A circulating heat dissipation fault detection system for an immersion liquid-cooled server

Abstract

The present invention discloses a circulating heat dissipation fault detection system for an immersion liquid-cooled server, and relates to the technical field of server heat dissipation fault detection. Existing server heat dissipation fault detection systems are too dependent on real-time temperature monitoring and lack long-term trend analysis and prediction capabilities, and are unable to fully prevent potential faults; the response speed and efficiency of the cooling system are insufficient, and it is difficult to respond to sudden heat increases in a timely manner; the system is complex and the degree of integration is not high, resulting in maintenance difficulties and increased costs. The present invention monitors the internal and external temperatures of the server in real time and combines the load conditions to dynamically adjust the data collection priority of the component to ensure timely acquisition of relevant data such as its performance and temperature, sets an early warning mechanism according to changes in power consumption, triggers corresponding alarms and makes corresponding adjustments, and increases the data collection frequency of the component to promptly detect abnormal conditions of the heat dissipation system and take corresponding measures to ensure stable operation of the server.

Description

Technical Field

[0001] The invention relates to the technical field of server heat dissipation fault detection, and in particular to a circulating heat dissipation fault detection system for an immersion liquid-cooled server. Background Art

[0002] The server circulating cooling fault detection system is a technology used to monitor the status of the server cooling system and detect faults. The system uses temperature sensors to monitor the temperature of key components inside the server, such as the CPU, memory, hard disk, etc. By regularly collecting temperature data, temperature anomalies can be discovered in a timely manner.

[0003] In this regard, the patent document with publication number CN104424058A discloses a server heat dissipation detection system, which includes a user terminal, an MCU, a temperature sensing module, a plurality of fans for heat dissipation, a switch module and a heating module. The user terminal is provided with a user interface for inputting preset parameters. The heating module includes a plurality of heating resistors distributed in a matrix. The switch module includes a plurality of switches for controlling the on and off of the heating resistors. The MCU opens the corresponding switches of the heating resistors corresponding to the preset positions of the heating elements according to the preset parameters, so that the heating resistors at the corresponding preset positions work to simulate the heating elements to generate heat, and at the same time controls the fans to work at a preset speed. The temperature sensing module is used to sense the temperature at the heating resistors and transmit the temperature to the MCU. The MCU compares the received temperature with a preset temperature in the preset parameters, and displays the comparison result on a display screen to determine whether the preset position of the heating element and the speed of the fan meet the requirements.

[0004] In this regard, the patent document with publication number CN106844155B discloses a server heat dissipation control method, including: detecting the current status of each server node, and determining whether it meets the preset status requirements; if so, continuing to determine whether the radiators corresponding to each of the current server nodes are all in place, if so, controlling the working state of each of the radiators according to the preset heat dissipation regulation strategy; if not, improving the heat dissipation efficiency of the radiators already in place. The server heat dissipation control method disclosed in the present invention improves the heat dissipation efficiency of the radiators already in place, so that the heat dissipation effect of the radiators adjacent to the server nodes with heat dissipation problems is enhanced, thereby spreading to the surrounding server nodes, so that each radiator also dissipates heat for the server nodes that originally do not correspond, thereby taking all server nodes into consideration. The present invention also discloses a server heat dissipation control system, and its beneficial effects are as described above.

[0005] The existing server heat dissipation fault detection system relies too much on real-time temperature monitoring and lacks long-term trend analysis and prediction capabilities, and cannot fully prevent potential failures; the cooling system's response speed and efficiency are insufficient, making it difficult to respond to sudden heat increases in a timely manner; the system is complex and not highly integrated, resulting in difficult maintenance and increased costs.

[0006] In view of the above problems, a circulating heat dissipation fault detection system for an immersion liquid-cooled server is proposed. Summary of the invention

[0007] The purpose of the present invention is to provide an immersion liquid cooling server circulation heat dissipation fault detection system, which solves the problem of excessive monitoring dependence and lack of long-term trend analysis and prediction capabilities in the background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: an immersion liquid cooling server circulation heat dissipation fault detection system, comprising:

[0009] Data collection module, deploy data collection nodes, identify key parts of the server, including CPU, GPU, memory, hard disk, power supply, deploy temperature sensors in key parts to ensure that data collection nodes obtain temperature information;

[0010] Initialize the data acquisition parameters, design the data acquisition frequency to be 1 second / time, the acquisition range to be 0-160 degrees Celsius, and the acquisition accuracy to be 0.1 degrees Celsius;

[0011] Hardware power adjusts collection priority and monitors the power consumption of server hardware;

[0012] The data collection node collects temperature data, stores the collected temperature data in the database, and updates the temperature data regularly;

[0013] The temperature prediction module divides the collected data into overlapping segments of 100 data points each, initializes an autoencoder neural network, and sets the architecture of the autoencoder neural network, including the input layer, encoding layer, decoding layer, and output layer;

[0014] Initialize the autoencoder neural network with 100 input neurons, 50 encoding neurons, and 100 output neurons. Input each small segment into the autoencoder neural network, which will learn the potential representation of the temperature data in the small segment, reconstruct the small segment through the decoder layer, and compare it with the original small segment.

[0015] The prediction error (the difference between the original and reconstructed segments) is used to update the weights of the autoencoder neural network;

[0016] The training goal of the autoencoder neural network is to minimize the prediction error, the mean square error between the original segment and the reconstructed segment.

[0017] Mean square error = (1 / n)*Σ(x_i-x_i')^2;

[0018] Where n is the number of data points in the segment; x_i is the i-th data point in the original segment;

[0019] x_i' is the i-th data point in the reconstructed segment; the autoencoder neural network updates its weights through the back-propagation algorithm to minimize the mean square error;

[0020] A heat dissipation fault detection module obtains a predicted temperature from a temperature prediction module;

[0021] Get the actual temperature from the data acquisition module, calculate the difference between the predicted temperature and the actual temperature, set the threshold, the threshold value is 3 degrees Celsius, the calculation formula is:

[0022] |Predicted temperature - actual temperature|>threshold;

[0023] Dynamic cooling control module, which monitors the power consumption of server hardware and adjusts cooling system parameters based on the calculated heat output;

[0024] System management module, real-time monitoring of system status, including data acquisition node status, temperature prediction model performance, cooling system performance, and server temperature;

[0025] Manage data collection nodes, add, delete or modify data collection nodes, configure data collection parameters (such as frequency, range and accuracy), monitor the health of data collection nodes, and configure prediction models and intervention measures;

[0026] Configure the temperature prediction model, including training data, model architecture, and training hyperparameters;

[0027] Configure intervention measures, including abnormal situation thresholds and trigger actions.

[0028] Preferably, the data acquisition node periodically acquires the data of the temperature sensor, stores the collected temperature data in the database, updates the data periodically, and adjusts the priority of data acquisition according to the change of hardware power. The exponentially weighted moving average algorithm can be expressed as: P(t)=(1 - α)×P(t - 1) + α × P new;

[0029] Where P(t) is the estimated power at time t, Pnew is the new power sample value, and α is the weighting coefficient.

[0030] Preferably, before implementing the temperature prediction module, the collected data needs to be preprocessed, and the preprocessing steps include data cleaning, normalization and data segmentation;

[0031] Data cleaning: dealing with outliers or missing values ​​in the data;

[0032] Normalization: scaling the data to a similar range;

[0033] Data segmentation: Divide the continuous data into overlapping segments of 100 data points each and input them into the autoencoder neural network.

[0034] Preferably, the input layer comprises 100 neurons for receiving the original small segment temperature data;

[0035] Encoding layer: contains 50 neurons, which are used to compress input data into potential representations;

[0036] Decoding layer: contains 100 neurons, used to decode the latent representation into raw data;

[0037] Output layer: Same as the input layer, containing 100 neurons, used to reconstruct small segments and initialize the weights and biases of the autoencoder neural network. The weights and biases are randomly initialized to start training.

[0038] Preferably, during the training process, each small segment of temperature data is input into the autoencoder neural network. The goal of the training is to minimize the difference between the reconstructed data and the original data. The training process can be divided into the following steps:

[0039] S1, forward propagation: input the original small segment of temperature data into the network, generate potential representation through the encoder layer, and reconstruct the data through the decoder layer;

[0040] S2, calculate the loss: calculate the mean square error between the reconstructed data and the original data as the prediction error;

[0041] S3, back propagation: Use the back propagation algorithm to update the weights and biases of the network to minimize the prediction error;

[0042] S4, repeat training: Repeat the above steps until the preset number of training times is reached or converges to a satisfactory error level.

[0043] Preferably, the forward propagation calculation formula is: h=f(Wx+b);

[0044] x is the input data; W is the weight matrix; b is the bias vector; f is the activation function; h is the output of the encoding layer;

[0045] In the autoencoder neural network, the back propagation calculation formula is:

[0046]

[0047]

[0048] Where L is the loss function; and They are the gradients of the loss function with respect to weights and biases respectively; is the gradient of the loss function with respect to the output of the encoding layer;

[0049] In the autoencoder neural network, the commonly used loss function is used to measure the difference between the reconstructed data and the original data. The calculation formula of the loss function is as follows:

[0050]

[0051] x i It is the original data; is the reconstruction data; N is the number of data points.

[0052] Preferably, a threshold is set to determine when a heat dissipation fault is considered to have occurred, and the threshold value is 3 degrees Celsius, that is, if the difference between the predicted temperature and the actual temperature is greater than 3 degrees Celsius, a heat dissipation fault is considered to have occurred, and the difference is greater than the threshold, triggering the detection of a heat dissipation fault.

[0053] Preferably, feedback and processing are performed based on the results of the fault detection, an alarm is issued to the operator, the fault event is recorded, and automatic or manual measures are taken to adjust the cooling system.

[0054] Preferably, the power consumption data is used to calculate the heat generated by the server, and the heat calculation formula is:

[0055] Q = P × t; where Q is the calorific value (unit: joule);

[0056] P is the power consumption (in watts);

[0057] t is time (in seconds).

[0058] Preferably, the parameters of the cooling system, including the fan speed and the water pump flow rate, are dynamically adjusted according to the calculated heat value;

[0059] In control theory, the feedback control algorithm, PID control algorithm, continuously adjusts the control variable (fan speed or water pump flow) to make the controlled variable (server temperature) close to the set value;

[0060] Feedback control algorithm calculation formula:

[0061]

[0062] U(t) is the controlled variable; e(t) is the error; K p is the ratio; K i is the integral; K d is the differential coefficient.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. The present invention provides an immersion liquid-cooled server circulation heat dissipation fault detection system. By real-time monitoring of the internal and external temperatures of the server and combining the load conditions, the system can dynamically adjust the fan speed and water pump flow, thereby effectively controlling the server temperature and avoiding performance degradation or even hardware damage due to overheating. Secondly, the system can reduce energy consumption as much as possible, improve energy efficiency, and reduce energy consumption and operating costs while ensuring the heat dissipation effect. In addition, it has fault detection and prediction functions, can promptly detect abnormal conditions in the heat dissipation system and take corresponding measures to ensure the stable operation of the server. In summary, it can effectively improve the efficiency and reliability of server heat dissipation management, provide enterprises with a more stable and efficient server operating environment, and thereby enhance the sustainable operation capability and competitiveness of the business. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0066] Figure 2 It is a structural schematic diagram of the present invention;

[0067] Figure 3 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0068] 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.

[0069] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings.

[0070] Combination Figure 1-Figure 3 , a circulating heat dissipation fault detection system for an immersion liquid cooling server of the present invention comprises:

[0071] In the collection of temperature data, by establishing data collection nodes, the designed data collection nodes are deployed in various key parts of the server to ensure that the data collection nodes can accurately obtain the status information of the key parts and continuously adjust the priority according to the changes in server hardware power. The adjustment of this priority is constantly changing according to the power changes. On this basis, the collected temperature is predicted, and the prediction task is divided into multiple small segments. Each segment is processed by a sub-neural network, and the results are summarized. Finally, according to the temperature prediction results, it is analyzed whether the heat dissipation of each part is normal. In the case of abnormalities, early intervention is made according to the prediction results to reduce losses. When it is judged that there is no abnormality, the server cooling system parameters are dynamically adjusted according to the hardware heat generation to cool and prevent excessive temperature.

[0072] The specific implementation steps are as follows:

[0073] Data collection module, deploy data collection nodes, identify key parts of the server, including CPU, GPU, memory, hard disk, power supply, deploy temperature sensors in key parts to ensure that data collection nodes obtain temperature information;

[0074] Initialize the data acquisition parameters, design the data acquisition frequency to be 1 second / time, the acquisition range to be 0-160 degrees Celsius, and the acquisition accuracy to be 0.1 degrees Celsius;

[0075] Hardware power adjusts collection priority and monitors the power consumption of server hardware;

[0076] The data collection node collects temperature data, stores the collected temperature data in the database, and updates the temperature data regularly;

[0077] The data acquisition node periodically obtains the data from the temperature sensor, stores the collected temperature data in the database, updates the data regularly, and adjusts the priority of data acquisition according to the change of hardware power. The exponentially weighted moving average algorithm can be expressed as: P(t) = (1 - α)×P(t - 1) + α × P new;

[0078] Where P(t) is the estimated power at time t, Pnew is the new power sample value, and α is the weighting coefficient.

[0079] The priority is constantly adjusted according to the power changes of the server hardware. When the CPU power increases, the CPU is adjusted to the first level. The first-level data is processed first, followed by other server hardware. This priority adjustment is constantly changing according to the power changes. According to the detected power changes, the system needs to adjust the priority of data processing in real time. When the CPU power increases, the CPU data will be adjusted to the first priority, and the data of other hardware will be lowered accordingly. According to the priority of each hardware component, the collected temperature data is processed, and those marked as high priority data are processed first to ensure that the system can respond quickly under high load;

[0080] The temperature prediction module divides the collected data into overlapping segments of 100 data points each. Before implementing the temperature prediction module, the collected data needs to be preprocessed. The preprocessing steps include data cleaning, normalization, and data segmentation.

[0081] Data cleaning: dealing with outliers or missing values ​​in the data;

[0082] Normalization: scaling the data to a similar range;

[0083] Data segmentation: Divide the continuous data into overlapping segments of 100 data points each, input them into the autoencoder neural network, initialize an autoencoder neural network, and set the architecture of the autoencoder neural network, including the input layer, encoding layer, decoding layer, and output layer;

[0084] Initialize the autoencoder neural network with 100 input neurons, 50 encoding neurons, and 100 output neurons. Input each small segment into the autoencoder neural network, which will learn the potential representation of the temperature data in the small segment, reconstruct the small segment through the decoder layer, and compare it with the original small segment.

[0085] Input layer: contains 100 neurons, which are used to receive the original small segment temperature data;

[0086] Encoding layer: contains 50 neurons, which are used to compress input data into potential representations;

[0087] Decoding layer: contains 100 neurons, used to decode the latent representation into raw data;

[0088] Output layer: Same as the input layer, containing 100 neurons, used to reconstruct small segments and initialize the weights and biases of the autoencoder neural network. The weights and biases are randomly initialized to start training.

[0089] The prediction error (the difference between the original and reconstructed segments) is used to update the weights of the autoencoder neural network;

[0090] The training goal of the autoencoder neural network is to minimize the prediction error, the mean square error between the original segment and the reconstructed segment.

[0091] Mean square error = (1 / n)*Σ(x_i-x_i')^2;

[0092] Where n is the number of data points in the segment; x_i is the i-th data point in the original segment;

[0093] x_i' is the i-th data point in the reconstructed segment; the autoencoder neural network updates its weights through the back-propagation algorithm to minimize the mean square error;

[0094] During the training process, each small segment of temperature data is input into the autoencoder neural network. The goal of the training is to minimize the difference between the reconstructed data and the original data. The training process can be divided into the following steps:

[0095] S1, forward propagation: input the original small segment of temperature data into the network, generate potential representation through the encoder layer, and reconstruct the data through the decoder layer;

[0096] S2, calculate the loss: calculate the mean square error between the reconstructed data and the original data as the prediction error;

[0097] S3, back propagation: Use the back propagation algorithm to update the weights and biases of the network to minimize the prediction error;

[0098] S4, repeat training: Repeat the above steps until the preset number of training times is reached or converges to a satisfactory error level.

[0099] Forward propagation calculation formula: h = f (Wx + b);

[0100] x is the input data; W is the weight matrix; b is the bias vector; f is the activation function; h is the output of the encoding layer;

[0101] In the autoencoder neural network, the back propagation calculation formula is:

[0102]

[0103]

[0104] Where L is the loss function; and They are the gradients of the loss function with respect to weights and biases respectively; is the gradient of the loss function with respect to the output of the encoding layer;

[0105] In the autoencoder neural network, the commonly used loss function is used to measure the difference between the reconstructed data and the original data. The calculation formula of the loss function is as follows:

[0106]

[0107] x i It is the original data; is the reconstructed data; N is the number of data points; the autoencoding neural network prediction model helps to accurately predict the temperature trend of the server, so that the cooling system can automatically adjust the cooling intensity according to actual needs, avoiding unnecessary overcooling, thereby saving energy. The dynamic cooling control strategy ensures that the cooling system operates at the most efficient state only when necessary, reducing overall energy consumption.

[0108] A heat dissipation fault detection module obtains a predicted temperature from a temperature prediction module;

[0109] Get the actual temperature from the data acquisition module, calculate the difference between the predicted temperature and the actual temperature, set the threshold, and set the threshold to determine when a heat dissipation fault occurs. The threshold is 3 degrees Celsius, that is, if the difference between the predicted temperature and the actual temperature is greater than 3 degrees Celsius, a heat dissipation fault is considered to have occurred. The difference is greater than the threshold, triggering the detection of a heat dissipation fault. The threshold is 3 degrees Celsius, and the calculation formula is:

[0110] |Predicted temperature - actual temperature|>threshold, based on the results of fault detection, feedback and processing are performed to alert the operator, record the fault event, and take manual measures to adjust the cooling system.

[0111] Dynamic cooling control module, which monitors the power consumption of server hardware and adjusts cooling system parameters based on the calculated heat output;

[0112] In server management, data collection priorities are set for different hardware components, and the data collection priority is dynamically adjusted according to changes in power consumption, so as to pay more attention to components with higher power consumption. By continuously monitoring the power consumption data of server hardware, the system can identify in real time which components have higher power consumption and which components have larger power changes. When the system detects that the power consumption of a hardware component has increased significantly or fluctuated greatly, the data collection priority of the component can be dynamically adjusted to ensure that relevant data such as its performance and temperature are obtained in a timely manner. An early warning mechanism is set according to changes in power consumption. When the power consumption of a component exceeds the set threshold, the corresponding alarm is triggered and corresponding adjustments are made to increase the data collection frequency of the component.

[0113] System management module, real-time monitoring of system status, including data collection node status, temperature prediction model performance, cooling system performance, server temperature, power consumption data to calculate the heat generated by the server, heat calculation formula:

[0114] Q = P × t; where Q is the calorific value (unit: joule);

[0115] P is the power consumption (in watts);

[0116] t is time (in seconds);

[0117] Dynamically adjust the cooling system parameters, including fan speed and water pump flow, based on the calculated heat output;

[0118] In control theory, the feedback control algorithm, PID control algorithm, continuously adjusts the control variable (fan speed or water pump flow) to make the controlled variable (server temperature) close to the set value;

[0119] Feedback control algorithm calculation formula:

[0120]

[0121] U(t) is the controlled variable; e(t) is the error; K p is the ratio; K i is the integral; K d is the differential coefficient;

[0122] Manage data collection nodes, add, delete or modify data collection nodes, configure data collection parameters (such as frequency, range and accuracy), monitor the health of data collection nodes, and configure prediction models and intervention measures;

[0123] Configure the temperature prediction model, including training data, model architecture, and training hyperparameters;

[0124] Configure intervention measures, including abnormal situation thresholds and trigger actions. By real-time monitoring of the temperature of key hardware components and adopting dynamic cooling control strategies, the system can effectively prevent overheating, thereby reducing hardware damage or performance degradation caused by overheating. The heat dissipation fault detection module can identify potential heat dissipation problems at an early stage, allowing the operation and maintenance team to intervene and repair the problem before it worsens, further improving the reliability of the system.

[0125] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0126] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An immersion liquid cooling server circulation heat dissipation fault detection system, characterized in that: include: Data collection module deploys data collection nodes, determines the key parts of the server, including CPU, GPU, memory, hard disk, power supply, and deploys temperature sensors in key parts to ensure that data collection nodes obtain temperature information. In the collection of temperature data, by establishing data collection nodes, the designed data collection nodes are deployed in various key parts of the server, and the priority is continuously adjusted according to the change of server hardware power. The collected temperature is predicted, and the prediction task is divided into multiple small segments. Each segment is processed by a sub-neural network, and the results are summarized and finally analyzed according to the temperature prediction results; Initialize the data acquisition parameters, design the data acquisition frequency to be 1 second / time, the acquisition range to be 0-160 degrees Celsius, and the acquisition accuracy to be 0.1 degrees Celsius; Adjust the collection priority of hardware power, monitor the power consumption of server hardware, and when the power rises, adjust it to the first level. The data of the first level will be processed first, followed by other server hardware. The priority is adjusted according to the power change. The data of other hardware will be lowered accordingly. Set an early warning mechanism according to the change of power consumption. When the power consumption exceeds the set threshold, trigger the corresponding alarm and make corresponding adjustments, and increase the frequency of data collection; The data collection node collects temperature data, stores the collected temperature data in the database, and updates the temperature data regularly; The temperature prediction module divides the collected data into overlapping segments of 100 data points each, initializes an autoencoder neural network, and sets the architecture of the autoencoder neural network, including the input layer, encoding layer, decoding layer, and output layer; Initialize the autoencoder neural network with 100 input neurons, 50 encoding neurons, and 100 output neurons. Input each small segment into the autoencoder neural network, which will learn the potential representation of the temperature data in the small segment, reconstruct the small segment through the decoder layer, and compare it with the original small segment. The prediction error is used to update the weights of the autoencoder neural network; The training goal of the autoencoder neural network is to minimize the prediction error, the mean square error between the original segment and the reconstructed segment. Mean square error = (1 / n)*Σ(x_i-x_i')^2; Where n is the number of data points in the segment; x_i is the i-th data point in the original segment; x_i' is the i-th data point in the reconstructed segment; the autoencoder neural network updates its weights through the back-propagation algorithm to minimize the mean square error; A heat dissipation fault detection module obtains a predicted temperature from a temperature prediction module; Get the actual temperature from the data acquisition module, calculate the difference between the predicted temperature and the actual temperature, set the threshold, the threshold value is 3 degrees Celsius, the calculation formula is: |Predicted temperature - actual temperature|>threshold; Dynamic cooling control module, which monitors the power consumption of server hardware and adjusts cooling system parameters based on the calculated heat output; System management module, real-time monitoring of system status, including data acquisition node status, temperature prediction model performance, cooling system performance, and server temperature; Manage data collection nodes, add, delete or modify data collection nodes, configure data collection parameters, monitor the health of data collection nodes, and configure prediction models and intervention measures; Configure the temperature prediction model, including training data, model architecture, and training hyperparameters; Configure intervention measures, including abnormal situation thresholds and trigger actions.

2. According to claim 1, a circulating heat dissipation fault detection system for an immersion liquid cooling server is characterized in that: The data acquisition node periodically obtains the data from the temperature sensor, stores the collected temperature data in the database, updates the data regularly, and adjusts the priority of data acquisition according to the change of hardware power. The exponentially weighted moving average algorithm can be expressed as: P(t) = (1-α) × P(t-1) + α × P new; Where P(t) is the estimated power at time t, Pnew is the new power sample value, and α is the weighting coefficient.

3. According to claim 1, a circulating heat dissipation fault detection system for an immersion liquid cooling server is characterized in that: Before implementing the temperature prediction module, the collected data needs to be preprocessed. The preprocessing steps include data cleaning, normalization, and data segmentation; Data cleaning: dealing with outliers or missing values ​​in the data; Normalization: scaling the data to a similar range; Data segmentation: Divide the continuous data into overlapping segments of 100 data points each and input them into the autoencoder neural network.

4. The immersion liquid cooling server circulation heat dissipation fault detection system according to claim 3 is characterized in that: Input layer: contains 100 neurons, which are used to receive the original small segment temperature data; Encoding layer: contains 50 neurons, which are used to compress input data into potential representations; Decoding layer: contains 100 neurons, used to decode the latent representation into raw data; Output layer: Same as the input layer, containing 100 neurons, used to reconstruct small segments and initialize the weights and biases of the autoencoder neural network. The weights and biases are randomly initialized to start training.

5. The immersion liquid cooling server circulation heat dissipation fault detection system according to claim 4, characterized in that: During the training process, each small segment of temperature data is input into the autoencoder neural network. The goal of the training is to minimize the difference between the reconstructed data and the original data. The training process can be divided into the following steps: S1, forward propagation: input the original small segment of temperature data into the network, generate potential representation through the encoder layer, and reconstruct the data through the decoder layer; S2, calculate the loss: calculate the mean square error between the reconstructed data and the original data as the prediction error; S3, back propagation: Use the back propagation algorithm to update the weights and biases of the network to minimize the prediction error; S4, repeated training: Repeat the above steps until the preset number of training times is reached or converges to a satisfactory error level.

6. The immersion liquid cooling server circulation heat dissipation fault detection system according to claim 5, characterized in that: Forward propagation calculation formula: h = f (Wx + b); x is the input data; W is the weight matrix; b is the bias vector; f is the activation function; h is the output of the encoding layer; In the autoencoder neural network, the back propagation calculation formula is: Where L is the loss function; and They are the gradients of the loss function with respect to weights and biases respectively; is the gradient of the loss function with respect to the output of the encoding layer; In the autoencoder neural network, the commonly used loss function is used to measure the difference between the reconstructed data and the original data. The calculation formula of the loss function is as follows: x i It is the original data; is the reconstruction data; N is the number of data points.

7. The immersion liquid cooling server circulation heat dissipation fault detection system according to claim 1, characterized in that: A threshold is set to determine when a heat dissipation fault is considered to have occurred. The threshold is 3 degrees Celsius, that is, if the difference between the predicted temperature and the actual temperature is greater than 3 degrees Celsius, a heat dissipation fault is considered to have occurred. The difference is greater than the threshold, triggering the detection of a heat dissipation fault.

8. The immersion liquid cooling server circulation heat dissipation fault detection system according to claim 7, characterized in that: Based on the results of fault detection, feedback and processing are performed to alert the operator, record the fault event, and take automatic or manual measures to adjust the cooling system.

9. The immersion liquid cooling server circulation heat dissipation fault detection system according to claim 1, characterized in that: The power consumption data is used to calculate the heat generated by the server. The heat calculation formula is: Q = P × t; Where Q is the heat generated; P is power consumption; t is the time.

10. The immersion liquid cooling server circulation heat dissipation fault detection system according to claim 1, characterized in that: Dynamically adjust the parameters of the cooling system, including fan speed and water pump flow, based on the calculated heat output; In control theory, the feedback control algorithm, PID control algorithm, continuously adjusts the control variable to make the controlled variable close to the set value; Feedback control algorithm calculation formula: U(t) is the controlled variable; e(t) is the error; K p is the ratio; K i is the integral; K d is the differential coefficient.

Images