Cloud platform data monitoring and management system of integrated immersion liquid cooling cabinet

Abstract

The invention discloses a cloud platform data monitoring and management system of an integrated immersion liquid cooling cabinet, and particularly relates to the technical field of cloud platform data management, and the cloud platform data monitoring and management system comprises the following steps: configuring a cooling liquid: configuring the type and quantity of the cooling liquid according to the heat dissipation demand of the cabinet; temperature, flow and cooling liquid consumption data in the cabinet are collected in real time, the flow and quantity of the cooling liquid are dynamically adjusted based on the data to ensure that the system maintains the optimal heat dissipation effect under different load conditions, historical data are analyzed through machine learning, the system can establish a cooling liquid aging state prediction model, and the cooling liquid aging state prediction accuracy is improved. A cooling liquid aging evaluation index is generated, the cooling liquid aging problem is early warned in advance, regular calibration of the sensor is based on dynamic adjustment of the cooling liquid aging state and data deviation, the data acquisition precision is ensured, and a data fusion algorithm is used for optimizing sensor data and eliminating the influence of failure or error of a single sensor.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud platform data management, and more specifically, to a cloud platform data monitoring and management system of an integrated immersion liquid cooling cabinet. Background Art

[0002] With the continuous growth of power consumption in data centers and high-performance computing equipment, traditional air cooling technology can no longer meet the heat dissipation needs. Immersion liquid cooling technology has become one of the efficient heat dissipation solutions for data centers due to its superior thermal conductivity and compact structural design. In particular, the integrated immersion liquid cooling cabinet uses coolant to completely immerse the equipment, and the liquid directly contacts the heat dissipation components to achieve efficient and low-noise heat dissipation effects. In order to improve the reliability and maintenance efficiency of this system, it is usually combined with the cloud platform data monitoring and management system, and sensors are used to collect data such as temperature, flow, and coolant consumption in real time for dynamic adjustment and optimization. However, although the immersion liquid cooling system has a strong heat dissipation capability, it faces many challenges in actual applications.

[0003] In daily use of the equipment, first of all, the type and flow of coolant need to be dynamically adjusted according to the heat dissipation requirements of the cabinet. If the flow is too large or too small, it will affect the cooling effect and may even cause the system to overheat. Traditional coolant flow adjustment relies on static settings, which are difficult to cope with rapid fluctuations in cabinet load or environmental changes. Secondly, the accuracy of the sensor is crucial to real-time monitoring of the system. However, the sensor may have data deviation or loss due to long-term use, external interference or equipment failure, resulting in inaccurate coolant configuration and flow adjustment. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a cloud platform data monitoring and management system for an integrated immersion liquid cooling cabinet. The temperature, flow rate and coolant consumption data inside the cabinet are collected in real time, and the flow rate and amount of the coolant are dynamically adjusted based on these data to ensure that the system maintains the best heat dissipation effect under different load conditions. Through machine learning and analysis of historical data, the system can establish a coolant aging status prediction model and generate a coolant aging evaluation index to provide early warning of coolant aging problems. The regular calibration of sensors is based on the dynamic adjustment of the coolant aging status and data deviation to ensure data acquisition accuracy. The data fusion algorithm is used to optimize sensor data and eliminate the impact of single sensor failure or error to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a cloud platform data monitoring and management system for an integrated immersion liquid cooling cabinet, comprising:

[0006] Coolant configuration: configure the type and amount of coolant according to the heat dissipation requirements of the cabinet, and dynamically adjust the flow or amount of coolant by monitoring the temperature, flow and coolant consumption in the cabinet;

[0007] Sensor data collection: Establish original threshold data by collecting temperature, flow, and coolant consumption data inside the cabinet;

[0008] Machine learning analysis: Analyze the deviation of the collected sensor historical data, establish a coolant aging status prediction model, and predict the coolant aging status by analyzing the historical data to generate a coolant aging evaluation index;

[0009] Regular calibration: Based on the results of the coolant aging status prediction model, the sensor is calibrated regularly to ensure the accuracy of the sensor data collected. The calibration interval is dynamically adjusted according to the coolant aging status and the deviation of the sensor's original threshold data;

[0010] Data feedback and optimization: Optimize the calibrated sensor data through data fusion algorithms to reduce the impact of individual sensor failures or errors;

[0011] Cloud platform monitoring: Upload the collected data to the cloud platform, and centrally process and analyze it through the cloud platform to monitor the status of the liquid cooling system and provide early warning and decision support.

[0012] In a preferred embodiment, the coolant configuration includes a temperature sensor and a flow meter, which monitor the temperature and flow rate of the coolant, and adjust the coolant flow and temperature so that the cabinet temperature is within the set threshold. The optimization adjustment formula is:

[0013]

[0014] Where, u(t) is the control output; K i is the integral gain; K d is the differential gain; ∫e(t)dt is the integral term; is the differential term; K p e(t) is proportional control; K p is the proportional gain; e(t) is the error; in the optimization adjustment formula, the target value is usually the desired temperature, and the actual value is the currently measured temperature. The temperature error expression formula is:

[0015] e(t)=T set -T actual

[0016] Among them, T set The threshold value set for the target temperature; T actual is the actual temperature.

[0017] In a preferred embodiment, the sensor data acquisition includes a temperature sensor, a humidity sensor and an airflow sensor, wherein the humidity sensor and the airflow sensor are used to monitor the humidity and airflow inside the cabinet, and the data of the temperature sensor, the humidity sensor and the airflow sensor are subjected to noise filtering and dynamic estimation by adopting the Kalman filtering step, and the Kalman filtering step is expressed as follows:

[0018]

[0019] in, is the state estimate at time k; is the prior estimate at time k-1; K k is the Kalman gain; P k|k-1 is the prediction error covariance matrix, H is the measurement matrix, and R is the measurement noise covariance; k is the observed data; the sensor data is estimated and corrected through the Kalman filter step; the dynamically estimated and corrected data are fused with the data of multiple sensors through the weighted average step, and the weighted average expression formula is:

[0020]

[0021] Among them, x final is the final data value after fusion; w i is the weight of the i-th sensor; x i is the output data of the i-th sensor; the temperature, humidity, and airflow multidimensional time series data are analyzed by the hidden Markov model, and the future data trend is predicted based on the time dependency of historical data. The state transition probability matrix A and observation probability matrix B of the hidden Markov model are combined with the observed data. The hidden Markov model expression formula is:

[0022]

[0023] Where P(O|λ) is the probability given the observation sequence O and the model parameter λ; α T (i) is the forward variable; the hidden pattern of multidimensional data is learned through the hidden Markov model to predict the long-term trend of the system. The dynamic estimation value of the Kalman filter step, the comprehensive data value of the weighted average step and the prediction result of the hidden Markov model step are fused through the weighted average step. The data fusion and optimization step expression formula is:

[0024]

[0025] Among them, w i The weight of each algorithm result; The output results of Kalman filter, weighted average and hidden Markov model.

[0026] In a preferred embodiment, the machine learning analysis uses a method combining a clustering model with a time series analysis model to optimize the prediction of the coolant aging state and the adjustment of the cooling strategy. The cluster analysis expression formula is:

[0027]

[0028] in, is the center point of the jth cluster; m j is the number of data points in the jth cluster; x i is the i-th data point of the j-th cluster; based on the data time series of each cluster center, a long short-term memory model is constructed to learn the dynamic change trend of sensor data and predict the changes of coolant aging related parameters in the future. The time series prediction formula is:

[0029] y t =f(W h h t-1 +W x x t +b)

[0030] Among them, y t is the predicted output at time t; x t is the input data at time t; h t-1 is the hidden state at the previous moment; W h With W x is the weight matrix; b is the bias term; f is the activation function.

[0031] In a preferred embodiment, the cluster centers The results of the long short-term memory model and the time series prediction results y t Perform weighted fusion and calculate comprehensive prediction results Its expression formula is:

[0032]

[0033] Among them, α is the weight coefficient dynamically adjusted based on the current data characteristics; is the comprehensive prediction result; t predicting outcomes for long-term and short-term memory; is the center point of the jth cluster; 1-α is the weight coefficient of the dynamic results of the time series.

[0034] In a preferred embodiment, regular calibration is performed according to the aging state of the coolant and the deviation of the temperature sensor, humidity sensor and airflow sensor, using an automated calibration device to perform physical calibration on the temperature sensor, humidity sensor and airflow sensor, and the calibration process is optimized by particle filtering, and the particle filter expression formula is:

[0035]

[0036] in, is the estimated value of the state; is the particle weight; is the particle position. During the coolant aging and temperature sensor, humidity sensor and airflow sensor calibration process, the temperature sensor, humidity sensor and airflow sensor data are observed by maximum likelihood estimation. The maximum likelihood estimation expression formula is:

[0037]

[0038] in, is the estimated parameter value; p(x i |θ) is the observed data x i The conditional probability density function under given parameters θ; Maximize the product of the probabilities of all observations.

[0039] In a preferred embodiment, the data feedback and optimization module includes data cleaning and noise removal, and uses a robust algorithm to remove abnormal noise and interference signals in the sensor data. The robust algorithm expression formula is:

[0040] y = argmin ∥ Ax - b ∥ 2

[0041] Among them, y is the result after fitting, usually the parameter or predicted value after fitting; A is the design matrix, which contains the eigenvector of sensor data; x is the parameter vector to be estimated; b is the observed data, that is, the original sensor data; in the data cleaning process, the noise and abnormal points in the data are identified and removed through the robust algorithm; weighted regression is performed in the local area of ​​the data set to calculate the weighted least squares solution, and the local weighted regression expression formula is:

[0042]

[0043] in, is the predicted value after weighted regression; w i is the weight of each data point; x i Input value for each data point.

[0044] Technical effects and advantages of the present invention:

[0045] 1. By dynamically adjusting the coolant flow and amount, the coolant configuration can be accurately controlled according to the real-time heat dissipation needs of the cabinet to ensure that the system is always in the best heat dissipation state. Through real-time monitoring based on temperature, flow and coolant consumption data, it can respond to changes in cabinet load in real time, effectively avoid overheating or overcooling, and ensure the stability and safety of equipment operation. Through machine learning, the coolant aging status is analyzed and the coolant aging evaluation index is generated, which can predict the aging of the coolant in advance;

[0046] 2. By dynamically adjusting the calibration interval and method, system operation deviations caused by sensor failure or errors are avoided, further improving the accuracy of monitoring. The data fusion algorithm can effectively reduce the impact of a single sensor failure or error on the overall performance of the system, ensuring the accuracy of coolant flow and temperature regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0048] 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.

[0049] Refer to the instruction manual Figure 1 , a cloud platform data monitoring and management system for an integrated immersion liquid cooling cabinet according to an embodiment of the present invention comprises:

[0050] Coolant configuration: configure the type and amount of coolant according to the heat dissipation requirements of the cabinet, and dynamically adjust the flow or amount of coolant by monitoring the temperature, flow and coolant consumption in the cabinet;

[0051] Sensor data collection: Establish original threshold data by collecting temperature, flow, and coolant consumption data inside the cabinet;

[0052] Machine learning analysis: Analyze the deviation of the collected sensor historical data, establish a coolant aging status prediction model, and predict the coolant aging status by analyzing the historical data to generate a coolant aging evaluation index;

[0053] Regular calibration: Based on the results of the coolant aging status prediction model, the sensor is calibrated regularly to ensure the accuracy of the sensor data collected. The calibration interval is dynamically adjusted according to the coolant aging status and the deviation of the sensor's original threshold data;

[0054] Data feedback and optimization: Optimize the calibrated sensor data through data fusion algorithms to reduce the impact of individual sensor failures or errors;

[0055] Cloud platform monitoring: Upload the collected data to the cloud platform, and centrally process and analyze it through the cloud platform to monitor the status of the liquid cooling system, provide early warning and decision support, and ensure the stability and long-term operation reliability of the liquid cooling system;

[0056] The coolant configuration includes a temperature sensor and a flow meter. The temperature and flow rate of the coolant are monitored by the temperature sensor and the flow meter. The cabinet temperature is optimized within the set threshold by adjusting the coolant flow and temperature. The optimization adjustment formula is:

[0057]

[0058] Where u(t) is the control output, which represents the signal output by the controller and is used to adjust the coolant flow or temperature to maintain the target temperature; K i is the integral gain; K d is the differential gain; ∫e(t)dt is the integral term, which represents the integration of the error e(t) over time; is the differential term; K p e(t) is proportional control, the current error value e(t) is multiplied by the proportional gain K p , and obtain a control signal whose size is proportional to the current error; K p is the proportional gain; e(t) is the error, which is the difference between the current state threshold and the target state threshold of the system; in the optimization adjustment formula, the target value is usually the desired temperature, and the actual value is the currently measured temperature. The temperature error expression formula is:

[0059] e(t)=T set -T actual

[0060] Among them, T set The threshold value set for the target temperature; T actual is the actual temperature; if the actual temperature is lower than the set temperature, the error e(t) is positive and the temperature needs to be increased; if the actual temperature is higher than the set temperature, the error e(t) is negative and the temperature needs to be lowered. Assuming the set temperature T set =30℃, actual temperature T actual =28℃, calculate the error e(t), which is expressed as: e(t)=30-28=2℃; Assuming the proportional gain K p =5, proportional term K p e(t) is: K pe(t)=5×2=10, which means the control system will output a control signal with a value of 10 to control the coolant flow or fan speed to reduce the temperature error. p e(t) indicates that the output of the system is proportional to the current error, that is, the larger the current error, the greater the adjustment of the control signal, thereby quickly correcting the temperature deviation, and the proportional gain K p It determines the response intensity of the control signal to the error. The key to coolant configuration is to monitor the temperature, flow and coolant consumption inside the cabinet in real time. The change of temperature reflects the heat dissipation demand of each device inside the cabinet. The flow directly affects the flow effect of the coolant inside the device, thus determining the heat dissipation efficiency. The consumption of coolant can reflect the attenuation of the system's heat dissipation capacity. Based on these data, through advanced control algorithms, the flow and amount of coolant can be adjusted dynamically in real time to ensure that the cooling effect is always maintained in the best state. For example, when the cabinet load increases, the system can automatically increase the coolant flow, otherwise it will reduce the flow, avoiding energy waste or poor cooling effect caused by excessive or insufficient coolant flow;

[0061] Sensor data collection includes temperature sensors, humidity sensors and airflow sensors. The humidity sensors and airflow sensors are used to monitor the humidity and airflow inside the cabinet to ensure the uniform distribution of temperature, humidity and airflow, and to ensure that the heat dissipation effect of the liquid cooling cabinet is not affected. The data of the temperature sensor, humidity sensor and airflow sensor are subjected to noise filtering and dynamic estimation by adopting the Kalman filtering step. The Kalman filtering step expression formula is:

[0062]

[0063] in, is the state estimate at time k; is the prior estimate at time k-1; K k is the Kalman gain, K k =P k|k-1 H T (HP k|k-1 H T +R) -1 ;P k|k-1 is the prediction error covariance matrix, H is the measurement matrix, and R is the measurement noise covariance; k The sensor data is the observed data; the sensor data is estimated and corrected in real time through the Kalman filter step to reduce the interference of sensor noise on sensor data such as temperature, humidity and airflow, so as to obtain certain real-time data; the dynamically estimated and corrected data is fused with the data of multiple sensors through the weighted average step, and the weighted average expression formula is:

[0064]

[0065] Among them, x final is the final data value after fusion; w i is the weight of the i-th sensor; x i is the output data of the i-th sensor; weight w i Dynamically adjust according to the accuracy, stability and historical performance of each sensor to make the high-precision sensor more influential, determine the accuracy after data fusion, analyze the temperature, humidity, and airflow multi-dimensional time series data through the hidden Markov model, and predict future data trends based on the time dependency of historical data. The state transition probability matrix A and observation probability matrix B of the hidden Markov model are combined with the observed data. The hidden Markov model expression formula is:

[0066]

[0067] Among them, λ=(A,B,π) are model parameters, including state transfer matrix A, observation probability matrix B, initial state probability π; P(O|λ) is the probability given observation sequence O and model parameter λ; α T (i) is a forward variable, which indicates the probability of being in state i at time T. The hidden Markov model is used to learn the implicit pattern of multidimensional data, predict the long-term trend of the system, and help identify potential faults such as coolant aging and sensor failure, so as to identify factors that may affect the heat dissipation effect in advance. The dynamic estimation value of the Kalman filter step, the comprehensive data value of the weighted average step, and the prediction result of the hidden Markov model step are fused through the weighted average step. The data fusion and optimization step expression formula is as follows:

[0068]

[0069] Among them, w i The weight of each algorithm's results is dynamically adjusted based on each algorithm's historical performance and the characteristics of the current input data; It is the output of Kalman filter, weighted average and hidden Markov model; by dynamically adjusting the weight w i , combining the advantages of multiple algorithms such as Kalman filtering, weighted averaging and hidden Markov model, determine the accuracy and reliability of data, and provide more stable and accurate real-time monitoring and management data. Sensor data acquisition is not only a simple physical quantity measurement process, it also involves how to convert these data into meaningful signals for subsequent processing and decision support. For example, the temperature sensor can monitor the temperature changes of various components inside the cabinet in real time, the flow sensor reflects the flow state of the coolant, and the coolant consumption sensor can determine the use of the coolant. Through these data, the system can establish the original threshold data, and based on these data, provide early warning and decision support to ensure that the cabinet is always in a suitable working environment;

[0070] By analyzing the sensor data, not only can the flow and amount of coolant be adjusted in real time, but it can also provide important reference for future maintenance work. For example, by monitoring the consumption of coolant, it can be determined whether there is leakage or efficiency decay in the liquid cooling system; by analyzing the temperature changes of various components inside the cabinet, potential overheating problems can be predicted and intervention can be made in advance. Data collection is not only a monitoring tool, it provides a basis for the intelligent and automated control of the system, and further improves the operating efficiency and reliability of the system.

[0071] Machine learning analysis uses a method that combines clustering models with time series analysis models to optimize the prediction of coolant aging status and the adjustment of cooling strategies. The clustering analysis expression formula is:

[0072]

[0073] in, is the center point of the jth cluster, and the cluster center value calculated by the clustering model represents the average static characteristics of the same type of data; m j is the number of data points in the jth cluster; x i is the i-th data point of the j-th cluster; based on the data time series of each cluster center, a long short-term memory model is constructed to learn the dynamic change trend of sensor data and predict the changes of coolant aging related parameters in the future. The time series prediction formula is:

[0074] y t =f(W h h t-1 +W x x t +b)

[0075] Among them, y t is the predicted output at time t; x t is the input data at time t; h t-1 is the hidden state at the previous moment; W h With W x is the weight matrix; b is the bias term, which is used to correct the predicted value; f is the activation function;

[0076] The cluster centers The results of the long short-term memory model and the time series prediction results y t Perform weighted fusion and calculate comprehensive prediction results Its expression formula is:

[0077]

[0078] Among them, α is a weight coefficient that is dynamically adjusted based on the characteristics of the current data, which is used to balance the impact of the clustering results and the dynamic time series prediction results, and is dynamically adjusted according to the characteristics of the current input data or historical performance; In order to comprehensively predict the results, the predicted values ​​of the final coolant-related parameters, coolant aging status, temperature, and humidity, are obtained by combining the analysis results of the clustering model and the dynamic prediction results of the long-short-term memory time series; y t It is the long short-term memory prediction result, which is the prediction value of the data at time t by the long short-term memory model, reflecting the time series change trend of the data; is the center point of the jth cluster; 1-α is the weight coefficient of the dynamic result of the time series, which dynamically adjusts the ratio of the clustering results and the long-short-term memory prediction results, and gives the long-short-term memory results a weight; by combining the static characteristic values ​​obtained by clustering and the time series forecast value y of the long short-term memory model t Fusion is performed according to the dynamic weight α to generate a comprehensive prediction result By dynamically adjusting the weight α, a balance is found between the results of the clustering model and the long-short-term memory model. This can take into account the global trend of static clustering and the real-time changes of dynamic prediction, thereby improving the accuracy and flexibility of prediction. Based on the historical data of sensors, machine learning analysis technology can automatically identify early signs of coolant aging. By training a large amount of historical data, the machine learning model can establish a coolant aging status prediction model to evaluate the current status of the coolant and its changing trend. The model analyzes multi-dimensional data such as temperature, flow rate, and coolant consumption to find out the relevant characteristics and laws of coolant aging, and generate a coolant aging evaluation index.

[0079] This evaluation index can provide system administrators with timely and accurate feedback on the coolant status, helping them determine whether the coolant needs to be replaced or replenished. Compared with traditional manual detection methods, machine learning analysis methods not only improve the accuracy of predictions, but also greatly improve the response speed of predictions. It can detect coolant problems earlier and avoid system heat dissipation failure or equipment damage caused by coolant aging. By establishing a data-driven coolant aging evaluation system, the system can achieve self-optimization, improve maintenance efficiency and long-term reliability of system operation.

[0080] Regular calibration According to the aging status of the coolant and the deviation of the temperature sensor, humidity sensor and airflow sensor, the temperature sensor, humidity sensor and airflow sensor are physically calibrated using automated calibration equipment. The calibration process is optimized by particle filtering, and the particle filter expression formula is:

[0081]

[0082] in, is the state estimation value, indicating the current calibration state of the sensor; is the particle weight, which indicates the contribution of each particle to the state estimation and is dynamically updated over time; is the particle position, which indicates the position of each particle in the state space and reflects the estimated state of the sensor. Particle weights and The particle position is used to track the system state. In the prediction of coolant aging and sensor calibration, particle filtering can efficiently estimate the state of the sensor, so as to regularly calibrate the temperature sensor, humidity sensor and airflow sensor to improve the accuracy of data acquisition. In the process of coolant aging and temperature sensor, humidity sensor and airflow sensor calibration, the temperature sensor, humidity sensor and airflow sensor data are observed by maximum likelihood estimation. The maximum likelihood estimation expression formula is:

[0083]

[0084] in, is the estimated parameter value, maximizing the likelihood function based on the observed data; p(x i |θ) is the observed data x i The conditional probability density function under given parameters θ; The probability product of all observed data is maximized; the maximum likelihood estimation can derive the optimal sensor parameters by observing the sensor data, and help determine the appropriate coolant aging state model. The particle filter can estimate the sensor state and coolant aging in real time in the dynamically changing data. When processing the sensor data, the accurate calibration value is determined. The maximum likelihood estimation helps to derive more accurate sensor parameters. Combined with the particle filter, it can effectively reduce the error in the sensor calibration process and improve the data accuracy. The system dynamically adjusts the sensor calibration interval according to the results of the coolant aging state prediction model. Different stages of coolant aging have different effects on the accuracy of the sensor, so the frequency of calibration should also be adjusted accordingly. For example, when the coolant is aged lightly, the accuracy of the sensor may remain stable and the calibration cycle can be longer; when the coolant is aged severely, the accuracy of the sensor may be more affected and more frequent calibration is required. Through this dynamic adjustment mechanism, it can ensure that the sensor data is always accurate and reliable.

[0085] Regular calibration can not only ensure the data accuracy of sensors such as temperature and flow, but also promptly detect possible sensor failures or errors, and restore them to normal working state by replacement or repair. Through this mechanism, the system can reduce misjudgment or delayed response caused by sensor problems, improve the accuracy of coolant flow and temperature regulation, and ensure the efficient operation of the liquid cooling system.

[0086] The data feedback and optimization module includes data cleaning and noise removal. The robust algorithm is used to remove abnormal noise and interference signals in the sensor data to further improve the accuracy of the data. The robust algorithm expression formula is:

[0087] y = argmin ∥ Ax - b ∥ 2

[0088] Among them, y is the result after fitting, usually the parameter or predicted value after fitting; A is the design matrix, which contains the eigenvector of the sensor data; x is the parameter vector to be estimated; b is the observed data, that is, the original sensor data; in the data cleaning process, the noise and outliers in the data are effectively identified and removed through the robust algorithm, the fitting error caused by outliers is reduced, and the data quality is improved; weighted regression is performed in the local area of ​​the data set to calculate the weighted least squares solution, and the local weighted regression expression formula is:

[0089]

[0090] in, is the predicted value after weighted regression; w i is the weight of each data point, weighted according to the distance from the target point; x i The input value for each data point is usually the data measured by the sensor; local weighted regression is used to flexibly fit the data in the local area, and different weights are assigned to different data points to accurately capture the local trend in the data. In the process of cleaning and denoising, local weighted regression can fit the data and remove abnormal points far from the true value by assigning higher weights to adjacent data points. Combined with the outlier removal of the robust algorithm, local weighted regression can effectively reduce the impact of noise on the final result and provide clearer data feedback. In a complex system, the failure or error of a single sensor may cause deviation in the system data, thereby affecting the adjustment of the coolant flow and temperature control. In order to solve this problem, the data fusion algorithm plays a vital role in the immersion liquid cooling system. Through data fusion technology, the system can comprehensively process data from multiple sensors, thereby overcoming the impact of single sensor failure or data error.

[0091] The data fusion algorithm can effectively eliminate inaccurate data signals and retain valid data that is critical to system operation by weighted averaging, filtering and correcting sensor data. In this way, even if some sensors fail, the system can still be corrected and optimized through data from other sensors to ensure the accuracy and stability of coolant flow and temperature control.

[0092] In addition, the data fusion algorithm can also adaptively optimize parameters such as coolant flow and temperature through intelligent algorithms, so that the entire system always remains in the best operating state. Through multi-sensor data fusion, the system can reduce error accumulation and ensure precise control of coolant flow and temperature regulation, thereby improving the overall efficiency and reliability of the liquid cooling system.

[0093] 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. A cloud platform data monitoring and management system for an integrated immersion liquid cooling cabinet, characterized in that: include: Coolant configuration: configure the type and amount of coolant according to the heat dissipation requirements of the cabinet, and dynamically adjust the flow or amount of coolant by monitoring the temperature, flow and coolant consumption in the cabinet; Sensor data collection: Establish original threshold data by collecting temperature, flow, and coolant consumption data inside the cabinet; Machine learning analysis: Analyze the deviation of the collected sensor historical data, establish a coolant aging status prediction model, and predict the coolant aging status by analyzing the historical data to generate a coolant aging evaluation index; Regular calibration: Based on the results of the coolant aging status prediction model, the sensor is calibrated regularly to ensure the accuracy of the sensor data collected. The calibration interval is dynamically adjusted according to the coolant aging status and the deviation of the sensor's original threshold data; Data feedback and optimization: Optimize the calibrated sensor data through data fusion algorithms to reduce the impact of individual sensor failures or errors; Cloud platform monitoring: Upload the collected data to the cloud platform, and centrally process and analyze it through the cloud platform to monitor the status of the liquid cooling system and provide early warning and decision support.

2. The cloud platform data monitoring and management system of the integrated immersion liquid cooling cabinet according to claim 1, characterized in that: The coolant configuration includes a temperature sensor and a flow meter. The temperature and flow rate of the coolant are monitored by the temperature sensor and the flow meter. The cabinet temperature is kept within the set threshold by adjusting the coolant flow and temperature. The optimization adjustment formula is: Where, u(t) is the control output; K i is the integral gain; K d is the differential gain; ∫e(t)dt is the integral term; is the differential term; K p e(t) is proportional control; K p is the proportional gain; e(t) is the error; in the optimization adjustment formula, the target value is usually the desired temperature, and the actual value is the currently measured temperature. The temperature error expression formula is: e(t)=T set -T actual Among them, T set The threshold value set for the target temperature; T actual is the actual temperature.

3. The cloud platform data monitoring and management system of the integrated immersion liquid cooling cabinet according to claim 2, characterized in that: Sensor data collection includes temperature sensors, humidity sensors and airflow sensors. The humidity sensors and airflow sensors are used to monitor the humidity and airflow inside the cabinet. The data of the temperature sensor, humidity sensor and airflow sensor are subjected to noise filtering and dynamic estimation by adopting the Kalman filtering step. The Kalman filtering step expression formula is: in, is the state estimate at time k; is the prior estimate at time k-1; K k is the Kalman gain; P k|k-1 is the prediction error covariance matrix, H is the measurement matrix, and R is the measurement noise covariance; k is the observed data; the sensor data is estimated and corrected through the Kalman filter step; the dynamically estimated and corrected data are fused with the data of multiple sensors through the weighted average step, and the weighted average expression formula is: Among them, x final is the final data value after fusion; w i is the weight of the i-th sensor; x i is the output data of the i-th sensor; the temperature, humidity, and airflow multidimensional time series data are analyzed by the hidden Markov model, and the future data trend is predicted based on the time dependency of historical data. The state transition probability matrix A and observation probability matrix B of the hidden Markov model are combined with the observed data. The hidden Markov model expression formula is: Where P(O|λ) is the probability given the observation sequence O and the model parameter λ; α T (i) is the forward variable; the hidden pattern of multidimensional data is learned through the hidden Markov model to predict the long-term trend of the system. The dynamic estimation value of the Kalman filter step, the comprehensive data value of the weighted average step and the prediction result of the hidden Markov model step are fused through the weighted average step. The data fusion and optimization step expression formula is: Among them, w i The weight of each algorithm result; The output results of Kalman filter, weighted average and hidden Markov model.

4. The cloud platform data monitoring and management system of the integrated immersion liquid cooling cabinet according to claim 3, characterized in that: Machine learning analysis uses a method that combines clustering models with time series analysis models to optimize the prediction of coolant aging status and the adjustment of cooling strategies. The clustering analysis expression formula is: in, is the center point of the jth cluster; m j is the number of data points in the jth cluster; x i is the i-th data point of the j-th cluster; based on the data time series of each cluster center, a long short-term memory model is constructed to learn the dynamic change trend of sensor data and predict the changes of coolant aging related parameters in the future. The time series prediction formula is: y t =f(W h h t-1 +W x x t +b) Among them, y t is the predicted output at time t; x t is the input data at time t; h t-1 is the hidden state at the previous moment; W h With W x is the weight matrix; b is the bias term; f is the activation function.

5. The cloud platform data monitoring and management system of the integrated immersion liquid cooling cabinet according to claim 4, characterized in that: The cluster centers The results of the long short-term memory model and the time series prediction results y t Perform weighted fusion and calculate comprehensive prediction results Its expression formula is: Among them, α is the weight coefficient dynamically adjusted based on the current data characteristics; is the comprehensive prediction result; t predicting outcomes for long-term and short-term memory; is the center point of the jth cluster; 1-α is the weight coefficient of the dynamic results of the time series.

6. The cloud platform data monitoring and management system of the integrated immersion liquid cooling cabinet according to claim 5, characterized in that: Regular calibration According to the aging status of the coolant and the deviation of the temperature sensor, humidity sensor and airflow sensor, the temperature sensor, humidity sensor and airflow sensor are physically calibrated using automated calibration equipment. The calibration process is optimized by particle filtering, and the particle filter expression formula is: in, is the estimated value of the state; is the particle weight; is the particle position. During the coolant aging and temperature sensor, humidity sensor and airflow sensor calibration process, the temperature sensor, humidity sensor and airflow sensor data are observed by maximum likelihood estimation. The maximum likelihood estimation expression formula is: in, is the estimated parameter value; p(x i |θ) is the observed data x i The conditional probability density function under given parameters θ; Maximize the product of the probabilities of all observations.

7. The cloud platform data monitoring and management system of the integrated immersion liquid cooling cabinet according to claim 6, characterized in that: The data feedback and optimization module includes data cleaning and noise removal. The robust algorithm is used to remove abnormal noise and interference signals in the sensor data. The robust algorithm expression formula is: y=argmin||Ax-b||2 Among them, y is the result after fitting, usually the parameter or predicted value after fitting; A is the design matrix, which contains the eigenvector of sensor data; x is the parameter vector to be estimated; b is the observed data, that is, the original sensor data; in the data cleaning process, the noise and abnormal points in the data are identified and removed through the robust algorithm; weighted regression is performed in the local area of ​​the data set to calculate the weighted least squares solution, and the local weighted regression expression formula is: in, is the predicted value after weighted regression; w i is the weight of each data point; x i Input value for each data point.

Images