Machine room water-cooling air conditioner adaptive control method based on big data analysis

Abstract

The invention discloses a machine room water-cooled air conditioner adaptive control method based on big data analysis, which relates to the technical field of adaptive regulation and control, is used for improving the problem of overlong response time caused by fixed thermal inertia, and comprises the following steps of: performing network deployment on distributed optical fiber sensing, screening environmental data in a water-cooled air conditioner adaptive system by utilizing a sliding time window algorithm, and determining the response time of the water-cooled air conditioner. Noise is suppressed, a thermal inertia compensation mechanism is entered, an LSTM neural network is constructed based on initial system thermal inertia to predict thermal disturbance of a plurality of subsequent control periods, and a space-time alignment compensation matrix is introduced to suppress harmonic oscillation in the thermal inertia compensation mechanism; the predicted thermal disturbance and the chilled water flow in the air conditioner are integrated, a double-closed-loop correction method is used for generating thermal compensation correction, gain switching is conducted on the double-closed-loop correction method, and synchronous alignment correction is conducted on asynchronously-collected data packets through a time sequence alignment engine; and setting a bending window and carrying out characterization fusion to judge an environment data acquisition point and an abnormal acquisition response rate thereof.

Description

Technical Field

[0001] The present invention relates to the field of adaptive control technology, and more specifically, to a method for adaptively controlling a water-cooled air conditioner in a computer room based on big data analysis. Background Art

[0002] When the load fluctuation changes significantly, the traditional adaptive control method of water-cooled air conditioners in computer rooms takes a long time to learn and adapt, and cannot respond quickly.

[0003] The prior art has the following deficiencies: The existing system has large thermal inertia. Even if the control system receives a signal of load change, the system's cooling capacity and reaction speed need time to adapt to the new conditions. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an adaptive control method for water-cooled air conditioners in computer rooms based on big data analysis. By introducing an adaptive control algorithm and a thermal inertia compensation mechanism, the thermal inertia of the water-cooled air conditioning system is recorded and screened in real time to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: The adaptive control method of water-cooled air conditioner in a computer room based on big data analysis includes the following steps: Step S1: Deploy the distributed optical fiber sensor network, use the sliding time window algorithm to filter the environmental data in the water-cooled air conditioning adaptive system, enable the IIR digital filter to suppress the noise and enter the thermal inertia compensation mechanism; Step S2: Based on the initial system thermal inertia, an LSTM neural network is constructed to predict the thermal disturbances of subsequent multiple control cycles, and a spatiotemporal alignment compensation matrix is ​​introduced to suppress harmonic oscillations in the thermal inertia compensation mechanism; Step S3: using a double closed-loop correction method to generate a thermal compensation correction value based on the predicted thermal disturbance and the chilled water flow in the air conditioner, detecting the environmental load, and setting a variable structure sliding film controller to perform gain switching on the double closed-loop correction method; Step S4: The improved DTW algorithm is applied through the timing alignment engine to perform synchronous alignment correction on the asynchronously collected data packets, set the bending window and perform feature fusion to determine the environmental data collection points and their abnormal collection response rates.

[0006] In a preferred embodiment, in step S1, when the distributed optical fiber sensor is deployed in a network, on the surface of the main pipe and branch pipe of the chilled water loop, each group of sensor nodes is designed to include 4 FBG sensors, and are arranged in a 120° annular distribution; The environmental data is acquired at a sampling rate of 5 Hz and transmitted to the central controller, where the environmental data is the temperature change rate of the cold storage water tank.

[0007] In a preferred embodiment, in step S1, dynamic data screening is performed according to environmental data using a sliding time window algorithm as follows: Preprocessing: Initialize the window length and step size, and mark the window length as T_win; Generate parameter sequence: Generate parameter sequence after de-dimensionalizing each environmental data ; It represents the data sequence within the sampling time window, that is, the standard deviation rate of the measurement data collected within a specific time window T_win ,in, is the standard deviation of the parameter sequence, is the parameter sequence mean; the kurtosis coefficient is ; Among them, kurtosis is used to measure the steepness of data distribution. Generally speaking, when the kurtosis coefficient is approximately equal to 3, the data follows a normal distribution. When the kurtosis coefficient is greater than 3, the data is concentrated near the mean. When the richness coefficient is less than 3, the data is flatly distributed. In the formula, For each data point in the sequence, is the mean of the series, is the standard deviation of the parameter sequence, n is the total number of data points; the calculation formula for the flow rate change rate is: ,in, is the flow rate at the current time t, For the previous moment of traffic, Sampling time interval.

[0008] In a preferred embodiment, in step S2, an LSTM neural network is constructed based on the initial system thermal inertia to predict thermal disturbances of subsequent multiple control cycles. The specific steps are as follows: Construct an LSTM network with multiple hidden units, and the input layer receives the system thermal inertia at multiple time points, which are marked as: ; in, For past load forecast data, is the past return water temperature, The data of heat exchange rate changes in the past; Hidden layer: multiple LSTM units are used to learn the thermal inertia characteristics of the system; The output layer predicts the thermal disturbance of the next n control cycles, that is, ; Set the initial learning rate and define the loss function as: ; In the formula, MAE is the average absolute difference between the predicted value and the true value; Among them, the calculation formula of MSLE mean square logarithmic error is: ; Among them, MSLE is used to reduce the impact of extreme errors; In the formula, is a regularization term used to control the complexity of the model.

[0009] In a preferred embodiment, in step S2, the compensation coefficients are defined in the spatiotemporal alignment compensation matrix Calculation rules: when When the time deviation is less than half of the thermal inertia time constant, the calculation rule of the compensation coefficient is: ; when When the time deviation is greater than half of the thermal inertia time constant, the calculation rule of the compensation coefficient is: ; in, In order to track the error, the above calculation rules are used to suppress the harmonic oscillation in the compensation process.

[0010] In a preferred embodiment, in step S3, the specific steps of using the double closed-loop correction method to generate the thermal compensation correction amount are as follows: Define the state vector: , where τ is the thermal inertia time constant, Ceq is the equivalent heat capacity, and Qacc is the accumulated thermal deviation; where τ reflects the system's response speed to temperature changes, Ceq measures the system's heat storage capacity and directly affects the thermal compensation calculation, and Qacc records the system's heat loss or accumulation to adjust the compensation strategy; Adaptive adjustment formula of the observation matrix: , in, When the thermal compensation amount changes, 0.8 is the preset fixed weight to stabilize the influence of the observed data. is the time decay factor; Set the trigger rule: If the residual covariance satisfies the following inequality, that is, when hour; The system triggers the parameter update mechanism, the gain matrix The calculation formula is as follows: , where the process noise covariance is set as: , the observation noise covariance is set as: ,in, = , 0.02, 0.01, and 0.05 are the disturbance of the preset thermal inertia time constant, the error of the equivalent heat capacity, and the correction of the accumulated thermal deviation, respectively.

[0011] In a preferred embodiment, in step S4, the asynchronously collected data packets are synchronously aligned and corrected, and a long short-term memory (LSTM) neural network containing 128 hidden units is constructed to predict the dynamic characteristics of thermal inertia in the chilled water system.

[0012] The asynchronously collected data packets include load forecast, return water temperature, supply water temperature, compressor power, condenser temperature difference, evaporator heat exchange efficiency, ambient temperature, cooling tower fan speed, chilled water flow and pump pressure, which are marked as , , , , , , , , , , all asynchronously collected data packets are regarded as ten-dimensional time series data; Define input features and define the ten-dimensional time series data as input features; Network training and optimization: The initial learning rate is set to 0.001, and the learning rate decay rate is dynamically adjusted to 0.95 every 10 iterations; the loss function uses the mean square error marked as MSE: ; The amount of training data is preset to 100,000 time steps, and data augmentation is performed according to the window size; Perform cross-validation: Use K-fold cross-validation to determine the environmental data collection points in the preset time step.

[0013] In a preferred embodiment, in step S4, the curved window is set by the finite difference method and the characteristic fusion is performed to determine the environmental data collection point and its abnormal collection response rate as follows: Calculate the thermal inertia change of the chilled water pipeline and the compensation formula is as follows: ;in, is the specific heat capacity, is the fluid mass flow rate, is the thermal conductivity of the tube wall, is the heat transfer area, is the temperature difference, is the infinitesimal length of the pipeline; The abnormal acquisition response rate is calculated using the power load field compensation method.

[0014] In a preferred embodiment, in step S4, the specific steps of calculating the abnormal acquisition response rate using the power load field compensation method are as follows: The transient fluctuation of compressor electric power is predicted by support vector regression model, and the compensation formula is as follows: ; where k is the adaptive adjustment factor, is the load power of the compressor after compensation, is the currently measured compressor power, is the change rate of the supply and return water temperature difference, and the change rate of the supply and return water temperature difference is used as the abnormal collection response rate.

[0015] Technical effects and advantages of the adaptive control method for water-cooled air conditioners in computer rooms based on big data analysis of the present invention: The invention deploys distributed optical fiber sensing in a network, uses a sliding time window algorithm to filter environmental data in a water-cooled air-conditioning adaptive system, enables an IIR digital filter to suppress noise and enter a thermal inertia compensation mechanism, constructs an LSTM neural network based on the initial system thermal inertia to predict thermal disturbances of subsequent multiple control cycles, introduces a time-space alignment compensation matrix to suppress harmonic oscillations in the thermal inertia compensation mechanism, improves data transmission efficiency, uses a double closed-loop correction method to generate a thermal compensation correction amount based on the predicted thermal disturbance and the chilled water flow in the air-conditioning, detects environmental loads, sets a variable structure sliding film controller to perform gain switching on the double closed-loop correction method, uses a timing alignment engine to perform synchronous alignment correction on asynchronously collected data packets using an improved DTW algorithm, sets a curved window and performs characterization fusion to judge environmental data collection points and their abnormal collection response rates, realizes a thermal inertia compensation closed loop of the water-cooled air-conditioning, and achieves an effect of accelerating response. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the adaptive control method of water-cooled air conditioner in a computer room based on big data analysis of the present invention.

[0017] Figure 2 This is a flow chart of the adaptive control method of water-cooled air conditioner in a computer room based on big data analysis of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The invention deploys distributed optical fiber sensing in a network, uses a sliding time window algorithm to filter environmental data in a water-cooled air-conditioning adaptive system, enables an IIR digital filter to suppress noise and enter a thermal inertia compensation mechanism, constructs an LSTM neural network based on the initial system thermal inertia to predict thermal disturbances of subsequent multiple control cycles, introduces a time-space alignment compensation matrix to suppress harmonic oscillations in the thermal inertia compensation mechanism, uses a double closed-loop correction method to generate a thermal compensation correction amount based on the predicted thermal disturbance and the chilled water flow in the air-conditioning, detects environmental loads, sets a variable structure sliding membrane controller to perform gain switching on the double closed-loop correction method, uses a timing alignment engine to apply an improved DTW algorithm to synchronously align and correct asynchronously collected data packets, sets a curved window and performs characterization fusion to judge environmental data collection points and their abnormal collection response rates, and realizes a thermal inertia compensation closed loop for the water-cooled air-conditioning.

[0020] Embodiment, a method for adaptively controlling water-cooled air conditioners in a computer room based on big data analysis, such as Figure 1 and Figure 2 As shown, the following steps are included: Step S1: Deploy the distributed optical fiber sensor network, use the sliding time window algorithm to filter the environmental data in the water-cooled air conditioning adaptive system, enable the IIR digital filter to suppress the noise and enter the thermal inertia compensation mechanism; Step S2: Based on the initial system thermal inertia, an LSTM neural network is constructed to predict the thermal disturbances of subsequent multiple control cycles, and a spatiotemporal alignment compensation matrix is ​​introduced to suppress harmonic oscillations in the thermal inertia compensation mechanism; Step S3: using a double closed-loop correction method to generate a thermal compensation correction value based on the predicted thermal disturbance and the chilled water flow in the air conditioner, detecting the environmental load, and setting a variable structure sliding film controller to perform gain switching on the double closed-loop correction method; Step S4: The improved DTW algorithm is applied through the timing alignment engine to perform synchronous alignment correction on the asynchronously collected data packets, set the bending window and perform feature fusion to determine the environmental data collection points and their abnormal collection response rates.

[0021] The specific implementation is as follows: In step S1, when the distributed optical fiber sensor is deployed in the network, a fiber Bragg grating (FBG) array with a pitch of 0.8 mm is selected on the surface of the main pipe and branch pipe of the chilled water loop. This technology has the characteristics of high precision and high stability, and can effectively monitor temperature changes; Each set of sensor nodes is designed to contain 4 FBG sensors, which are arranged in a 120° circular distribution to ensure that the temperature distribution information of the pipe section can be fully captured. For example, in a chilled water pipe, the change in water temperature may be unevenly distributed due to the influence of flow rate, pipe wall thermal conductivity and external ambient temperature. The use of a circular arrangement can more accurately obtain these temperature gradient information and provide basic data for subsequent analysis; In order to achieve multiplexing, the sensor network uses wavelength division multiplexing (WDM) technology. The central wavelength is set in the range of 1529.55nm to 1567.36nm, and the wavelength interval is set to 0.8nm. In this way, multiple signals can be transmitted simultaneously in the same optical fiber, greatly improving the efficiency and accuracy of data collection; For example, in a large chilled water network, a control unit may contain hundreds of FBG sensors. If each sensor needs to be connected independently, the wiring will be complicated and costly. WDM technology allows multiple sensors to share the same optical fiber, greatly reducing the amount of optical fiber used and reducing system costs.

[0022] A sampling rate of 5 Hz and a resolution of 0.01°C are used to obtain environmental data and transmit it to the central controller, where the environmental data is the temperature change rate of the cold storage water tank; The sliding time window algorithm is used to perform dynamic data screening based on environmental data as follows: Preprocessing: Initialize the window length and step size, and mark the window length as T_win; Generate parameter sequence: Generate parameter sequence after de-dimensionalizing each environmental data ; It represents the data sequence within the sampling time window, that is, the measurement data collected within a specific time window T_win Calculate the standard deviation rate ,in, is the standard deviation of the parameter sequence, is the mean of the parameter sequence; Kurtosis ; Among them, kurtosis is used to measure the steepness of data distribution. Generally speaking, when the kurtosis coefficient is approximately equal to 3, the data follows a normal distribution. When the kurtosis coefficient is greater than 3, the data is concentrated near the mean. When the richness coefficient is less than 3, the data is flatly distributed. In the formula, For each data point in the sequence, is the mean of the series, is the standard deviation of the parameter sequence, and n is the total number of data points; For example, you can set valid data filtering conditions: When SDR<0.15 and |Kurtosis-3|<1, it is determined to be stable operating data and used for system analysis and decision-making; When the flow rate change rate is >15% / min, shorten T_win to half of the previous time, enable the IIR digital filter to suppress high-frequency noise, and set the cutoff frequency to 0.5Hz; The calculation formula of flow rate change rate is: ; In the formula, is the flow rate at the current time t, For the previous moment of traffic, Sampling time interval; It should be noted that the IIR digital filter is a common signal processing tool used to remove high-frequency noise and smooth data. Its characteristics include using recursive calculations to make the output value dependent on the current input value and the past output value; In this system, the IIR digital filter is enabled for high-frequency noise suppression, and the cutoff frequency is set to 0.5Hz, which means that the system will filter out fluctuations above 0.5Hz, thereby reducing short-term noise interference, making the flow data more stable, and facilitating subsequent analysis and control; In step S2, an LSTM neural network is constructed based on the initial system thermal inertia to predict the thermal disturbances of subsequent multiple control cycles. The specific steps are as follows: Construct an LSTM network with multiple hidden units, and the input layer receives the system thermal inertia at multiple time points, which are marked as: ; in, Load forecast data for the past 10 seconds. is the return water temperature in the past 5 seconds, The heat exchange rate change data for the past 3 seconds; Hidden layer: multiple LSTM units are used to learn the thermal inertia characteristics of the system; The number of LSTM units in the specific hidden layer is not limited, but is set by the experimenter according to the specific implementation method, which will not be described here; The output layer predicts the thermal disturbance of the next n control cycles, that is, ; The training adopts the Nadam optimizer, which combines the advantages of NAG accelerated gradient and Adam algorithm to improve the convergence speed and stability of training; Assume the initial learning rate is 0.001, the batch size is 256, and the loss function is defined as: ; In the formula, MAE is the average absolute difference between the predicted value and the true value; Among them, the calculation formula of MSLE mean square logarithmic error is: ; Among them, MSLE is suitable for situations with a large prediction range and can reduce the impact of extreme errors; In the formula, is a regularization term used to control model complexity and prevent overfitting. Specifically, the regularization coefficient λ=0.01; Defining compensation coefficients within the spatiotemporal alignment compensation matrix Calculation rules: when When the time deviation is less than half of the thermal inertia time constant, the calculation rule of the compensation coefficient is: ; when When the time deviation is greater than half of the thermal inertia time constant, the calculation rule of the compensation coefficient is: ; in, To track the error, the above calculation rules are used to suppress the harmonic oscillation during the compensation process; It should be noted that in data flow analysis, if the SDR of temperature data within a certain period of time is less than 0.15 and |Kurtosis-3|<1, the data is judged to be stable working condition data, otherwise it is necessary to further screen or filter abnormal data; For example, during the operation of the chilled water network, if there is a sudden increase in flow when the water pump is started, the system can automatically shorten the time window to capture key data and suppress high-frequency noise to improve monitoring accuracy; In step S3, the double closed-loop correction method is used to generate the thermal compensation correction value. The specific steps are as follows: Definition of the state vector: ; Where, τ is the thermal inertia time constant, Ceq is the equivalent heat capacity, and Qacc is the accumulated thermal deviation; Among them, τ reflects the response speed of the system to temperature changes, Ceq measures the heat storage capacity of the system and directly affects the thermal compensation calculation, and Qacc records the heat loss or accumulation of the system to adjust the compensation strategy; For example, in actual applications, if the equivalent heat capacity Ceq of the cold storage water tank changes beyond the normal range (such as an increase of more than 10%), it may mean that scaling occurs inside the water tank or the heat exchange efficiency is reduced, and maintenance or cleaning is required.

[0023] Adaptive adjustment formula of the observation matrix: ; In the formula, When the thermal compensation changes greatly, the system reduces the compensation intensity to avoid over-adjustment. 0.8 is the preset fixed weight to stabilize the influence of the observed data. is the time decay factor, which ensures that the impact of new observation data on the compensation amount is gradually reduced to prevent short-term fluctuations from interfering with system stability; Set the trigger rule: If the residual covariance satisfies the following inequality, that is, when hour; The system triggers the parameter update mechanism, the gain matrix The calculation formula is as follows: ; Among them, the process noise covariance is set as: , the observation noise covariance is set as: ; In the formula, 0.02 corresponds to the disturbance of the thermal inertia time constant, 0.01 corresponds to the error of the equivalent heat capacity, and 0.05 corresponds to the correction of the accumulated thermal deviation; Among them, the observation noise covariance is used to limit the impact of measurement errors and ensure system stability; It should be noted that the core of the dual closed-loop correction method is to adjust the compensation parameters in real time and optimize the system response, thereby ensuring the efficient and stable operation of the chilled water system, which will not be elaborated here; In step S4, the asynchronously collected data packets are synchronized and corrected, and a long short-term memory (LSTM) neural network with 128 hidden units is constructed to predict the dynamic characteristics of thermal inertia in the chilled water system.

[0024] The asynchronously collected data packets include load forecast, return water temperature, supply water temperature, compressor power, condenser temperature difference, evaporator heat exchange efficiency, ambient temperature, cooling tower fan speed, chilled water flow and pump pressure, which are marked as , , , , , , , , , , all asynchronously collected data packets are regarded as ten-dimensional time series data; Define input features and define the ten-dimensional time series data as input features; For example, in a chilled water system, the ambient temperature and cooling tower fan speed The dynamic changes will directly affect the heat exchange efficiency of the condenser , which in turn affects the overall cooling performance. Therefore, the dynamic correlation of these parameters must be incorporated into the input layer of the LSTM neural network for accurate modeling.

[0025] Network training and optimization: The Adam optimization algorithm is used, the initial learning rate is set to 0.001, and the learning rate decay rate is dynamically adjusted to 0.95 every 10 iterations; the loss function uses the mean square error (MSE): ; The amount of training data was set to 100,000 time steps, and the sliding window method (window size = 50s) was used for data augmentation; Perform cross validation: Use K-fold cross validation (K=5) to determine the environmental data collection points in the preset time step; For example, during the experiment, if the error of the LSTM model on the test data is continuously higher than 2.5%, it may indicate that there is a lag effect in the input data, and it is necessary to adjust the time window size or add data normalization preprocessing to improve the prediction accuracy; Among them, the normalization process can use min-max normalization to scale the input data to the [0,1] interval to reduce the learning bias caused by numerical differences; The dynamic thermal inertia of the chilled water system is mainly affected by the coupling of multiple physical fields such as fluid mechanics, heat transfer, and electrical load characteristics. In order to improve the compensation accuracy, a multi-physical field coupling compensation mechanism is adopted, and a multi-field model based on the finite element method (FEM) is designed, including: The thermal field compensation uses the finite difference method (FDM) to calculate the thermal inertia change of the chilled water pipe. The compensation formula is as follows: ; In the formula, is the specific heat capacity, is the fluid mass flow rate, is the thermal conductivity of the tube wall, is the heat transfer area, is the temperature difference, is the infinitesimal length of the pipeline; For example: when the chilled water flow rate decreases (such as 0.2m / s→0.1m / s), the heat exchange efficiency of the pipeline decreases, and the compensation parameters need to be automatically adjusted to avoid overcooling or overheating.

[0026] Power load field compensation: The transient fluctuation of compressor power is predicted by support vector regression model, and the compensation formula is as follows: ; Among them, k is the adaptive adjustment factor, which is updated in real time as the load changes. is the compressor power after compensation, is the currently measured compressor power, is the rate of change of the supply and return water temperature difference; For example, when the power grid in the refrigeration room fluctuates greatly (such as a 5% voltage drop), the system automatically adjusts the operating curve of the cooling tower fan to ensure power stability.

[0027] It should be noted that the support vector regression (SVR) for power load field compensation is only applicable to linear or weakly nonlinear short-term power fluctuation prediction. If the power disturbance is large or the nonlinearity is high, the LSTM or Transformer structure may be better. This solution uses SVR for prediction, but for complex power grid disturbance scenarios, more complex deep learning methods can be used for optimization, which will not be described here.

[0028] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0029] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application of the technical solution and the invention constraints. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0030] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0031] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0032] Finally: 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 in the protection scope of the present invention.

Claims

1. The adaptive control method of water-cooled air conditioner in computer room based on big data analysis is characterized in that: The following steps are included: Step S1: Deploy the distributed optical fiber sensor network, use the sliding time window algorithm to filter the environmental data in the water-cooled air conditioning adaptive system, enable the IIR digital filter to suppress the noise and enter the thermal inertia compensation mechanism; Step S2: Based on the initial system thermal inertia, an LSTM neural network is constructed to predict the thermal disturbances of subsequent multiple control cycles, and a spatiotemporal alignment compensation matrix is ​​introduced to suppress harmonic oscillations in the thermal inertia compensation mechanism; Step S3: using a double closed-loop correction method to generate a thermal compensation correction value based on the predicted thermal disturbance and the chilled water flow in the air conditioner, detecting the environmental load, and setting a variable structure sliding film controller to perform gain switching on the double closed-loop correction method; Step S4: The improved DTW algorithm is applied through the timing alignment engine to perform synchronous alignment correction on the asynchronously collected data packets, set the bending window and perform feature fusion to determine the environmental data collection points and their abnormal collection response rates.

2. The method for adaptive control of water-cooled air conditioners in a computer room based on big data analysis according to claim 1 is characterized in that: In step S1, when the distributed optical fiber sensor is deployed in the network, on the surface of the main pipe and branch pipe of the chilled water loop, each group of sensor nodes is designed to include 4 FBG sensors, and are arranged in a 120° ring distribution; The environmental data is acquired at a sampling rate of 5 Hz and transmitted to the central controller, where the environmental data is the temperature change rate of the cold storage water tank.

3. The method for adaptive control of water-cooled air conditioners in a computer room based on big data analysis according to claim 1 is characterized in that: In step S1, dynamic data screening is performed based on environmental data using a sliding time window algorithm as follows: Preprocessing: Initialize the window length and step size, and mark the window length as T_win; Generate parameter sequence: Generate parameter sequence after de-dimensionalizing each environmental data ; It represents the data sequence within the sampling time window, that is, the standard deviation rate of the measurement data collected within a specific time window T_win ,in, is the standard deviation of the parameter sequence, is the parameter sequence mean; the kurtosis coefficient is ; Among them, kurtosis is used to measure the steepness of data distribution. Generally speaking, when the kurtosis coefficient is approximately equal to 3, the data follows a normal distribution. When the kurtosis coefficient is greater than 3, the data is concentrated near the mean. When the richness coefficient is less than 3, the data is flatly distributed. In the formula, For each data point in the sequence, is the mean of the series, is the standard deviation of the parameter sequence, n is the total number of data points; the calculation formula for the flow rate change rate is: ,in, is the flow rate at the current time t, For the previous moment of traffic, Sampling time interval.

4. The method for adaptive control of water-cooled air conditioners in a computer room based on big data analysis according to claim 1 is characterized in that: In step S2, an LSTM neural network is constructed based on the initial system thermal inertia to predict the thermal disturbances of subsequent multiple control cycles. The specific steps are as follows: Construct an LSTM network with multiple hidden units, and the input layer receives the system thermal inertia at multiple time points, which are marked as: ; in, For past load forecast data, is the past return water temperature, The data of heat exchange rate changes in the past; Hidden layer: multiple LSTM units are used to learn the thermal inertia characteristics of the system; The output layer predicts the thermal disturbance of the next n control cycles, that is, ; Set the initial learning rate and define the loss function as: ； In the formula, MAE is the average absolute difference between the predicted value and the true value; Among them, the calculation formula of MSLE mean square logarithmic error is: ； Among them, MSLE is used to reduce the impact of extreme errors; In the formula, is a regularization term used to control the complexity of the model.

5. The method for adaptive control of water-cooled air conditioners in computer rooms based on big data analysis according to claim 1 is characterized in that: In step S2, the compensation coefficients are defined in the spatiotemporal alignment compensation matrix Calculation rules: when When the time deviation is less than half of the thermal inertia time constant, the calculation rule of the compensation coefficient is: ； when When the time deviation is greater than half of the thermal inertia time constant, the calculation rule of the compensation coefficient is: ； in, In order to track the error, the above calculation rules are used to suppress the harmonic oscillation in the compensation process.

6. The method for adaptive control of water-cooled air conditioners in computer rooms based on big data analysis according to claim 1 is characterized in that: In step S3, the double closed-loop correction method is used to generate the thermal compensation correction value. The specific steps are as follows: Define the state vector: , where τ is the thermal inertia time constant, Ceq is the equivalent heat capacity, and Qacc is the accumulated thermal deviation; where τ reflects the system's response speed to temperature changes, Ceq measures the system's heat storage capacity and directly affects the thermal compensation calculation, and Qacc records the system's heat loss or accumulation to adjust the compensation strategy; Adaptive adjustment formula of the observation matrix: , in, When the thermal compensation amount changes, 0.8 is the preset fixed weight to stabilize the influence of the observed data. is the time decay factor; Set the trigger rule: If the residual covariance satisfies the following inequality, that is, when hour; The system triggers the parameter update mechanism, the gain matrix The calculation formula is as follows: , where the process noise covariance is set as: , the observation noise covariance is set as: ,in, = , 0.02, 0.01, and 0.05 are the disturbance of the preset thermal inertia time constant, the error of the equivalent heat capacity, and the correction of the accumulated thermal deviation, respectively.

7. The method for adaptive control of water-cooled air conditioners in computer rooms based on big data analysis according to claim 1 is characterized in that: In step S4, the asynchronously collected data packets are synchronized and corrected, and a long short-term memory (LSTM) neural network with 128 hidden units is constructed to predict the dynamic characteristics of thermal inertia in the chilled water system; The asynchronously collected data packets include load forecast, return water temperature, supply water temperature, compressor power, condenser temperature difference, evaporator heat exchange efficiency, ambient temperature, cooling tower fan speed, chilled water flow rate and pump pressure, which are marked as , , , , , , , , , , all asynchronously collected data packets are regarded as ten-dimensional time series data; Define input features and define the ten-dimensional time series data as input features; Network training and optimization: The initial learning rate is set to 0.001, and the learning rate decay rate is dynamically adjusted to 0.95 every 10 iterations; the loss function uses the mean square error marked as MSE: ； The amount of training data is preset to 100,000 time steps, and data augmentation is performed according to the window size; Perform cross-validation: Use K-fold cross-validation to determine the environmental data collection points in the preset time step.

8. The method for adaptive control of water-cooled air conditioners in computer rooms based on big data analysis according to claim 7 is characterized in that: In step S4, the curved window is set by the finite difference method and the characteristic fusion is performed to determine the environmental data collection point and its abnormal collection response rate as follows: Calculate the thermal inertia change of the chilled water pipeline and the compensation formula is as follows: ;in, is the specific heat capacity, is the fluid mass flow rate, is the thermal conductivity of the tube wall, is the heat transfer area, is the temperature difference, is the infinitesimal length of the pipeline; The abnormal acquisition response rate is calculated using the power load field compensation method.

9. The method for adaptive control of water-cooled air conditioners in a computer room based on big data analysis according to claim 8 is characterized in that: In step S4, the specific steps of calculating the abnormal acquisition response rate using the power load field compensation method are as follows: The transient fluctuation of compressor electric power is predicted by support vector regression model, and the compensation formula is as follows: ; where k is the adaptive adjustment factor, is the load power of the compressor after compensation, is the currently measured compressor power, is the change rate of the supply and return water temperature difference, and the change rate of the supply and return water temperature difference is used as the abnormal collection response rate.

Images