Data center anti-freezing energy-saving control device and control method
By combining multi-dimensional sensing monitoring and digital twin simulation modules, the cooling tower flow rate is dynamically adjusted, solving the problem of high icing risk in data center backup cooling towers during winter, achieving a balance between energy saving and antifreeze, and improving the system's adaptability and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-03-17
AI Technical Summary
Existing data center backup cooling towers are at high risk of freezing in low-temperature winter environments. Traditional antifreeze solutions are energy-intensive, have rigid control logic, and lack predictive capabilities, making it difficult to balance antifreeze safety with energy efficiency.
The system combines a multi-dimensional sensing and monitoring module with a digital twin simulation module. The intelligent control module predicts the internal temperature field distribution of the standby cooling tower and dynamically adjusts the hot water flow of the main cooling tower and the standby cooling tower. The linear regression model and NSGA-II algorithm are used to optimize temperature, flow rate and energy consumption, and adaptive PID and DQN reinforcement learning strategies are combined for control.
It achieves proactive prediction and dynamic adaptation, integrates multi-objective optimization, improves control accuracy and robustness, adapts to different climate regions and cooling tower specifications, reduces energy consumption and ensures stable system operation.
Smart Images
Figure CN121677463A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data center cooling technology, and in particular to a data center anti-freezing and energy-saving control device and control method. Background Technology
[0002] As the computing power density of data centers continues to increase, the stable operation of the cooling system directly determines the availability of the data center. As a key redundant device in the cooling system, the backup cooling tower faces a severe risk of freezing in the low temperature environment of winter. If the cooling water in the water collection pan and connecting pipes freezes below 0°C, it will cause equipment damage, pipe rupture, or even paralysis of the cooling system, resulting in significant economic losses.
[0003] Current mainstream backup cooling tower antifreeze solutions in the industry have significant drawbacks, making it difficult to balance antifreeze safety with energy efficiency: 1. Serious energy waste: Traditional solutions use electric heating for insulation or continuous low flow operation mode. Regardless of the ambient temperature and equipment status, they maintain a fixed heating power or flow rate, resulting in winter anti-freezing energy consumption accounting for 8%-15% of the total energy consumption of the data center, with extremely low energy utilization. 2. Rigid control logic: Using a fixed antifreeze temperature threshold (e.g., 5℃) without considering regional climate differences, changes in the main cooling tower's operating load, and real-time fluctuations in ambient temperature and humidity can easily lead to problems such as overheating (wasting energy) or a threshold that is too low (risk of freezing). 3. Insufficient predictive capability: Relying on real-time temperature sensor feedback, it belongs to passive response control and lacks the ability to predict the internal temperature field of the backup cooling tower. When the ambient temperature drops sharply, the response lag can easily lead to icing. Summary of the Invention
[0004] The purpose of this invention is to provide a data center anti-freezing and energy-saving control device and method to solve the above-mentioned technical problems.
[0005] To achieve the above objectives, the present invention provides a data center anti-freezing and energy-saving control device, including a water supply connection pipe and a water return connection pipe connecting the main cooling tower and the backup cooling tower, a multi-dimensional sensing and monitoring module, an intelligent control module, and a digital twin simulation module. The water supply connection pipe and the water return connection pipe are respectively equipped with a water supply electric regulating valve and a water return electric regulating valve. The multidimensional sensing and monitoring module, the electric regulating valve for water supply, the electric regulating valve for water return, and the digital twin simulation module are all connected to the intelligent control module. This enables the intelligent control module to transmit the multidimensional data collected by the multidimensional sensing and monitoring module to the digital twin simulation module for predicting the internal temperature field distribution of the standby cooling tower. Based on the prediction results, the hot water flow of the main cooling tower and the standby cooling tower is controlled to prevent the standby cooling tower from freezing.
[0006] A control method for a data center anti-freezing and energy-saving control device includes the following steps: S1. Based on historical winter meteorological data of the area where the data center is located, the structural parameters of the backup cooling tower and the operating parameters of the main cooling tower, and considering the constraints of safe operation, the initial antifreeze target temperature, flow balance threshold and initial PID parameters are calibrated through the heat balance equation. S2. The real-time data collected by the multi-dimensional sensing and monitoring module and the predicted data of the digital twin model are integrated, and the data is cleaned by the 3σ criterion and weighted by the AHP-entropy weight method to generate the working condition dataset. S3. Based on the working condition dataset output in step S2, the initial antifreeze target temperature output in step S1 is dynamically corrected through a linear regression model to obtain the dynamic target temperature. Combined with historical operating results, the NSGA-II algorithm is used to determine the multi-objective optimization weights for temperature stability, flow balance, and minimum energy consumption. S4. Guided by the dynamic target temperature and multi-objective optimization weights output in step S3, the opening control signals of the water supply electric regulating valve and the return water electric regulating valve are generated and executed through adaptive PID parameter adjustment, DQN multi-objective optimization and flow balance constraint calculation.
[0007] Therefore, the present invention, employing the above-mentioned data center anti-freezing and energy-saving control device and method, has the following beneficial effects: 1. Proactive prediction + dynamic adaptation: Integrating digital twin simulation and multi-dimensional sensor data, it predicts the temperature field distribution of the backup cooling tower in advance, breaking the traditional passive response mode; it dynamically corrects the antifreeze target temperature through a linear regression model to adapt to changes in climate, load and other operating conditions. 2. Multi-objective collaborative optimization: The NSGA-II algorithm is used to quantify the weight allocation of temperature stability, flow balance and minimum energy consumption, so as to overcome the limitations of single-objective control and achieve the optimal balance of the three core requirements. 3. Combining control precision and robustness: The innovative adaptive PID+DQN reinforcement learning composite control strategy ensures control stability under normal operating conditions while also meeting the dynamic adjustment requirements of sudden operating conditions (such as sudden temperature drop or sudden flow change). 4. High versatility: Based on regional historical meteorological data and equipment structural parameters, the initial parameters are calibrated, which can be adapted to different climate regions (severe cold / mild winter) and different specifications of cooling towers, without the need for extensive customization.
[0008] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0009] Figure 1 This is a connection block diagram of a data center anti-freezing and energy-saving control device according to the present invention. Detailed Implementation
[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.
[0011] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.
[0012] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0013] like Figure 1 As shown, a data center anti-freezing and energy-saving control device includes a water supply connection pipe and a return water connection pipe connecting the main cooling tower and the standby cooling tower, a multi-dimensional sensing and monitoring module, an intelligent control module, and a digital twin simulation module. The water supply connection pipe and the return water connection pipe are respectively equipped with an electric regulating valve for water supply and an electric regulating valve for return water. The multi-dimensional sensing and monitoring module, the electric regulating valves for water supply and return water, and the digital twin simulation module are all connected to the intelligent control module. This enables the intelligent control module to transmit the multi-dimensional data collected by the multi-dimensional sensing and monitoring module to the digital twin simulation module for predicting the internal temperature field distribution of the standby cooling tower. Based on the prediction results, the device controls the hot water flow rate of the main cooling tower and the standby cooling tower to prevent the standby cooling tower from freezing.
[0014] The multi-dimensional sensing and monitoring module includes: a water collection pan temperature sensor for real-time monitoring of the standby cooling tower water collection pan temperature; a water supply flow sensor for real-time monitoring of the cooling water flow in the water supply connection pipe; a return water flow sensor for real-time monitoring of the cooling water flow in the return water connection pipe; an ambient temperature and humidity sensor for real-time monitoring of outdoor temperature and humidity; and a water supply temperature sensor for real-time monitoring of the main cooling tower water supply temperature.
[0015] It should be noted that the above electronic components are all mature products on the market. This embodiment only requires purchasing them and connecting them according to the instruction manual. No changes have been made to their circuit connection structure, so they will not be described in detail here.
[0016] A control method for a data center anti-freezing and energy-saving control device includes the following steps: S1. Based on historical winter meteorological data of the area where the data center is located, the structural parameters of the backup cooling tower and the operating parameters of the main cooling tower, and considering the constraints of safe operation, the initial antifreeze target temperature, flow balance threshold and initial PID parameters are calibrated through the heat balance equation. The historical winter weather data for the area where the data center is located, as described in step S1, includes the lowest temperature. and the cumulative duration below 0°C ; The structural parameters of the standby cooling tower include the volume of the water collection tray. Water supply connection pipe diameter , return water connecting pipe diameter and maximum allowed traffic ; Main cooling tower operating parameters include water supply temperature range and main cooling tower load rate ; The safety operation constraint is the minimum allowable temperature of the standby cooling tower sump. And satisfy ; The initial antifreeze target temperature expression is as follows: ; In the formula, Indicates the safety factor; Traffic balancing threshold The expression is as follows: ; In the formula, This indicates the allowable percentage of flow deviation, used to ensure that the supply and return water flow rates are basically consistent and to maintain the hydraulic balance of the system; The initial PID parameter expression is as follows: proportionality coefficient : ; Integral coefficient : ; Differential coefficients : ; In the formula, This indicates the specific heat capacity of the cooling water; Indicates the density of cooling water; and These represent the integral time constant and the derivative time constant, respectively.
[0017] S2. The real-time data collected by the multi-dimensional sensing and monitoring module and the predicted data of the digital twin model are integrated, and the data is cleaned by the 3σ criterion and weighted by the AHP-entropy weight method to generate the working condition dataset. In step S2, the temperature of the standby cooling tower water collection basin is collected synchronously. Cooling water flow rate in the water supply connection pipeline Cooling water flow rate in the return water connection pipe Outdoor temperature and humidity , and the main cooling tower water supply temperature The raw data was obtained, and the moving average filtering method was used to eliminate random noise in the raw data. The 3σ criterion was used to remove outliers. Finally, the AHP-entropy weighting method was used to weight and fuse the multi-sensor data to generate the working condition dataset. , These represent the pre-treated standby cooling tower water collection pan temperature, the cooling water flow rate in the supply water connection pipe, the cooling water flow rate in the return water connection pipe, the outdoor temperature and humidity, and the main cooling tower supply water temperature, respectively.
[0018] S3. Based on the working condition dataset output in step S2, the initial antifreeze target temperature output in step S1 is dynamically corrected through a linear regression model to obtain the dynamic target temperature. Combined with historical operating results, the NSGA-II algorithm is used to determine the multi-objective optimization weights for temperature stability, flow balance, and minimum energy consumption. Step S3 specifically includes the following steps: S31. Filter out real-time operating condition features that are strongly correlated with the antifreeze target temperature; S311. From historical operating data, select the operating condition where the standby cooling tower has no icing, the flow rate is balanced, and the energy consumption increase of the main cooling tower is minimal. Obtain the temperature of the standby cooling tower's water collection basin under this condition and regard it as the historical optimal target temperature. The flow balancing expression is as follows: ; S312, Set the initial feature set including outdoor temperature and humidity. Main cooling tower water supply temperature Main cooling tower load rate Temperature change rate of standby cooling tower water collection basin and average flow rate of supply and return water Among them, the temperature change rate of the standby cooling tower water collection basin The calculation formula is as follows: ; In the formula, and They represent Time and The temperature of the standby cooling tower's water collection basin at any given time; Indicates the sampling period; The formula for calculating the average flow rate of supply and return water is as follows: ; In the formula, and They represent The flow rate of cooling water in the supply water connection pipe and the flow rate of cooling water in the return water connection pipe are constantly monitored. S313. Calculate the Pearson correlation coefficient between the features in the initial feature set and the historical best target temperature. and retain The features constitute the feature matrix. , Indicates the filtered first One feature; Indicates the filtering threshold; S314. Standardize the feature matrix to obtain the standardized feature matrix. and standardize the feature matrix The samples are divided into a training set and a validation set; S32. Learn the mapping relationship between historical operating conditions and historical optimal target temperature through a linear regression model, and dynamically correct the initial target temperature in step S1. S321. Construct the following linear regression model: ; In the formula, Represents the intercept term; Indicates the first Standardized features The regression coefficients; Represents the random error term; S322, Using the training set as input, and the historical best target temperature Given the label vectors, train a linear regression model using the following loss function: ; In the formula, Indicates the loss value; Indicates the maximum number of iterations; Indicates the iteration count index; The regression coefficients are solved using the following least squares method: ; In the formula, This represents the estimated value of the regression coefficients; Represents the normalized feature matrix Transpose of; Represents a label vector; S323, Using the validation set The linear regression model was validated using MAE as the evaluation index, and the validated linear regression model was used to predict the dynamic target temperature. : ; ; In the formula, Indicates the threshold for waste heat; S33. Construct the following multi-objective optimization objective function. : ; in, ; ; ; Constraints: ; In the formula, , and These represent the weights of the temperature stability target, the flow balance target, and the energy minimization target, respectively. , and Let these represent the objective functions for temperature stability, flow balance, and energy minimization, respectively. This represents the number of sampling points within the optimization period; This indicates the real-time energy consumption when the main cooling tower transfers waste heat to the standby cooling tower. This represents the baseline energy consumption of the main cooling tower when there is no waste heat transfer. This indicates the maximum allowable temperature of the standby cooling tower's water collection basin; and These represent the opening degrees of the water supply electric valve and the return water electric valve, respectively. S34. Use the NSGA-II algorithm to search for Pareto optimal solutions in the feasible solution space and determine... , and Optimal weights: ; In the formula, This represents the optimal weight, and , These represent the weights of the optimized temperature stability target, flow balance target, and energy minimization target, respectively. This represents the final Pareto optimal solution set; It indicates the degree of crowding of an individual.
[0019] S4. Guided by the dynamic target temperature and multi-objective optimization weights output in step S3, the opening control signals of the water supply electric regulating valve and the return water electric regulating valve are generated and executed through adaptive PID parameter adjustment, DQN multi-objective optimization and flow balance constraint calculation.
[0020] Step S4 specifically includes the following steps: S41, Based on optimal weights Adjusting PID parameters: Adjusted scaling factor : ; Adjusted integral coefficients : ; Adjusted differential coefficients : ; In the formula, , and These represent the response coefficients for the temperature stability target, the flow balance target, and the energy minimization target, respectively. S42. Output the initial opening degree of the water supply electric valve based on the PID algorithm after parameter adjustment. and the opening degree of the initial return water electric valve : ; In the formula, Indicates temperature deviation; This represents the differential term of the temperature deviation, and , and They represent and Temperature deviation at any given time; This represents the integral term of the temperature deviation, and , Indicates the cumulative number of samples; S43. Utilize the DQN reinforcement learning algorithm with optimal weights. Guided by this, the initial water supply electric valve opening output in step S42 is corrected. and the opening degree of the initial return water electric valve : ; ; In the formula, and These represent the corrected opening degrees of the water supply electric valve and the return water electric valve, respectively. and These represent the adjustment amounts of the water supply electric valve opening and the return water electric valve opening, respectively, output by the DQN reinforcement learning algorithm. S44. Determine the corrected result. and Does it meet the requirements? If so, return to step S43 until the condition is met, and output the corrected result. and ; S45, Flow balance constraint verification and secondary correction; S451, Flow deviation verification: like If the verification is successful, output the result. , , and These represent the opening degrees of the water supply electric valve and the return water electric valve after the second correction, respectively. Otherwise, the following correction strategy will be implemented: ; ; in, ; In the formula, Indicates the current flow deviation; This represents the correction factor, and ; Indicates the safety threshold; S452, Secondary Verification: Determine the flow deviation after secondary correction. Does it meet the requirements? If so, the verification is considered successful, and the result is output. and Otherwise, send a pipeline pressure warning signal; Among them, the flow deviation after secondary correction The calculation formula is as follows: ; In the formula, This indicates that when the water supply electric valve is opened to the specified degree... Water supply flow rate at that time; This indicates that when the return water electric valve opening degree is executed... The return water flow rate at that time; S46, will and The signal is converted into a standard electrical signal that the electric valve can recognize, which drives the electric valve to perform the action and monitors the opening error. If the opening error is within the set threshold range, the execution is considered successful; otherwise, a valve fault alarm signal is sent and the previous opening is maintained.
[0021] Simulation experiment: Verify the superiority of the present invention in terms of antifreeze safety, energy consumption control, flow balance, and response speed.
[0022] Experimental parameter settings Historical winter meteorological data for the area where the digital center is located: , ; Backup cooling tower parameters: water collection tray volume Water supply pipe diameter return water pipe diameter , ; Main cooling tower parameters: water supply temperature range (32℃, 38℃), load rate ; Safety constraints: , , ; Algorithm parameters: sampling period Number of iterations of the NSGA-II algorithm The DQN reinforcement learning reward function is based on a multi-objective weight design.
[0023] Comparison Models: Comparison Group 1: Traditional PID control (fixed antifreeze threshold of 5℃, no flow balance constraint); Comparison Group 2: Fixed threshold control (industry-common 5℃ threshold, electric heating insulation); Comparison Group 3: Single objective optimization control (only optimizes the antifreeze objective, ignoring energy consumption and flow balance); Experimental Group: Digital twin + linear regression + NSGA-II + PID + DQN composite control of this invention.
[0024] Experimental scenario Simulates typical winter operating conditions: ambient temperature fluctuates between -15℃ and 5℃ (including a sudden drop scenario: temperature drops from 3℃ to -8℃ within 1 hour), and the main cooling tower load rate dynamically changes from 60% to 80% to 100%.
[0025] Experimental results
[0026] Results Analysis 1. Freeze protection safety: There was no risk of freezing in both control group 2 (fixed threshold) and the experimental group. However, control group 2 experienced a surge in energy consumption due to its higher threshold (5℃). The experimental group, through a dynamic target temperature (0.5℃), ensured freeze protection safety while avoiding overheating. 2. Energy-saving performance: The energy consumption of the experimental group was reduced by 46.7% compared with the control group 2, 25% compared with the control group 1, and 20% compared with the control group 3. The core reason is that the multi-objective optimization weight allocation effectively reduced the energy consumption of ineffective heating and redundant flow. 3. Flow balance: The flow deviation rate of the experimental group was only 2.8%, which was much lower than that of the control group (7.1%-10.5%), demonstrating the effectiveness of the flow balance constraint and the secondary correction strategy, and avoiding hydraulic imbalance. 4. Response speed: In the scenario of sudden temperature drop, the response time of the experimental group (1.8s) is much faster than that of the traditional solution (8.5s-12.3s). The predictive ability of digital twin and the dynamic correction ability of DQN greatly improve the robustness of the system. 5. System stability: The main cooling tower water supply temperature fluctuation (±0.4℃) in the experimental group was minimal, ensuring the stable operation of the data center cooling system and avoiding the impact of load changes and temperature fluctuations on computing power output.
[0027] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A data center freeze protection and energy saving control apparatus, characterized by: The application comprises a water supply communication pipeline and a backwater communication pipeline connected between a main cooling tower and a standby cooling tower, a multidimensional sensing monitoring module, an intelligent control module, and a digital twin simulation module, wherein the water supply communication pipeline and the backwater communication pipeline are respectively provided with a water supply electric regulating valve and a backwater electric regulating valve. The multidimensional sensing monitoring module, the water supply electric regulating valve, the backwater electric regulating valve, and the digital twin simulation module are connected with the intelligent control module, so that the multidimensional data collected by the multidimensional sensing monitoring module are transmitted to the digital twin simulation module for standby cooling tower internal temperature field distribution prediction through the intelligent control module, and the hot water flow of the main cooling tower and the standby cooling tower is controlled according to the prediction result to avoid freezing of the standby cooling tower.
2. The data center freeze protection and energy saving control device of claim 1, wherein: The multidimensional sensing monitoring module comprises: a water collecting disc temperature sensor for monitoring the standby cooling tower water collecting disc temperature in real time; a water supply flow sensor for monitoring the cooling water flow in the water supply communication pipeline in real time; a backwater flow sensor for monitoring the cooling water flow in the backwater communication pipeline in real time; an environmental temperature and humidity sensor for monitoring the outdoor temperature and humidity in real time; a water supply temperature sensor for monitoring the main cooling tower water supply temperature in real time.
3. The control method of the data center freeze protection and energy saving control device according to claim 1 or 2, characterized in that: The application comprises the following steps: S1, based on the historical winter meteorological data of the region where the data center is located, the standby cooling tower structure parameters and the main cooling tower operation parameters, and considering the safety operation constraint conditions, the initial anti-freezing target temperature, the flow balance threshold and the initial PID parameters are calibrated through the heat balance equation; S2, the real-time collected data of the multidimensional sensing monitoring module and the prediction data of the digital twin model are fused, the data are cleaned through the 3σ criterion and weighted through the AHP-entropy weight method to generate a working condition data set; S3, based on the working condition data set output in step S2, the initial anti-freezing target temperature output in step S1 is dynamically corrected through a linear regression model to obtain a dynamic target temperature, and the multi-objective optimization weight with stable temperature, balanced flow and minimum energy consumption is determined by combining the historical operation results and using the NSGA-II algorithm; S4, taking the dynamic target temperature and the multi-objective optimization weight output in step S3 as the guide, the opening control signals of the water supply electric regulating valve and the backwater electric regulating valve are generated and executed through adaptive PID parameter adjustment, DQN multi-objective optimization and flow balance constraint calculation.
4. The control method of the data center freeze protection and energy saving control device according to claim 3, characterized in that: The historical winter weather data of the area where the data center described in step S1 is located includes the minimum temperature and the cumulative duration below 0°C ; The standby cooling tower structure parameters include a water collecting pan volume , a water supply communication pipe diameter , a backwater communication pipe diameter , and a maximum allowable flow rate ; Main cooling tower operating parameters include the supply water temperature range and the main cooling tower load rate ; The safe operation constraint condition is the minimum allowable temperature of the reserve cooling tower water collecting disc , and satisfies ; The initial anti-freezing target temperature expression is as follows: ； In the formula, represents the safety factor; Flow balance threshold The expression is as follows: ； In the formula, represents the flow deviation allowance ratio; The initial PID parameter expression is as follows: proportionality factor : ; Integration coefficient : ; derivative coefficient : ; wherein represents the specific heat capacity of the cooling water; represents the density of the cooling water; and respectively represent the integral time constant and the derivative time constant.
5. The control method of the data center freeze protection and energy saving control device according to claim 4, characterized in that: In step S2, the standby cooling tower sump temperature is synchronously collected , the cooling water flow rate in the water supply communication pipeline , the cooling water flow rate in the return water communication pipeline , the outdoor temperature and humidity , , and the main cooling tower water supply temperature are obtained as original data, random noise in the original data is eliminated by using a moving average filtering method, abnormal values are removed by using a 3σ criterion, and finally, the working condition data set is generated by using an AHP-entropy weight method to weight and fuse multiple sensing data , respectively represent the standby cooling tower sump temperature, the cooling water flow rate in the water supply communication pipeline, the cooling water flow rate in the return water communication pipeline, the outdoor temperature and humidity, and the main cooling tower water supply temperature after preprocessing.
6. The control method of the data center freeze protection and energy saving control device according to claim 5, characterized in that: Step S3 specifically comprises the following steps: S31, screening out the real-time working condition characteristics strongly related to the anti-freezing target temperature; S311, from the historical operation data, the standby cooling tower is screened out without icing, flow balance, the minimum energy consumption increment of the main cooling tower, and the water collecting plate temperature of the standby cooling tower under the working condition is obtained, which is regarded as the historical optimal target temperature ; wherein, the flow balance expression is as follows: ; S312, setting the preliminary feature set including outdoor temperature and humidity , main cooling tower water supply temperature , main cooling tower load rate , standby cooling tower sump temperature change rate and average flow of supply and return water ; wherein the standby cooling tower sump temperature change rate The calculation formula is as follows: ； wherein and respectively represent the time of day and the standby cooling tower sump temperature at the time of day; denotes the sampling period; The average water supply and backwater flow calculation formula is as follows: ； In the formula, and respectively represent the flow rate of cooling water in the water supply communication pipe and the flow rate of cooling water in the backwater communication pipe at the time S313、Calculate the Pearson correlation coefficient between the features in the preliminary feature set and the historical optimal target temperature , and reserve the features to form a feature matrix , represents the first feature after screening; represents the screening threshold; S314, normalizing the feature matrix to obtain a normalized feature matrix and the normalized feature matrix The samples are divided into a training set and a validation set; S32, the mapping relationship between the historical working conditions and the historical optimal target temperature is learned through a linear regression model to dynamically correct the initial target temperature of step S1; S321, the following linear regression model is constructed: ； wherein represents an intercept term; represents the regression coefficient of the th standardized feature ; and represents a random error term. S322, Using the training set as input, and the historical best target temperature Given the label vectors, train a linear regression model using the following loss function: ； In the formula, represents a loss value; represents a maximum number of iterations; represents an iteration number index; The least square method is used to solve the regression coefficients as follows: ； In the formula, This represents the estimated value of the regression coefficients; Represents the normalized feature matrix transpose; Represents a label vector; S323, using the validation set The linear regression model is verified with MAE as the evaluation index, and the linear regression model that passes the verification is used to predict the dynamic target temperature : ； ; In the formula, represents a waste heat waste threshold value; S33, construct the following multi-objective optimization objective function : ； wherein, ； ； ； Constraints: ; In the formula, , and respectively represent the weights of the temperature stability target, the flow balance target and the minimum energy consumption target; , and respectively represent the temperature stability target function, the flow balance target function and the minimum energy consumption target function; represents the number of sampling points in the optimization period; represents the real-time energy consumption when the main cooling tower delivers residual heat to the standby cooling tower; represents the reference energy consumption when the main cooling tower has no residual heat delivery; represents the maximum allowable temperature of the standby cooling tower sump; and respectively represent the opening degrees of the water supply electric valve and the return water electric valve; S34, search for the Pareto optimal solution in the feasible solution space by using the NSGA-II algorithm, determine 、 and optimal weight: ； In the formula, represents the optimal weight, and , respectively represent the weights of the optimized temperature stability target, the flow balance target and the minimum energy consumption target; represents the final Pareto optimal solution set; represents the crowding degree of the individual.
7. The control method of the data center freeze protection and energy saving control device according to claim 6, characterized in that: Step S4 specifically comprises the following steps: S41、based on the optimal weight Adjusting PID parameters: adjusted scaling factor : ; adjusted integration factor : ; adjusted differential coefficient : ; wherein, , and respectively represent the response coefficients of the temperature stabilization target, the flow balance target, and the minimum energy consumption target. S42, output initial water supply electric valve opening degree based on parameter adjusted PID algorithm and initial return water electric valve opening degree : ； wherein represents a temperature deviation; represents a temperature deviation differential term, and , and respectively represent and a temperature deviation at a time t; represents a temperature deviation integral term, and , represents a cumulative sampling number; S43, using a DQN reinforcement learning algorithm to optimize the weights S42, the initial supply valve opening degree and the initial return valve opening degree are corrected based on the target water supply temperature and the target water return temperature and the initial return valve opening degree : ； ； In the formula, and respectively represent the corrected water supply electric valve opening and the return water electric valve opening; and respectively represent the adjustment amount of the water supply electric valve opening and the return water electric valve opening output by the DQN reinforcement learning algorithm; S44, judging whether the corrected and satisfy , if yes, returning to step S43 until the conditions are satisfied, outputting the corrected and ; S45, flow balance constraint verification and secondary correction; S451, flow deviation verification: If , then determine that the check is qualified, output , , and respectively represent the secondary modified water supply electric valve opening and the return water electric valve opening; Otherwise, the following correction strategy is executed: ； ； wherein, ； wherein represents the current flow deviation; represents a correction factor, and ; represents a safety threshold; S452, second check: judging the flow deviation after the second correction whether the condition is satisfied , if yes, determining that the check is qualified, and outputting and ; otherwise, sending a pipeline pressure warning signal; wherein the flow deviation after the second correction The calculation formula is as follows: ； In the formula, This indicates that when the water supply electric valve is opened to the specified degree... Water supply flow rate at that time; This indicates that when the return water electric valve opening degree is executed... The return water flow rate at that time; S46, to and Converts to the standard electrical signal that the electric valve can recognize, drives the electric valve to execute the action, and monitors the opening error. If the opening error is within the set threshold range, it is determined that the execution is successful, otherwise a valve failure alarm signal is sent, and the last round opening is maintained.