Multi-node cooperative monitoring and intelligent regulation and control method and system based on array server

Abstract

The invention discloses a multi-node cooperative monitoring and intelligent regulation and control method and system based on an array server, and relates to the technical field of intelligent thermal management, and the method comprises the steps: collecting node operation state data of the array server and machine room environment data, and generating a standardized data set; calculating a temperature change trend and a utilization rate change trend based on the standardized data set, and obtaining a joint weight coefficient by combining an environment parameter distribution weight; predicting the processor temperature by using the joint weight coefficient to obtain a processor predicted temperature, and generating a target fan rotating speed value based on the processor predicted temperature and the joint weight coefficient; inputting the target fan rotating speed value into a fan execution mechanism, collecting actual operation data, and comparing the actual operation data with a prediction result to calculate a prediction error and an execution error; according to the prediction error and the execution error, the weight distribution parameters and the prediction model parameters are dynamically corrected, an updated control strategy is obtained, and self-adaptive adjustment of the fan is achieved. According to the invention, stable operation and energy efficiency optimization in the heat dissipation process of the machine room are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent thermal management, in particular to a multi-node cooperative monitoring and intelligent control method and system based on an array server. BACKGROUND

[0002] With the continuous improvement of computing node performance and the rapid expansion of data center scale, the problem of computer room heat dissipation has become a key link affecting system stability and energy efficiency performance. Traditional heat dissipation control methods rely on static thresholds or single sensor parameters for adjustment, and do not fully consider the coupling effect of processor load dynamic changes and environmental factors. In recent years, fan speed control methods based on predictive control and adaptive optimization have gradually emerged, which realize the forward-looking prediction and control of processor temperature by introducing time series analysis, trend extraction and machine learning algorithms.

[0003] However, the existing methods still have two limitations: first, the existing heat dissipation control methods often rely on processor temperature curves or fan feedback signals for single-point prediction and adjustment, and do not realize the multi-source information fusion of environmental temperature, humidity and processor, graphics processor load trends. Second, the adaptive ability of the existing method is limited, mostly only through static parameters or fixed thresholds for control optimization, lacking a dynamic correction mechanism for prediction error and execution error. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a multi-node cooperative monitoring and intelligent control method based on an array server to solve the problems of insufficient prediction accuracy caused by single data source in existing computer room heat dissipation control methods and how to realize fan prediction and control and energy efficiency optimization through environmental-load joint weight.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a multi-node cooperative monitoring and intelligent control method based on an array server, which comprises collecting node running state data and computer room environment data of the array server, standardizing preprocessing, and obtaining a standardized data set;

[0008] Through the standardized data set, the processor temperature change trend and the graphics processor utilization rate change trend are calculated to obtain trend data;

[0009] Based on the trend data and environmental parameters, weight distribution is performed to obtain joint weight coefficients;

[0010] The processor predicted temperature is obtained by predicting the processor temperature through the joint weight coefficients and the standardized data set.

[0011] The fan rotating speed is obtained through the control signal generation process of the temperature prediction and the joint weight coefficient by the processor;

[0012] The fan rotating speed is input to the fan actuator to obtain the actual fan rotating speed and the node temperature;

[0013] The joint weight distribution strategy is adaptively corrected through the actual fan rotating speed and the node temperature to obtain an updated control strategy, and the multiple nodes are cooperatively regulated through the updated control strategy to continuously drive the fan actuator to adaptively adjust under multiple environmental and load conditions.

[0014] As a preferred scheme of the multi-node cooperative monitoring and intelligent regulation method based on the array server, the node running state data and the machine room environment data of the array server are collected, standardized pretreated, and standardized data sets are obtained, and the specific steps are as follows,

[0015] The processor temperature, the graphics processor utilization rate, the environmental temperature, the environmental humidity and the air flow rate data are collected to obtain the node running state data and the machine room environment data of the array server.

[0016] The node running state data and the machine room environment data of the array server are time-stamped to ensure the time sequence consistency of the data;

[0017] The marked node running state data and the machine room environment data of the array server are summarized to obtain the initial data set with uniform time sequence labels.

[0018] As a preferred scheme of the multi-node cooperative monitoring and intelligent regulation method based on the array server, the node running state data and the machine room environment data of the array server are collected, standardized pretreated, and standardized data sets are obtained, and the specific steps are as follows,

[0019] The initial data set is filtered to obtain a filtered data set;

[0020] The filtered data set is normalized to obtain normalized data, and the normalized data is sequentially processed according to the collection time to obtain the standardized data set with uniform time sequence labels.

[0021] As a preferred scheme of the multi-node cooperative monitoring and intelligent regulation method based on the array server, the node running state data and the machine room environment data of the array server are collected, standardized pretreated, and standardized data sets are obtained, and the specific steps are as follows,

[0022] Extract time series data in the standardized data set, calculate the change rate by using the time series trend extraction method, obtain the temperature trend value and the utilization trend value;

[0023] The temperature trend value and the utilization trend value are summarized to obtain trend data.

[0024] As a preferred scheme of the multi-node cooperative monitoring and intelligent control method based on the array server, wherein: based on the trend data and the environmental temperature and humidity parameters, the weight distribution is performed to obtain a joint weight coefficient, and the specific steps are as follows,

[0025] The trend data, the environmental temperature and the environmental humidity parameters are fused to obtain a comprehensive input data set;

[0026] Based on the comprehensive input data set, the processor temperature trend weight and the graphics processor utilization trend weight are obtained by using a trend calculation function, and the environmental temperature, the environmental humidity and the PDU current parameters are converted into environmental weights by using an environmental response function to obtain a joint weight coefficient.

[0027] As a preferred scheme of the multi-node cooperative monitoring and intelligent control method based on the array server, wherein: the processor temperature is predicted by the joint weight coefficient and the standardized data set to obtain a processor predicted temperature, and the specific steps are as follows,

[0028] The joint weight coefficient and the standardized data set are predicted and calculated to obtain the processor predicted temperature, and the processor predicted temperature is output as a temperature sequence.

[0029] As a preferred scheme of the multi-node cooperative monitoring and intelligent control method based on the array server, wherein: the fan rotating speed is obtained by performing a control signal generation process on the processor predicted temperature and the joint weight coefficient, and the specific steps are as follows,

[0030] The processor predicted temperature and the joint weight coefficient are fused and processed to establish a correlation between the temperature change trend and the environmental-load factor, and the target fan rotating speed value is calculated.

[0031] As a preferred scheme of the multi-node cooperative monitoring and intelligent control method based on the array server, wherein: the fan rotating speed is input to a fan execution mechanism to obtain an actual fan rotating speed and a node temperature, and the specific steps are as follows,

[0032] The target fan rotating speed value is input to the fan execution mechanism to adjust the actual rotating speed of the fan;

[0033] The fan actual rotating speed, the processor temperature and the graphics processor temperature are collected to obtain running state data.

[0034] As a preferred scheme of the array server-based multi-node cooperative monitoring and intelligent regulation method, the joint weight distribution strategy is adaptively corrected through the actual fan rotating speed and the node temperature, an updated control strategy is obtained, the multi-node cooperative regulation is performed through the updated control strategy, and the fan actuator is continuously driven to adaptively adjust under various environmental and load conditions.

[0035] The running state data is compared with the processor predicted temperature and the target rotating speed to obtain a prediction error and an execution error.

[0036] Based on the prediction error and the execution error, the weight distribution strategy and the prediction model parameters are dynamically corrected to obtain an updated control strategy.

[0037] The multi-node cooperative regulation is performed through the updated control strategy, and the fan actuator is continuously driven to adaptively adjust under various environmental and load conditions.

[0038] In a second aspect, the application provides an array server-based multi-node cooperative monitoring and intelligent regulation system, which comprises a data acquisition module, a trend calculation module, a weight distribution module, a temperature prediction module, a signal generation module, an execution feedback module and a cooperative regulation module.

[0039] The trend calculation module calculates the processor temperature change trend and the graphics processor utilization change trend through the standardized data set to obtain trend data.

[0040] The weight distribution module performs weight distribution based on the trend data and the environmental parameters to obtain a joint weight coefficient.

[0041] The temperature prediction module predicts the processor temperature through the joint weight coefficient and the standardized data set to obtain a processor predicted temperature.

[0042] The signal generation module performs control signal generation processing through the processor predicted temperature and the joint weight coefficient to obtain a fan rotating speed.

[0043] The execution feedback module inputs the fan rotating speed to the fan actuator to obtain an actual fan rotating speed and a node temperature.

[0044] The cooperative regulation module adaptively corrects the joint weight distribution strategy through the actual fan rotating speed and the node temperature, obtains an updated control strategy, performs multi-node cooperative regulation through the updated control strategy, and continuously drives the fan actuator to adaptively adjust under various environmental and load conditions.

[0045] The application has the beneficial effects that: by collecting node running state data and machine room environment data of the array server and performing standardized preprocessing, data time sequence consistency and comparability are realized, and the accuracy of subsequent prediction and control process is ensured; by calculating joint weight coefficients based on trend data and environment temperature and humidity parameters, processor temperature prediction and dynamic adaptation of environment-load characteristics are realized, and prediction accuracy and control sensitivity are improved; by dynamically correcting weight distribution strategy and prediction model parameters based on prediction error and execution error, adaptive optimization of the control strategy is realized, and stability and energy efficiency improvement of the machine room heat dissipation process are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0047] Fig. 1 The flow chart of the multi-node cooperative monitoring and intelligent regulation method based on the array server.

[0048] Fig. 2 The flow chart of the collection and preprocessing.

[0049] Fig. 3 The flow chart of the weight distribution and temperature prediction.

[0050] Fig. 4 The flow chart of the target fan speed and closed-loop correction. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.

[0052] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0053] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0054] REFERENCE Figs. 1-4For an embodiment of the present application, the embodiment provides a multi-node cooperative monitoring and intelligent control method based on an array server, comprising the following steps:

[0055] S1: Collect node running state data and machine room environment data of the array server, perform standardized preprocessing, and obtain a standardized data set.

[0056] The node running state data and machine room environment data of the array server are collected, including SoC core temperature, memory bandwidth utilization, processor temperature, graphics processor utilization, environment temperature, environment humidity, and cabinet PDU current data.

[0057] Further, by customizing a heterogeneous node perception module in the BMC, SoC core temperature and memory bandwidth utilization are collected in real time using physical channels, temperature sensors are fixedly arranged on the surface of the processor shell of the computing node to obtain processor temperature data through real-time detection, a performance monitoring interface is called in the instruction scheduling unit of the graphics processor to collect graphics processor utilization data, temperature sensors are arranged in the air circulation area of the machine room to collect environment temperature data, and humidity sensors are arranged at the same collection point to collect environment humidity data. The SoC core temperature, memory bandwidth utilization, processor temperature data, graphics processor utilization data, environment temperature data, and environment humidity data are combined to obtain the node running state data and machine room environment data of the array server.

[0058] The node running state data and machine room environment data of the array server are time-stamped to ensure the time sequence consistency of the data.

[0059] Further, the node running state data and machine room environment data of the array server are respectively input into a timestamp generator to generate node running state data of the array server with a timestamp and machine room environment data with a timestamp. The node running state data of the array server with a timestamp and the machine room environment data with a timestamp are compared to verify whether the timestamp interval is within a unified sampling period. Through comparison and verification, the node running state data of the array server with a timestamp and the machine room environment data with a timestamp are synchronized to ensure the time sequence consistency of the data.

[0060] The marked node running state data and machine room environment data of the array server are summarized to obtain an initial data set with a unified time sequence label.

[0061] Further, the node running state data of the array server with a timestamp and the data center environment data with a timestamp are input to the cache area of the data aggregation processor, the node running state data of the array server and the data center environment data are compared in the cache area according to the timestamp sequence, data records collected at the same time point or adjacent time points are identified, the data aggregation processor combines the node running state data of the array server and the data center environment data after the timestamp alignment according to the preset time alignment rule to form a data pair with a unified time sequence label, and the data pair is continuously arranged in the time sequence to obtain an initial data set with a unified time sequence label.

[0062] The initial data set is filtered to obtain a filtered data set.

[0063] Further, the initial data set with a unified time sequence label is input to the filter processor for noise suppression processing, and a filtering operation is performed to obtain a filtered data set.

[0064] The filtered data set is normalized to obtain normalized data, and the normalized data is sequentially processed according to the collection time to obtain a standardized data set with a unified time sequence label.

[0065] Further, the filtered data set is normalized to obtain normalized data, and the normalized data is arranged in the collection time sequence to obtain serialized data with the collection time sequence, and a unified time sequence label is superimposed to obtain a standardized data set with a unified time sequence label.

[0066] S2: The trend data is obtained by calculating the processor temperature change trend and the graphics processor utilization rate change trend based on the standardized data set.

[0067] The time sequence data is extracted from the standardized data set, the change rate is calculated by using a time sequence trend extraction method, and the temperature trend value and the utilization rate trend value are obtained.

[0068] Further, the temperature time sequence data and the graphics processor utilization rate time sequence data are extracted from the standardized data set, the change rate is calculated by using a time sequence trend extraction method, and the temperature trend value and the utilization rate trend value are obtained.

[0069] The temperature trend value and the utilization rate trend value are aggregated to obtain the trend data.

[0070] Further, the temperature trend value and the utilization rate trend value both have a unified time index, and under the same time index t, the temperature trend value T t and the utilization rate trend value U tPairing, the paired temperature trend value and the utilization trend value are combined in the form of a binary vector to form trend data, denoted as:

[0071] D t = {T t , U t}.

[0072] S3: Based on the trend data and the environmental temperature and humidity parameters, weight distribution is performed to obtain a joint weight coefficient.

[0073] The trend data is fused with the environmental temperature and humidity parameters in the computer room environment data to obtain a comprehensive input data set.

[0074] Further, the trend data is read point by point according to the time index to obtain a trend data sequence with a time index. Under the same time index, the environmental temperature parameter and the environmental humidity parameter are read point by point and correspond to the trend data. Under the unified time index, the trend data, the environmental temperature parameter and the environmental humidity parameter are fused to obtain a comprehensive input data set.

[0075] Based on the comprehensive input data set, the processor temperature trend weight and the graphics processor utilization trend weight are obtained by using a trend calculation function. An environmental state matrix is constructed by using the environmental temperature, the environmental humidity and the cabinet PDU current parameter. The environmental state matrix is converted into an environmental weight by using an environmental response function to obtain a joint weight coefficient.

[0076] Further, the SoC core temperature and the memory bandwidth utilization rate are normalized before entering the trend calculation. The comprehensive input data set is input into the trend calculation function to calculate the processor temperature trend weight W temp (t), and the expression is:

[0077]

[0078] Among them, represents the processor temperature change trend value adjusted by the correction factor, represents the graphics processor utilization rate change trend value adjusted by the correction factor, T (t) represents the processor temperature change trend value under the time index t, U trend (t) represents the graphics processor utilization rate change trend value under the time index t, represents the exponential moving average value of the processor temperature trend calculated by the window length k, represents the weight coefficient of the processor temperature instantaneous trend, represents the weight coefficient of the processor temperature smooth trend, and k represents the window length of the exponential moving average, Z SoC (t) represents the normalized SoC core temperature, Z MEM (t) represents the normalized memory bandwidth utilization rate, represents a SoC core temperature correction factor, represents a memory bandwidth utilization correction factor.

[0079] It should be noted that, and In the historical running data, the processor temperature, the graphics processor utilization, the SoC core temperature and the memory bandwidth utilization are selected respectively to construct a cost function, the square integral of the reference temperature of the processor temperature and the square sum of the fan instruction change are used as the optimization objective, the initial value range of the correction factor is determined in the bounded domain by using coarse grid search, and then the Nelder-Mead derivative-free optimization method is used for iteration and refinement until the cost function converges to the minimum value, and finally the optimal correction factor is obtained, k is cross-validated on the historical processor temperature sequence, the window length that achieves the best balance between the minimum prediction error or the smoothness and real-time of the temperature fluctuation is selected, and α1 and α2 are normalized feature data obtained by selecting a time window containing low load to high load, low environment temperature to high environment temperature as a calibration data set from the standardized data set, and the temperature trend value and the exponential moving average in the calibration data set are unified in feature scale by using the zero mean unit variance method, the processor temperature trend weight is obtained by using the standardized features in the trend calculation function, and the square integral of the difference between the processor temperature and the reference temperature and the square sum of the fan instruction change are used as the cost function J in the playback process of the calibration data set, and the expression is:

[0080]

[0081] wherein, T c (t) represents the processor temperature, T ref represents the safe running temperature, Δs cmd (t) represents the change amount of the fan speed instruction at the time index t, and λ represents the balance coefficient.

[0082] It should be noted that T ref is the safe running temperature specified by the processor manufacturer, and λ is used to balance the temperature control accuracy and the fan speed smoothness, and λ is determined by observing the compromise point of the temperature deviation and the fan energy consumption through multiple experiments.

[0083] The minimum cost starting point is obtained by coarse grid search in the bounded domain, and then the Nelder-Mead derivative-free optimization is used for optimization, which is represented as:

[0084] α1∈[L1,U1],α2∈[L2,U2];

[0085] Wherein, the coarse grid step example is 0.25, L1 represents the lower bound of alpha1, U1 represents the upper bound of alpha1, L2 represents the lower bound of alpha2, and U2 represents the upper bound of alpha2.

[0086] It should be noted that L1, U1, L2 and U2 are obtained by statistical analysis of a large number of historical data of fan operation and processor temperature control. If alpha1 is too large, the fan will be adjusted frequently, and if alpha1 is too small, the temperature will be difficult to control. If alpha2 is too large, the fan speed will change too slowly, and if alpha2 is too small, the fan will be switched frequently.

[0087] To avoid excessive dependence on a single feature, a coefficient ratio constraint and a light regularization are added, which is expressed as:

[0088]

[0089] Wherein, alpha1 >= alpha2 in order to reflect that the instantaneous trend influence is not lower than the smooth trend, and eta represents the regularization coefficient.

[0090] The average temperature deviation and command variance are calculated in the sliding window of the online running and small step correction is performed, which is expressed as:

[0091]

[0092] Wherein, rho represents the learning rate.

[0093] According to the above process, alpha1 and alpha2 can be finally obtained.

[0094] The comprehensive input data set is input into the trend calculation function, and the graphics processor utilization rate trend weight W util (t) is calculated, and the expression is:

[0095]

[0096] Wherein, U trend (t) represents the graphics processor utilization rate trend value at time index t, beta1 represents the weight coefficient of the quadratic term of the graphics processor utilization rate trend, and beta2 represents the weight coefficient of the positive trend of the graphics processor utilization rate.

[0097] It should be noted that beta1 and beta2 are based on the playback of the calibration data set. After the graphics processor utilization rate trend value is standardized, it is input into the Sigmoid function of the combination of the quadratic term and the linear term. The initial value is obtained by setting the small signal slope and the target point, and then the target function containing the temperature limit constraint and the forward alignment constraint is iteratively optimized to obtain the final value.

[0098] The comprehensive input data set is input into the trend calculation function, and the environment weight W env (t) is calculated, and the expression is:

[0099]

[0100] wherein, T env (t) represents the ambient temperature parameter at time index t, H env (t) represents the ambient humidity parameter at time index t, M env (t) represents the PDU current parameter at time index t, M ref represents the reference PDU current parameter, T ref represents the reference ambient temperature, T max represents the maximum allowed ambient temperature, H ref represents the reference ambient humidity, H max represents the maximum allowed ambient humidity, M max represents the maximum allowed PDU current, γ1 represents the weight coefficient of the ambient temperature factor, γ2 represents the weight coefficient of the ambient humidity factor, and γ3 represents the weight coefficient of the PDU current.

[0101] It should be noted that γ1, γ3 and γ2 are coefficients obtained by fitting the influence relationship of the ambient temperature, the ambient humidity and the PDU current on the fan speed in the calibration data set, the initial value is obtained by regression calculation, and the final value is determined by optimization of the objective function with regularization constraint.

[0102] The processor temperature trend weight, the graphics processor utilization trend weight and the environment weight are fused to calculate a joint weight coefficient W joint (t), and the expression is:

[0103]

[0104] wherein, δ1 represents the distribution coefficient of the processor temperature trend weight, δ2 represents the distribution coefficient of the graphics processor utilization trend weight, and δ3 represents the distribution coefficient of the environment weight.

[0105] It should be noted that δ1, δ2 and δ3 are balance coefficients obtained by jointly optimizing the temperature prediction error and the fan energy consumption, the initial value is set by experience, and the final value is determined by multi-objective optimization convergence.

[0106] S4: predicting the processor temperature by the joint weight coefficient and the standardized data set to obtain a processor predicted temperature.

[0107] The joint weight coefficient and the standardized data set are predicted and calculated to obtain the processor predicted temperature, and the processor predicted temperature is output as a temperature sequence.

[0108] Further, the joint weight coefficient, the processor temperature standardized value and the graphics processor utilization standardized value in the standardized data set are substituted into the processor temperature prediction increment formula to obtain a processor temperature prediction increment The expression is:

[0109]

[0110] wherein X T (t) represents a normalized value of the processor temperature in the normalized data set, X U (t) represents a normalized value of the GPU utilization in the normalized data set, represents a predicted temperature of the processor, θ1 represents a weight coefficient of the normalized value of the processor temperature to the predicted temperature change, and θ2 represents a weight coefficient of the normalized value of the GPU utilization to the predicted temperature change.

[0111] It should be noted that θ1 and θ2 are determined by bounded optimization of initial values obtained through playback of the calibration data set and minimization of the objective function, balancing the prediction accuracy and fan instruction smoothness.

[0112] The processor temperature prediction increment is added to the processor predicted temperature at the previous time to obtain a new processor predicted temperature is expressed as:

[0113]

[0114] wherein Δ represents a time step.

[0115] S5: performing a control signal generation process on the processor predicted temperature and the joint weight coefficient to obtain a fan rotation speed.

[0116] The processor predicted temperature and the joint weight coefficient are fused to establish a correlation between the temperature change trend and the environmental-load factor, and a target fan rotation speed value is calculated.

[0117] Further, the processor predicted temperature change rate is calculated by processing the processor predicted temperature, and the processor predicted temperature change rate V(t) is obtained, which is expressed as:

[0118]

[0119] The processor predicted temperature change rate and the joint weight coefficient are fused to obtain a correlation value between the temperature change trend and the environmental-load factor, which is expressed as:

[0120]

[0121] wherein R(t) represents a correlation value between the temperature change trend and the environmental-load factor, W j (t) represents a value of the joint weight coefficient at time t, k1 represents a sensitivity of controlling the processor predicted temperature change rate, and k2 represents a sensitivity of controlling the temperature deviation normalized quantity.

[0122] It should be noted that k1 and k2 are fitted by minimizing the deviation between the predicted correlation value and the target response during the calibration dataset playback process using the optimization objective function.

[0123] The correlation value between the temperature change trend and the environment-load factor is mapped to the processor predicted temperature to calculate the target fan speed value s(t), expressed as:

[0124]

[0125] clamp(x,0,1)=min(1,max(0,x));

[0126] where s min represents the lower limit of the target fan speed value, s max represents the upper limit of the target fan speed value, η1 represents the weight of the processor predicted temperature in the target fan speed value mapping, η2 represents the weight of the correlation value between the temperature change trend and the environment-load factor in the target fan speed value mapping, and clamp represents a limiting function to prevent abnormal control instructions caused by exceeding the set range.

[0127] It should be noted that η1 and η2 are obtained by minimizing the mean square error between the predicted target fan speed value and the recorded fan speed instruction on the calibration dataset and using the gradient descent method for iterative optimization.

[0128] In a multi-node scenario, the target fan speed values calculated by each node are shared through inter-node communication to form a collaborative control instruction, ensuring balanced overall heat dissipation efficiency.

[0129] S6: Input the fan speed to the fan actuator to obtain the actual fan speed and node temperature.

[0130] Input the target fan speed value to the fan actuator to adjust the actual fan speed.

[0131] Further, the target fan speed value is first differentiated with the last actual fan speed to obtain a speed error, then a feedforward-feedback fusion is calculated using the speed error and the target fan speed value to obtain a control duty cycle original value, and the control duty cycle is obtained by slope limiting and upper and lower boundary truncation, finally the control duty cycle is converted to a driving voltage and input to the fan actuator to adjust the actual fan speed. In the multi-node collaborative control mode, the fan actuators of multiple nodes are uniformly deployed, the running states of each node are comprehensively considered, and global heat dissipation energy efficiency optimization is ensured.

[0132] It should be noted that the fan actuator is a device that receives control instructions and directly drives the fan to adjust the speed.

[0133] Collect the actual fan speed, current processor temperature, and graphics processor temperature to obtain operating status data.

[0134] Furthermore, the fan speed sensor is continuously read during the sampling period Δ to obtain the actual fan speed. At the same time index t, the processor temperature sensor code value and the graphics processor temperature sensor code value are read to obtain the current processor temperature and graphics processor temperature. The current processor temperature, graphics processor temperature and the actual fan speed are aligned and summarized into operating status data.

[0135] S7: Adaptively adjust the joint weight allocation strategy based on actual fan speed and node temperature to obtain an updated control strategy.

[0136] The operating status data is compared with the processor's predicted temperature and target speed to obtain the prediction error and execution error.

[0137] Furthermore, the operating status data is aligned with the processor's predicted temperature and target rotational speed at time index t to obtain a comparison input set. Based on the comparison input set, the prediction error e is calculated. T (t), the expression is:

[0138]

[0139] Among them, T proc (t) represents the processor temperature, ε T k represents the temperature dead zone threshold. T This represents the positive temperature deviation amplification factor, 1 condition This indicates a conditional indicator function. If the condition is true, the value is 1; otherwise, the value is 0.

[0140] It should be noted that ε T The threshold value, k, is determined by gradually increasing it and observing whether the fan speed oscillation significantly decreases. The minimum value that still ensures stable fan speed is then selected. T In the calibration dataset, when T proc When (t)>T(t), gradually increase k T Observe the fan speed response speed and temperature recovery speed to determine the optimal k. T .

[0141] The execution error e is calculated by comparing the input set and the prediction error. S (t), the expression is:

[0142]

[0143] Among them, S fan (t) represents the actual fan speed, ε S k represents the engine speed dead zone threshold.S Represents the execution deviation coupling amplification coefficient.

[0144] It should be noted that ε S is obtained by gradually increasing the threshold, observing the fan regulation frequency and noise level. When the fan jitter is significantly reduced, the minimum effective value is selected, and k S is in the calibration dataset. When S fan (t) < s(t), k is gradually increased S , and the time and stability of the fan speed tracking the target are observed to determine the optimal value.

[0145] Based on the prediction error and execution error, the weight allocation strategy and prediction model parameters are dynamically corrected to obtain the updated control strategy.

[0146] Furthermore, through the prediction error and execution error, the comprehensive error E(t) is calculated and expressed as:

[0147] E(t) = ε T (t) + ε S (t);

[0148] Through the comprehensive error, each weight allocation parameter is corrected, and the expression is:

[0149]

[0150] Among them, f i (t) represents the corresponding trend value (such as temperature trend value, utilization trend value, ambient temperature parameter), and η δ represents the learning rate of weight update.

[0151] It should be noted that η δ is the initial value obtained by debugging on the calibration dataset and is obtained by dynamically adjusting using a decay strategy during actual operation.

[0152] Through the prediction error and execution error, the comprehensive loss function L(t) is obtained, and the expression is:

[0153]

[0154] The gradient of the prediction model parameter Θ(t) is calculated, and the expression is:

[0155]

[0156] Among them, represents the gradient vector of the comprehensive loss function L(t) with respect to the parameter Θ(t) at the time index t, represents taking the partial derivative of each parameter in Θ(t) respectively to obtain the sensitivity of the comprehensive loss function with respect to each parameter.

[0157] The prediction model parameters are updated using gradient descent, expressed as follows:

[0158]

[0159] Where, η θ This represents the learning rate used to update the parameters of the prediction model.

[0160] Substitute the new prediction model parameter Θ(t+1) into the prediction temperature calculation formula to obtain the new processor prediction temperature. Combine the new processor prediction temperature with the joint weighting coefficient to obtain the new target fan speed value and obtain the new control strategy.

[0161] S8: Through the updated control strategy, in the multi-node collaborative control mode, the fan actuator is continuously driven to maintain adaptive adjustment under different operating conditions, so that temperature management and energy efficiency performance can achieve dynamic balance, thereby realizing stable operation and energy consumption optimization of the overall heat dissipation process.

[0162] During the generation of the updated control strategy, continuous monitoring of prediction and execution errors enables dynamic correction of the weight allocation strategy and prediction model parameters. This allows the control logic to continuously optimize with changes in operating status and environmental conditions. In a multi-node collaborative control mode, multiple computing nodes run the updated control strategy on different device units. Nodes share operating status data and environmental parameters in real time, and synchronized adjustments ensure that each node maintains consistent control direction within the same timescale, thus forming a unified fan speed regulation and temperature management action at the overall level. After correction, the control strategy is simultaneously applied to the fan regulation process of all nodes, and continuous feedback accumulates operating experience, enabling the entire multi-node environment to maintain an adaptive closed-loop control mechanism that ensures heat dissipation efficiency and equipment stability even under load fluctuations and environmental changes.

[0163] This embodiment also provides a computer device applicable to the multi-node collaborative monitoring and intelligent control method based on an array server, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the multi-node collaborative monitoring and intelligent control method based on an array server as proposed in the above embodiment.

[0164] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0165] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the multi-node collaborative monitoring and intelligent control method based on an array server as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0166] In summary, this invention achieves data time-series consistency and comparability by: collecting node operating status data and data center environmental data from the array server and performing standardized preprocessing, thus ensuring the accuracy of subsequent prediction and control processes; dynamically adapting processor temperature prediction to environmental-load characteristics by calculating joint weighting coefficients based on trend data and environmental temperature and humidity parameters, thereby improving prediction accuracy and control sensitivity; and dynamically correcting the weight allocation strategy and prediction model parameters based on prediction and execution errors, thereby achieving adaptive optimization of the control strategy and ensuring the stability and energy efficiency improvement of the data center heat dissipation process.

[0167] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for multi-node collaborative monitoring and intelligent control based on an array server, characterized in that: Comprising, Collecting node running state data and machine room environment data of the array server, performing standardized preprocessing, and obtaining a standardized data set; Through the standardized data set, the processor temperature change trend and the graphics processor utilization rate change trend are calculated, and trend data is obtained; Based on the trend data and the environment parameters, weight distribution is performed to obtain a joint weight coefficient; The processor predicted temperature is obtained by predicting the processor temperature through the joint weight coefficient and the standardized data set; The fan speed is obtained by performing control signal generation processing on the processor predicted temperature and the joint weight coefficient; The actual fan speed and node temperature are obtained by inputting the fan speed to the fan actuator; The updated control strategy is obtained by adaptively correcting the joint weight distribution strategy through the actual fan speed and node temperature, and the multi-node collaborative regulation is performed through the updated control strategy to continuously drive the fan actuator to maintain adaptive adjustment under various environmental and load conditions.

2. The method of claim 1, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The specific steps of collecting node running state data and machine room environment data of the array server are as follows: Collecting processor temperature, graphics processor utilization rate, environment temperature and environment humidity data to obtain node running state data and machine room environment data of the array server; The node running state data and machine room environment data of the array server are time-stamped, and the marked node running state data and machine room environment data of the array server are data-aggregated to obtain an initial data set with a unified time sequence label.

3. The method of claim 2, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The specific steps of performing standardized preprocessing to obtain a standardized data set are as follows: Filtering the initial data set to obtain a filtered data set; The filtered data set is normalized to obtain normalized data, and is sequentially processed according to the collection time to obtain a standardized data set with a unified time sequence label.

4. The method of claim 3, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The specific steps of calculating the processor temperature change trend and the graphics processor utilization rate change trend through the standardized data set to obtain trend data are as follows: Extracting time sequence data from the standardized data set, calculating the change rate using a time sequence trend extraction method to obtain temperature trend values and utilization rate trend values; The temperature trend values and utilization rate trend values are aggregated to obtain trend data.

5. The method of claim 4, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The specific steps of performing weight distribution based on the trend data and the environment parameters to obtain a joint weight coefficient are as follows: Fusing the trend data with the environment temperature and environment humidity parameters to obtain a comprehensive input data set; Based on the comprehensive input data set, the processor temperature trend weight and the graphics processor utilization rate trend weight are obtained using a trend calculation function, and the environment temperature, environment humidity and PDU current parameters are converted into environment weights using an environment response function to obtain a joint weight coefficient.

6. The method of claim 5, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The specific steps of predicting the processor temperature through the joint weight coefficient and the standardized data set to obtain a processor predicted temperature are as follows: The joint weight coefficient and the standardized data set are predicted and calculated to obtain a processor predicted temperature, and the processor predicted temperature is output as a temperature sequence.

7. The method of claim 6, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The specific steps of generating a control signal through the processor predicted temperature and the joint weight coefficient to obtain a fan speed are as follows: The processor predicted temperature is fused with the joint weight coefficient to establish a correlation between the temperature change trend and the environment-load factor, and the target fan rotating speed value is calculated.

8. The method of claim 7, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The fan rotating speed is input to the fan actuator to obtain the actual fan rotating speed and the node temperature, and the specific steps are, The target fan rotating speed value is input to the fan actuator to adjust the actual rotating speed of the fan. The fan actual rotating speed, the processor temperature and the graphics processor temperature are collected to obtain the running state data.

9. The method of claim 8, wherein the array server-based multi-node cooperative monitoring and intelligent regulation method is characterized by: The joint weight distribution strategy is adaptively corrected through the actual fan rotating speed and the node temperature to obtain an updated control strategy, and through the updated control strategy, multi-node collaborative regulation is performed to continuously drive the fan actuator to maintain adaptive adjustment under various environmental and load conditions. The running state data, the processor predicted temperature and the target rotating speed are compared to obtain prediction errors and execution errors; Based on the prediction errors and the execution errors, the weight distribution strategy and the prediction model parameters are dynamically corrected to obtain an updated control strategy; Through the updated control strategy, multi-node collaborative regulation is performed to continuously drive the fan actuator to maintain adaptive adjustment under various environmental and load conditions.

10. An array server-based multi-node cooperative monitoring and intelligent control system based on the file encryption method of any one of claims 1-9. The data acquisition module acquires the node running state data and the computer room environment data of the array server, performs standardized preprocessing to obtain a standardized data set; The trend calculation module calculates the processor temperature change trend and the graphics processor utilization change trend through the standardized data set to obtain trend data; The weight distribution module distributes weights based on the trend data and the environment parameters to obtain joint weight coefficients; The temperature prediction module predicts the processor temperature through the joint weight coefficients and the standardized data set to obtain a processor predicted temperature; The signal generation module generates a control signal through the processor predicted temperature and the joint weight coefficients to obtain a fan rotating speed; The execution feedback module inputs the fan rotating speed to the fan actuator to obtain the actual fan rotating speed and the node temperature; The collaborative regulation module adaptively corrects the joint weight distribution strategy through the actual fan rotating speed and the node temperature to obtain an updated control strategy, and through the updated control strategy, multi-node collaborative regulation is performed to continuously drive the fan actuator to maintain adaptive adjustment under various environmental and load conditions.

Images