperature data, and relative humidity data. The time data and the power data, temperature data and relative humidity data of the cleaned equipment are subjected to feature engineering and standardization to obtain standardized single-dimensional features. A feature matrix is ​​constructed using the standardized single-dimensional features, resulting in a standardized feature matrix. The standardized feature matrix is ​​represented as follows: ; In the formula, express The standardized feature matrix at time step; express Device power data at any given time express Device power data at any given time; express Temperature data at any given time express Temperature data at any given time; express Relative humidity data at any given time, express Relative humidity data at any given time; Indicates hours (0~23). It indicates the weekdays (1-7). Indicates the month (1-12), and , and All data is time data encoded using sine / cosine encoding; taking hours as an example, the formula for sine / cosine encoding is: , .

[0033] Specifically, the edge preprocessing in S2 is to address the problems of high noise, inconsistent format, and inability to be directly input into the model from the original data. It also performs pre-calculation of basic physical quantities, reducing the burden on subsequent model decisions. The core operations of edge preprocessing are shown in Figure 2.

[0034] Table 2: Core Operation Table for Edge Preprocessing Preferably, the formula for calculating the real-time air enthalpy is: ; In the formula, This indicates the real-time air enthalpy value. This represents real-time temperature data. This represents real-time relative humidity data.

[0035] In S3 of this embodiment: S3 specifically includes: Offline CFD airflow simulation is performed based on the basic parameters of the computer room, the environmental data (temperature data, relative humidity data, pressure difference data), the equipment power data, and the predicted heat generation of the equipment to simulate the airflow distribution before and after the thermal aisle is closed in the computer room, thereby obtaining the air mass flow rate. Calculate the cooling requirements before and after the hot aisle is closed based on the air mass flow rate and the real-time air enthalpy. Based on the cooling demand before and after the hot aisle is closed, the baseline values ​​for the floor opening ratio and damper opening are calculated.

[0036] Specifically, the CFD airflow simulation uses the k-ε turbulence model with a mesh size ≤ 50 mm and sets the boundary conditions as follows: supply air temperature 18~25℃ and return air temperature 35~40℃.

[0037] The S3 verifies the hot and cold aisle closure scheme and quantifies the cooling demand through numerical simulation, providing optimal reference parameters for real-time control and avoiding blind adjustments.

[0038] Preferably, the formula for calculating the cooling demand is: ; In the formula, This indicates the cooling demand. This indicates the air mass flow rate. This indicates the real-time air enthalpy value at the server's exhaust vent. This indicates the real-time air enthalpy value at the cold aisle inlet.

[0039] In S4 of this embodiment: The LSTM multi-scale device power prediction is based on a 3-layer shared LSTM network (64 neurons per layer) to extract temporal features. Through the structure of shared feature layers and independent output heads (with an independent fully connected layer configured for each prediction cycle), it outputs device power prediction values ​​at multiple scales of 15min / 1h / 24h. The training process adopts the mean squared error loss function and L2 regularization, and the prediction accuracy is verified by MAPE and RMSE.

[0040] In an LSTM network, the activation function of the hidden layer is the Tanh activation function (i.e., the hyperbolic tangent activation function), and the activation function of the output layer is the Sigmoid activation function (predicted values ​​are compressed to [0,1]); where: ; ; ; ; ; ; In the formula, for..., It is the Sigmoid activation function. The hyperbolic tangent activation function is used. For Hadama accumulation, , , , These represent the current time-stamp output vectors for the forget gate, input gate, output gate, and candidate cell state, respectively. , , , These are the weight matrices for the forget gate, input gate, output gate, and candidate cell states, respectively. , , , These are the bias vectors for the forget gate, input gate, output gate, and candidate cell state, respectively. The hidden state of the previous moment With the current input feature vector The concatenated vector, Let be the cell state vector at the current moment. This is the cell state vector from the previous time step. This is the hidden state vector at the current time step.

[0041] Design a multi-scale output layer: ;in, For the model's predicted output, This is the output layer weight matrix. For the first The hidden state of the layer This is the output bias.

[0042] The mean squared error loss function is expressed as: ; In the formula, This is the total loss function value. The total number of samples in the training batch. For the first The model prediction value for each sample. For the first The true label value of each sample For the first The squared prediction error for each sample, This is the regularization coefficient (typical value 0.001). For network layer indexing, For the first l The weight matrix of the layer network, For the first l The squared L2 norm of the weight matrix of the layer network.

[0043] When verifying prediction accuracy using MAPE and RMSE: , .

[0044] The standardized feature matrix is ​​directly input into the LSTM layer for temporal feature extraction. Multi-scale prediction achieves prediction at different periods through independent output heads, but shares the LSTM feature extractor. The prediction results... , Predicted power value of drive equipment.

[0045] Preferably, the power-heat generation mapping model is expressed as: ; In the formula, This represents the predicted heat output of the device. This represents the predicted power value of the device. This represents the model parameters of the power-calorific value mapping model, and The value range is [0.8, 1.0].

[0046] In S5 of this embodiment: S5 specifically includes: Based on the supply air temperature optimization model, the optimal supply air temperature setpoint is calculated according to the predicted power of the equipment, real-time temperature data and air mass flow rate. Based on the optimal supply air temperature setpoint and the cooling capacity requirement, the target values ​​for the air supply floor opening ratio and the air valve opening degree are dynamically calculated.

[0047] Preferably, the air supply temperature optimization model is expressed as: ; In the formula, This represents the optimal supply air temperature setpoint. This represents the predicted power value of the device. This indicates the air mass flow rate. This represents the specific heat capacity of air at constant pressure (1.005 kJ / kg·℃). This represents real-time temperature data; The model parameters of the air supply temperature optimization model are represented, and their values ​​range from [0.5, 1.2].

[0048] In S6 of this embodiment: S6 specifically includes: The pressure difference data before and after the cabinet, the temperature difference data of the hot and cold aisles, and the real-time equipment power data are used as real-time feedback quantities. The predicted power value of the equipment is used as the feedforward compensation amount; The target values ​​for the floor perforation rate and damper opening are used as target quantities. Use the baseline values ​​of the air supply floor opening ratio and the air valve opening degree as initial quantities; Fuzzy PID control is performed based on the real-time feedback, the feedforward compensation, the target, and the initial values ​​to obtain energy-saving control commands; The energy-saving control commands include control commands for air conditioning frequency adjustment, air guide plate angle, electric air valve opening degree, and air supply floor opening ratio.

[0049] Specifically, in the fuzzy PID control process, the PID parameters are dynamically adjusted according to the temperature difference data ΔT between the hot and cold channels and the pressure difference data ΔP between the front and rear of the cabinet. The specific parameter tuning is shown in Table 3 below.

[0050] Table 3: Parameter Tuning Table After receiving the energy-saving control command, it is broken down into the corresponding actuator (equipment) action signals: Variable frequency air conditioner controller: frequency adjustment command (20-120Hz), air supply temperature setpoint; Intelligent air guide plate system: air supply angle adjustment command; Electric air valve matrix: Air supply floor opening ratio adjustment command (0%-50%); Action execution: Each actuator completes precise adjustments according to instructions, realizing dynamic optimization of air supply parameters and airflow organization; Status feedback: Real-time acquisition of actual operating data of the actuator (actual frequency, opening degree, air supply temperature) and feedback to the environmental sensing stage to form a control feedback closed loop and verify the adjustment effect.

[0051] Preferably, the environmental data further includes infrared thermal imaging data; the multimodal data center air conditioning energy-saving control method further includes: dynamically adjusting the air supply direction based on the infrared thermal imaging data.

[0052] Specifically, infrared thermal imaging data is obtained by measuring the surface temperature of non-contact devices, which is used to locate overheated components and adjust the airflow direction to quickly dissipate heat from the overheated components.

[0053] In this embodiment, S3 to S6 are the virtual-real fusion model calculation and decision generation process. This process is divided into two parallel branches: offline CFD simulation optimization (S3) and real-time intelligent control decision (S4 to S6). Offline CFD simulation provides a reference boundary for real-time control, and real-time equipment operating data corrects the simulation model in reverse, forming a virtual-real linkage digital twin capability.

[0054] In this embodiment, the multimodal data center air conditioning energy-saving control method of the present invention further includes S7: Calculate the rate of decrease in cooling demand based on the cooling demand before and after the hot aisle is closed. Obtain the air conditioning energy consumption before and after the equipment is operated, and calculate the air conditioning energy saving rate accordingly; The power usage efficiency after the equipment is activated is obtained, and the overall energy efficiency is calculated by combining the cooling demand reduction rate and the air conditioning energy saving rate. Determine whether the overall energy efficiency is greater than or equal to a preset energy efficiency threshold (e.g., 15%). If not, return to start re-running the energy-saving control of the computer room air conditioning. If yes, end the energy-saving control of the computer room air conditioning.

[0055] In addition, model accuracy can be evaluated: calculate the MAPE and RMSE of the LSTM prediction to verify whether the prediction error is less than 5%.

[0056] Preferably, the formula for calculating the rate of decrease in cooling demand is: ; In the formula, This indicates the rate of decrease in cooling demand. This indicates the cooling demand before the hot aisle is closed. This indicates the cooling demand after the hot aisle is closed; The formula for calculating the air conditioner's energy-saving rate is as follows: ; In the formula, This indicates the energy-saving rate of the air conditioner. This indicates the air conditioning energy consumption after the equipment has been activated. This indicates the air conditioning energy consumption before the equipment was activated; The formula for calculating comprehensive energy efficiency is as follows: ; In the formula, This indicates the overall energy efficiency. This indicates the power efficiency after the device is activated.

[0057] In this embodiment, the multimodal data center air conditioning energy-saving control method of the present invention further includes S8: Model parameter correction: The model parameters of the power-heat generation mapping model and the air supply temperature optimization model are refitted based on real-time equipment operating data; actual airflow data is obtained, and the boundary conditions and turbulence parameters of the CFD model are corrected in reverse based on the actual airflow data.

[0058] Control strategy optimization: Based on the adjustment effect, optimize the parameter tuning rules of fuzzy PID, adjust the hyperparameters and time window length of LSTM model, and improve prediction and control accuracy.

[0059] Digital twin linkage: Through the digital twin platform, the real-time mapping, simulation and strategy pre-play of the data center's operating status can be realized, and the optimal control strategy can be continuously output.

[0060] The corrected model parameters and optimized control strategies are fed back into CFD simulation and intelligent control to complete the closed-loop iteration of the entire process and achieve continuous improvement in energy efficiency.

[0061] In summary, the method of this invention achieves the following: On-site survey: Measure the dimensions of the computer room, the layout of the server racks, and the existing airflow organization; CFD modeling: Simulate the airflow distribution before and after closure to determine the optimal closure scheme; Hardware deployment: Installation of acrylic panels, fireproof curtains, sensors, and electric air valves; Control system development: implementation of fuzzy PID algorithm and LSTM prediction model; Parameter tuning: Adjust empirical coefficients such as α and β based on actual operating data; Continuous optimization: Monitor and adjust control strategies in real time through a digital twin platform; This invention reduces the mixing of hot and cold air through physical isolation, achieves dynamic optimization of air supply parameters through intelligent control, and finally quantifies the energy-saving effect through a comprehensive energy efficiency model, forming a complete optimization closed loop.

[0062] Example 2: Based on the above-mentioned multimodal energy-saving control method for computer room air conditioning, the present invention also provides a multimodal energy-saving control system for computer room air conditioning.

[0063] like Figure 2 As shown, a multimodal data center air conditioning energy-saving control system, applied to the multimodal data center air conditioning energy-saving control method described above, includes: The data acquisition module is used to acquire basic parameters of the computer room, as well as environmental data, equipment power data, and equipment operating condition data of the computer room in all dimensions; wherein, the environmental data includes temperature data, relative humidity data, and pressure difference data; An edge preprocessing module is used to perform edge preprocessing on the environmental data and the equipment power data to obtain a standardized feature matrix, and to obtain the real-time air enthalpy values ​​of the cold aisle inlet and server exhaust vent in the computer room. The offline CFD airflow simulation module is used to perform offline CFD airflow simulation based on the basic parameters of the computer room, the environmental data, the equipment power data and the predicted value of equipment heat generation, and to calculate the cooling demand and the benchmark values ​​of the floor opening ratio and the opening degree of the air valve in combination with the real-time air enthalpy value. The online LSTM multi-scale prediction module is used to perform online LSTM multi-scale equipment power prediction using historical equipment power data and the standardized feature matrix to obtain the equipment power prediction value for future periods, and calculate the equipment heat generation prediction value according to the power-heat generation mapping model. The air supply temperature optimization module is used to dynamically calculate the target values ​​of the air supply floor opening ratio and the air valve opening degree based on the air supply temperature optimization model, combined with the predicted equipment power value and the cooling demand. The fuzzy PID control module is used to perform fuzzy PID control based on the predicted power value of the equipment, the target values ​​of the air supply floor opening ratio and the air valve opening, and the reference values ​​of the air supply floor opening ratio and the air valve opening, to obtain energy-saving control commands.

[0064] Specifically, the data acquisition module includes: (1) Temperature and humidity sensor array: Distribution: One set is deployed every 3 meters in the hot and cold aisles (accuracy ±0.5℃, ±2%RH); Function: To monitor temperature gradient and humidity changes and identify local hotspots.

[0065] (2) Differential pressure sensor: Deployment: At the junction of hot and cold aisles (range 0-50Pa); Function: To detect airflow leakage and assess the sealing effect.

[0066] (3) Infrared thermal imaging devices: Installation: Directly in front of the rack (16×16 resolution); Function: Non-contact measurement of surface temperature of equipment, and location of overheated components.

[0067] (4) Power monitoring devices: Access: PDU (Power Distribution Unit); Function: To obtain device power in real time.

[0068] Additionally, it should be noted that the specific functions of each module in the multimodal data center air conditioning energy-saving control system of the present invention are described in the specific steps of the multimodal data center air conditioning energy-saving control method of the present invention, and will not be repeated here.

[0069] Example 3: Based on the above-mentioned multimodal data center air conditioning energy-saving control method, the present invention also provides a multimodal data center air conditioning energy-saving control device.

[0070] like Figure 3 As shown, a multimodal data center air conditioning energy-saving control device includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the multimodal data center air conditioning energy-saving control method described above.

[0071] In one optional embodiment, a multimodal data center air conditioning energy-saving control device is provided, such as... Figure 3 As shown. Figure 3The multimodal data center air conditioning energy-saving control device shown includes a processor and a memory. The processor and memory are connected, for example, via a bus. Optionally, the multimodal data center air conditioning energy-saving control device may further include a transceiver, which can be used for data interaction between the multimodal data center air conditioning energy-saving control device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver is not limited to one unit, and the structure of this multimodal data center air conditioning energy-saving control device does not constitute a limitation on the embodiments of the present invention.

[0072] The processor can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a PLC (Programmable Logic Controller), a FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. The processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0073] A bus can include a pathway for transmitting information between the aforementioned components. The bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0074] The memory may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these.

[0075] The memory stores application code (computer program) that executes the present invention, and its execution is controlled by a processor. The processor executes the application code stored in the memory to implement the content shown in the foregoing method embodiments.

[0076] Among them, the multimodal data center air conditioning energy-saving control device can also be a terminal device. The terminal device can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0077] It should be noted that, Figure 3 The multimodal data center air conditioning energy-saving control device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0078] In summary, in the multimodal energy-saving control method, system, and device for computer room air conditioning of the present invention: (1) The present invention has a complete digital twin closed loop from physical perception → digital modeling → decision generation → physical execution → effect verification → model iteration, with no data breakpoints; (2) The offline CFD simulation in this invention provides a reference boundary for real-time control, and the real-time running data corrects the simulation model in reverse, solving the problem of the disconnect between simulation and reality in traditional schemes; (3) The LSTM prediction in this invention provides feedforward compensation for PID control, solves the pain points of lag adjustment and overcooling in traditional air conditioning control, and achieves predictive and precise temperature control. (4) Status data of all execution links are fed back throughout the process to realize real-time closed loop of adjustment-feedback-re-optimization, ensuring that the solution always maintains optimal energy efficiency when the load and environment of the computer room change dynamically.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multimodal energy-saving control method for computer room air conditioning, characterized in that, include: Acquire basic parameters of the data center, as well as comprehensive environmental data, equipment power data, and equipment operating condition data; wherein, the environmental data includes temperature data, relative humidity data, and differential pressure data; Edge preprocessing is performed on the environmental data and the equipment power data to obtain a standardized feature matrix, and the real-time air enthalpy values ​​of the cold aisle inlet and server exhaust vent in the computer room are obtained. Offline CFD airflow simulation is performed based on the basic parameters of the computer room, the environmental data, the equipment power data, and the predicted heat generation of the equipment. Combined with the real-time air enthalpy value, the cooling demand and the baseline values ​​of the floor opening ratio and the opening degree of the air valve are calculated. Online LSTM multi-scale equipment power prediction is performed using historical equipment power data and the standardized feature matrix to obtain the predicted equipment power value for future periods, and the predicted equipment heat generation value is calculated based on the power-heat generation mapping model. Based on the supply air temperature optimization model, combined with the predicted equipment power and the cooling demand, the target values ​​of the supply air floor opening ratio and the air valve opening are dynamically calculated. Fuzzy PID control is performed based on the predicted power of the equipment, the target values ​​of the air supply floor opening ratio and the air valve opening, and the reference values ​​of the air supply floor opening ratio and the air valve opening to obtain energy-saving control commands.

2. The multimodal data center air conditioning energy-saving control method according to claim 1, characterized in that, Edge preprocessing is performed on the environmental data and the device power data to obtain a standardized feature matrix, specifically including: Data cleaning is performed on the device power data, temperature data, and relative humidity data. The time data and the power data, temperature data and relative humidity data of the cleaned equipment are subjected to feature engineering and standardization to obtain standardized single-dimensional features. A feature matrix is ​​constructed using the standardized single-dimensional features, resulting in a standardized feature matrix. The standardized feature matrix is ​​represented as follows: ； In the formula, express The standardized feature matrix at time step; express Device power data at any given time express Device power data at any given time; express Temperature data at any given time express Temperature data at any given time; express Relative humidity data at any given time, express Relative humidity data at any given time; Indicates hours, Indicates the day of the week. Indicates the month, and , and All data are time data encoded using sine / cosine encoding.

3. The multimodal data center air conditioning energy-saving control method according to claim 1, characterized in that, The formula for calculating the real-time air enthalpy is as follows: ； In the formula, This indicates the real-time air enthalpy value. This represents real-time temperature data. This represents real-time relative humidity data.

4. The multimodal data center air conditioning energy-saving control method according to claim 1, characterized in that, Offline CFD airflow simulation is performed based on the basic parameters of the computer room, the environmental data, the equipment power data, and the predicted heat generation of the equipment. Combined with the real-time air enthalpy, the baseline values ​​for cooling demand, floor perforation ratio, and damper opening are calculated, specifically including: Offline CFD airflow simulation is performed based on the basic parameters of the computer room, the environmental data, the equipment power data, and the predicted heat generation of the equipment to simulate the airflow distribution before and after the thermal aisle is closed in the computer room, thereby obtaining the air mass flow rate. Calculate the cooling requirements before and after the hot aisle is closed based on the air mass flow rate and the real-time air enthalpy. Based on the cooling demand before and after the hot aisle is closed, the baseline values ​​for the floor opening ratio and damper opening are calculated.

5. The multimodal data center air conditioning energy-saving control method according to claim 4, characterized in that, The formula for calculating the cooling demand is as follows: ； In the formula, This indicates the cooling demand. This indicates the air mass flow rate. This indicates the real-time air enthalpy value at the server's exhaust vent. This indicates the real-time air enthalpy value at the cold aisle inlet.

6. The multimodal data center air conditioning energy-saving control method according to claim 1, characterized in that, The power-heat generation mapping model is expressed as follows: ； In the formula, This represents the predicted heat output of the device. This represents the predicted power value of the device. This represents the model parameters of the power-calorific value mapping model, and The value range is [0.8, 1.0].

7. The multimodal data center air conditioning energy-saving control method according to claim 4, characterized in that, Based on the supply air temperature optimization model, and combining the predicted equipment power and the cooling demand, the target values ​​for the supply air floor opening ratio and damper opening are dynamically calculated, specifically including: Based on the supply air temperature optimization model, the optimal supply air temperature setpoint is calculated according to the predicted power of the equipment, real-time temperature data and air mass flow rate. Based on the optimal supply air temperature setpoint and the cooling capacity requirement, the target values ​​for the air supply floor opening ratio and the air valve opening degree are dynamically calculated.

8. The multimodal data center air conditioning energy-saving control method according to claim 7, characterized in that, The supply air temperature optimization model is expressed as follows: ； In the formula, This represents the optimal supply air temperature setpoint. This represents the predicted power value of the device. This indicates the air mass flow rate. This indicates the specific heat capacity of air at constant pressure. This represents real-time temperature data; The model parameters of the air supply temperature optimization model are represented, and their values ​​range from [0.5, 1.2].

9. The multimodal data center air conditioning energy-saving control method according to claim 1, characterized in that, Based on the predicted power of the equipment, the target values ​​of the air supply floor opening ratio and the damper opening, and the reference values ​​of the air supply floor opening ratio and the damper opening, fuzzy PID control is performed to obtain energy-saving control commands, specifically including: The pressure difference data before and after the cabinet, the temperature difference data of the hot and cold aisles, and the real-time equipment power data are used as real-time feedback quantities. The predicted power value of the equipment is used as the feedforward compensation amount; The target values ​​for the floor perforation rate and damper opening are used as target quantities. Use the baseline values ​​of the air supply floor opening ratio and the air valve opening degree as initial quantities; Fuzzy PID control is performed based on the real-time feedback, the feedforward compensation, the target, and the initial values ​​to obtain energy-saving control commands; The energy-saving control commands include control commands for air conditioning frequency adjustment, air guide plate angle, electric air valve opening degree, and air supply floor opening ratio.

10. The multimodal data center air conditioning energy-saving control method according to claim 4, characterized in that, Also includes: Calculate the rate of decrease in cooling demand based on the cooling demand before and after the hot aisle is closed. Obtain the air conditioning energy consumption before and after the equipment is operated, and calculate the air conditioning energy saving rate accordingly; The power usage efficiency after the equipment is activated is obtained, and the overall energy efficiency is calculated by combining the cooling demand reduction rate and the air conditioning energy saving rate. Determine whether the overall energy efficiency is greater than or equal to the preset energy efficiency threshold. If not, return to start re-running the energy-saving control of the computer room air conditioning. If yes, end the energy-saving control of the computer room air conditioning.

11. The multimodal data center air conditioning energy-saving control method according to claim 10, characterized in that, The formula for calculating the rate of decrease in cooling demand is as follows: ； In the formula, This indicates the rate of decrease in cooling demand. This indicates the cooling demand before the hot aisle is closed. This indicates the cooling demand after the hot aisle is closed; The formula for calculating the air conditioner's energy-saving rate is as follows: ； In the formula, This indicates the energy-saving rate of the air conditioner. This indicates the air conditioning energy consumption after the equipment has been activated. This indicates the air conditioning energy consumption before the equipment was activated; The formula for calculating comprehensive energy efficiency is as follows: ； In the formula, This indicates the overall energy efficiency. This indicates the power efficiency after the device is activated.

12. The multimodal data center air conditioning energy-saving control method according to claim 1, characterized in that, After the equipment is activated, the following is also included: The model parameters of the power-heat generation mapping model and the air supply temperature optimization model are refitted based on real-time equipment operating data. Acquire actual airflow data, and then use the actual airflow data to reverse-correct the boundary conditions and turbulence parameters of the CFD model.

13. The multimodal data center air conditioning energy-saving control method according to claim 1, characterized in that, The environmental data also includes infrared thermal imaging data; The multimodal data center air conditioning energy-saving control method also includes: The air supply direction is dynamically adjusted based on the infrared thermal imaging data.

14. A multimodal energy-saving control system for computer room air conditioning, characterized in that: The energy-saving control method for multimodal computer room air conditioning, as described in any one of claims 1 to 13, includes: The data acquisition module is used to acquire basic parameters of the computer room, as well as environmental data, equipment power data, and equipment operating condition data of the computer room in all dimensions; wherein, the environmental data includes temperature data, relative humidity data, and pressure difference data; An edge preprocessing module is used to perform edge preprocessing on the environmental data and the equipment power data to obtain a standardized feature matrix, and to obtain the real-time air enthalpy values ​​of the cold aisle inlet and server exhaust vent in the computer room. The offline CFD airflow simulation module is used to perform offline CFD airflow simulation based on the basic parameters of the computer room, the environmental data, the equipment power data and the predicted value of equipment heat generation, and to calculate the cooling demand and the benchmark values ​​of the floor opening ratio and the opening degree of the air valve in combination with the real-time air enthalpy value. The online LSTM multi-scale prediction module is used to perform online LSTM multi-scale equipment power prediction using historical equipment power data and the standardized feature matrix to obtain the equipment power prediction value for future periods, and calculate the equipment heat generation prediction value according to the power-heat generation mapping model. The air supply temperature optimization module is used to dynamically calculate the target values ​​of the air supply floor opening ratio and the air valve opening degree based on the air supply temperature optimization model, combined with the predicted equipment power value and the cooling demand. The fuzzy PID control module is used to perform fuzzy PID control based on the predicted power value of the equipment, the target values ​​of the air supply floor opening ratio and the air valve opening, and the reference values ​​of the air supply floor opening ratio and the air valve opening, to obtain energy-saving control commands.

15. A multimodal energy-saving control device for computer room air conditioning, characterized in that: The system includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the multimodal data center air conditioning energy-saving control method as described in any one of claims 1 to 13.

Images