CN117170980A

CN117170980A - Early warning method, device, equipment and storage medium for server hardware abnormality

Info

Publication number: CN117170980A
Application number: CN202311433545.7A
Authority: CN
Inventors: 李盛新; 李道童; 陈衍东; 韩红瑞; 艾山彬
Original assignee: Suzhou Metabrain Intelligent Technology Co Ltd
Current assignee: Suzhou Metabrain Intelligent Technology Co Ltd
Priority date: 2023-10-31
Filing date: 2023-10-31
Publication date: 2023-12-05
Anticipated expiration: 2043-10-31
Also published as: CN117170980B

Abstract

The embodiment of the application relates to the technical field of computers, in particular to a server hardware abnormality early warning method, device, equipment and storage medium, aiming at improving the accuracy and efficiency of server bottom hardware fault early warning. The method comprises the following steps: respectively inputting the data of the processing information quantity change rate complaint bureau and the terminal current change rate of each hardware on a server in a preset time period into a pre-trained fuzzy inference model, wherein the fuzzy inference model is obtained by training based on a mixed element heuristic optimization algorithm; the fuzzy inference model obtains a predicted temperature value of the hardware according to the processed information quantity change rate data and the terminal current change rate data; marking the hardware as abnormal hardware under the condition that the difference value between the predicted temperature value and the actual temperature value of the hardware is larger than a preset temperature difference threshold value; and generating early warning information corresponding to the abnormal hardware.

Description

Early warning method, device, equipment and storage medium for server hardware abnormality

Technical Field

The embodiment of the application relates to the technical field of computers, in particular to a server hardware abnormality early warning method, device and equipment and a storage medium.

Background

The server is a high-performance computer for providing various services to the outside, the daily calculation amount of the server is huge, if the bottom hardware of the server is abnormal to cause unexpected temporary halt, the problems of data loss, application interruption and the like can be caused, the task which depends on the server to be carried out is seriously influenced, and the working efficiency is reduced, so that the bottom hardware of the server needs to be monitored in real time, early warning is carried out in time when the hardware abnormality is found, and the normal operation of the server is ensured. In the related art, abnormal early warning is carried out on the bottom hardware of the server through a neural network, so that the method is a relatively efficient early warning mode.

In the related art, the limitation of the algorithm used by the neural network influences the accuracy of the neural network in performing abnormal early warning on the bottom hardware of the server, so that the abnormal early warning can not be well performed on devices with low circuit unit repeatability and small transistor density and scale, and the server can not be effectively prevented from being failed.

Disclosure of Invention

The embodiment of the application provides a server hardware abnormality early warning method, device, equipment and storage medium, aiming at improving the accuracy and efficiency of server bottom hardware fault early warning.

An embodiment of the present application provides a method for early warning of server hardware abnormality, where the method includes:

respectively inputting the processing information quantity change rate data and the terminal current change rate data of each hardware on a server in a preset time period into a pre-trained fuzzy inference model, wherein the fuzzy inference model is obtained by training based on a mixed element heuristic optimization algorithm;

the fuzzy inference model obtains a predicted temperature value of the hardware according to the processed information quantity change rate data and the terminal current change rate data;

marking the hardware as abnormal hardware under the condition that the difference value between the predicted temperature value and the actual temperature value of the hardware is larger than a preset temperature difference threshold value;

and generating early warning information corresponding to the abnormal hardware.

Optionally, the training step of the fuzzy inference model includes:

collecting training samples to obtain a training sample data set, wherein the training samples comprise processing information quantity change rate data, end current change rate data and actual measurement temperature values of each piece of hardware on the server in a preset time period;

inputting the training sample data set into a fuzzy inference model to be trained;

Carrying out parameter initialization on the fuzzy inference model to be trained according to the training sample by using a fuzzy clustering method to obtain an initial fuzzy inference model;

setting initial parameters of the mixed element heuristic optimization algorithm according to the parameters of the initial fuzzy inference model;

and carrying out parameter optimization on the initial fuzzy inference model through the mixed element heuristic optimization algorithm to obtain the trained fuzzy inference model.

Optionally, the method further comprises:

and setting a corresponding temperature difference threshold according to a preset temperature difference threshold setting rule.

Optionally, the setting the corresponding temperature difference threshold according to a preset temperature difference threshold setting rule includes:

determining root mean square error between a predicted temperature value obtained by the fuzzy inference model and a corresponding measured temperature value;

and setting the corresponding temperature difference threshold according to the root mean square error.

Optionally, the using a fuzzy clustering method, according to the training sample, performs parameter initialization on the fuzzy inference model to be trained to obtain an initial fuzzy inference model, including:

randomly selecting a first preset number of initial clustering centers from the training samples;

According to a preset optimization target, carrying out iterative optimization on the initial clustering center to obtain an optimized clustering center and corresponding clustering parameters;

according to the optimized cluster center, generating a corresponding network structure by cluster division;

and assigning the network parameters of the fuzzy inference model according to the clustering parameters to obtain the initial fuzzy inference model.

Optionally, the optimization objective is to minimize a weighted sum of the distance from the data point in the training sample to its corresponding cluster center and the membership.

Optionally, the setting initial parameters of the hybrid heuristic optimization algorithm according to the parameters of the initial fuzzy inference model includes:

determining the number of particles of the particle swarm in the mixed element heuristic optimization algorithm according to the parameter number of the initial fuzzy inference model;

and determining a particle parameter value corresponding to each particle in the particle swarm according to the parameter value of the initial fuzzy inference model.

Optionally, the method further comprises:

determining a numerical range of particle parameter values of each particle in the particle swarm according to the parameter values of the initial fuzzy inference model;

And carrying out random assignment on each particle in the particle swarm according to the numerical range.

Optionally, the performing parameter optimization on the initial fuzzy inference model through the mixed element heuristic optimization algorithm to obtain the trained fuzzy inference model includes:

determining parameters of each particle in a particle swarm in the mixed element heuristic optimization algorithm;

iteratively updating parameters of the particle swarm through a preset fitness function to obtain result particles;

and taking the parameters of the result particles as the parameters of the fuzzy inference model to obtain the trained fuzzy inference model.

Optionally, the iteratively updating parameters of the particle swarm through a preset fitness function to obtain result particles includes:

determining the fitness value of each particle in the particle swarm through a preset fitness function;

under the condition that the fitness value does not meet the iteration stop criterion, updating the particle positions in the particle swarm by using an optimized particle swarm algorithm;

updating the particle swarm through a genetic algorithm to obtain an updated particle swarm;

Determining an fitness value of each particle in the updated particle swarm through the fitness function;

and iteratively executing the steps, and obtaining the result particles when the fitness value meets a preset iteration stop criterion.

Optionally, updating the particle swarm by a genetic algorithm to obtain an updated particle swarm, including:

updating each particle in the particle swarm by a selection operator, a crossing operator and a mutation operator to obtain a new particle swarm;

determining an fitness value of each particle in the new particle swarm through the fitness function;

and iteratively executing the steps, and obtaining the updated particle swarm under the condition that the fitness value meets a preset stopping rule.

Optionally, the updating each particle in the particle swarm by the selection operator, the crossover operator and the mutation operator to obtain a new particle swarm includes:

selecting a plurality of particles in the particle swarm through the selection operator to obtain a plurality of target particles;

performing pairwise intersection calculation on the plurality of target particles through the intersection operator to obtain particles after intersection;

Placing the crossed particles into the particle swarm instead of original particles to obtain a primary updated particle swarm;

and carrying out mutation on the particle swarm after preliminary updating through the mutation operator to obtain the new particle swarm.

Optionally, the fitness function is used to determine a root mean square error corresponding to each particle in the particle swarm.

Optionally, the method further comprises:

and storing the abnormal information corresponding to the abnormal hardware into a background log.

Optionally, the method further comprises:

and sending the early warning information to a corresponding remote address in a preset early warning information sending mode.

Optionally, the method further comprises:

the information quantity change rate data and the terminal current change rate data of each piece of hardware are put into a training set used for training the fuzzy inference model, and an updated training set is obtained;

and carrying out parameter updating on the fuzzy inference model by using the updated training set to obtain a fuzzy inference model with updated parameters.

Optionally, the method further comprises:

determining root mean square error between a predicted temperature value obtained by the model reasoning model and a corresponding measured temperature value;

And updating the preset temperature difference threshold according to the root mean square error.

A second aspect of the embodiment of the present application provides a server hardware abnormality early warning device, including:

the data input module is used for respectively inputting the processing information quantity change rate data and the terminal current change rate data of each piece of hardware on the server in a preset time period into a pre-trained fuzzy inference model, wherein the fuzzy inference model is obtained by training based on a mixed element heuristic optimization algorithm;

the predicted temperature value acquisition module is used for acquiring a predicted temperature value of the hardware according to the processing information quantity change rate data and the terminal current change rate data by the fuzzy inference model;

the abnormal hardware marking module is used for marking the hardware as abnormal hardware under the condition that the difference value between the predicted temperature value and the actual temperature value of the hardware is larger than a preset temperature difference threshold value;

and the early warning information generation module is used for generating early warning information corresponding to the abnormal hardware.

Optionally, the apparatus further comprises a fuzzy inference model training module, the fuzzy inference model training module comprising:

the training sample collection sub-module is used for collecting training samples to obtain a training sample data set, wherein the training samples comprise processing information quantity change rate data, end current change rate data and actually measured temperature values of each piece of hardware on the server in a preset time period;

The training sample input sub-module is used for inputting the training sample data set into a fuzzy inference model to be trained;

the model initialization sub-module is used for initializing parameters of the fuzzy inference model to be trained according to the training sample by using a fuzzy clustering method to obtain an initial fuzzy inference model;

the parameter setting sub-module is used for setting initial parameters of the mixed element heuristic optimization algorithm according to the parameters of the initial fuzzy inference model;

and the parameter optimization sub-module is used for carrying out parameter optimization on the initial fuzzy inference model through the mixed element heuristic optimization algorithm to obtain the trained fuzzy inference model.

Optionally, the fuzzy inference model training module further includes:

the temperature difference threshold setting submodule is used for setting a corresponding temperature difference threshold according to a preset temperature difference threshold setting rule.

Optionally, the temperature difference threshold setting submodule includes:

the root mean square error determination submodule is used for determining root mean square error between a predicted temperature value obtained by the fuzzy inference model and a corresponding measured temperature value;

and the temperature difference threshold determining submodule is used for setting the corresponding temperature difference threshold according to the root mean square error.

Optionally, the model initialization submodule includes:

the cluster center selecting sub-module is used for randomly selecting a first preset number of initial cluster centers from the training samples;

the iterative optimization sub-module is used for carrying out iterative optimization on the initial clustering center according to a preset optimization target to obtain an optimized clustering center and corresponding clustering parameters;

the network structure generation sub-module is used for generating a corresponding network structure by clustering division according to the optimized clustering center;

and the initial fuzzy inference model acquisition sub-module is used for assigning values to the network parameters of the fuzzy inference model according to the clustering parameters to obtain the initial fuzzy inference model.

Optionally, the parameter setting submodule includes:

the particle number setting sub-module is used for determining the particle number of the particle group in the mixed element heuristic optimization algorithm according to the parameter number of the initial fuzzy inference model;

and the particle parameter value determining submodule is used for determining a particle parameter value corresponding to each particle in the particle swarm according to the parameter value of the initial fuzzy inference model.

Optionally, the parameter setting sub-module further includes:

the value range determination submodule is used for determining the value range of the particle parameter value of each particle in the particle swarm according to the parameter value of the initial fuzzy inference model;

and the random assignment sub-module is used for carrying out random assignment on each particle in the particle swarm according to the numerical range.

Optionally, the parameter optimization submodule includes:

a particle parameter determination sub-module for determining a parameter of each particle in a particle population in the mixed element heuristic optimization algorithm;

the result particle acquisition sub-module is used for carrying out iterative updating on the parameters of the particle swarm through a preset fitness function to obtain result particles;

and the model acquisition sub-module is used for taking the parameters of the result particles as the parameters of the fuzzy inference model to obtain the trained fuzzy inference model.

Optionally, the result particle acquisition submodule includes:

the first fitness value determining submodule is used for determining the fitness value of each particle in the particle swarm through a preset fitness function;

a particle position updating sub-module, configured to update a particle position in the particle swarm using an optimized particle swarm algorithm if the fitness value does not meet the iteration stop criterion;

The genetic computing sub-module is used for updating the particle swarm through a genetic algorithm to obtain an updated particle swarm;

a second fitness value determining submodule, configured to determine, according to the fitness function, a fitness value of each particle in the updated particle swarm;

and the result particle determination submodule is used for iteratively executing the steps, and obtaining the result particles when the fitness value meets a preset iteration stop criterion.

Optionally, the genetic computing submodule includes:

the genetic updating sub-module is used for updating each particle in the particle swarm through a selection operator, a crossover operator and a mutation operator to obtain a new particle swarm;

a third fitness value determining submodule, configured to determine a fitness value of each particle in the new particle swarm through the fitness function;

and the updated particle swarm acquisition sub-module is used for iteratively executing the steps, and obtaining the updated particle swarm under the condition that the fitness value meets a preset stopping rule.

Optionally, the genetic update submodule includes:

the target particle acquisition submodule is used for selecting a plurality of particles in the particle swarm through the selection operator to obtain a plurality of target particles;

The cross particle acquisition submodule is used for carrying out pairwise cross calculation on the plurality of target particles through the cross operator to obtain crossed particles;

the population acquisition submodule after preliminary updating is used for placing the crossed particles into the particle swarm instead of the original particles to obtain a particle swarm after preliminary updating;

and the particle swarm mutation sub-module is used for mutating the particle swarm after preliminary updating through the mutation operator to obtain the new particle swarm.

Optionally, the fitness function is used to determine a root mean square error for each particle in the population of particles.

Optionally, the apparatus further comprises:

and the abnormal information storage module is used for storing the abnormal information corresponding to the abnormal hardware into a background log.

Optionally, the method further comprises:

and the early warning information sending module is used for sending the early warning information to the corresponding remote address in a preset early warning information sending mode.

Optionally, the apparatus further comprises:

the training set updating module is used for putting the processing information quantity change rate data and the terminal current change rate data of each hardware into a training set used for training the fuzzy inference model to obtain an updated training set;

And the parameter updating module is used for updating the parameters of the fuzzy inference model by using the updated training set to obtain the fuzzy inference model with updated parameters.

Optionally, the apparatus further comprises:

the root mean square error determining module is used for determining root mean square error between the predicted temperature value obtained by the model reasoning model and the corresponding measured temperature value;

and the temperature difference threshold updating module is used for updating the preset temperature difference threshold according to the root mean square error.

A third aspect of the embodiments of the present application provides a readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method according to the first aspect of the present application.

A fourth aspect of the embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps of the method according to the first aspect of the application when the processor executes the computer program.

By adopting the hardware abnormality early warning method provided by the application, the processing information quantity change rate data and the terminal current change rate data of each hardware on a server in a preset time period are respectively input into a pre-trained fuzzy inference model, and the fuzzy inference model is obtained by training based on a mixed element heuristic optimization algorithm; the fuzzy inference model obtains a predicted temperature value of the hardware according to the processed information quantity change rate data and the terminal current change rate data; marking the hardware as abnormal hardware under the condition that the difference value between the predicted temperature value and the actual temperature value of the hardware is larger than a preset temperature difference threshold value; and generating early warning information corresponding to the abnormal hardware. According to the method, a pre-trained fuzzy inference model is used for obtaining a predicted temperature value of hardware according to the processing information amount change rate data and the terminal current change rate data of each piece of hardware on a server in a preset time period, comparing the actual temperature value of the current piece of hardware with the predicted temperature value, further determining whether the piece of hardware is abnormal, generating corresponding early warning information when the piece of hardware is abnormal, wherein the pre-trained fuzzy inference model is obtained by training based on a mixed element heuristic optimization algorithm (IQPSO-GA), and the mixed element heuristic optimization algorithm is combined with an IQPSO (optimized particle swarm algorithm) and a GA (genetic algorithm), wherein the IQPSO algorithm is obtained by optimizing on the basis of the QPSO (particle swarm algorithm), global searching of the data is facilitated, the training effect of the model is improved remarkably, and the trained fuzzy inference model can better capture the relation between the information amount, the power-off current value and the hardware temperature, and the accuracy and efficiency of hardware abnormal early warning are improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments of the present application will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of a fuzzy inference model structure;

FIG. 2 is a diagram illustrating network result generation according to an embodiment of the present application;

FIG. 3 is a flowchart of a hybrid heuristic optimization algorithm according to an embodiment of the present application;

FIG. 4 is a flowchart of a hardware anomaly early warning method according to an embodiment of the present application;

FIG. 5 is a schematic diagram of a sample dataset according to an embodiment of the present application;

FIG. 6 is a diagram of training test results of a fuzzy inference model according to an embodiment of the present application;

FIG. 7 is a diagram of a model training test result based on hybrid heuristic optimization algorithm training according to an embodiment of the present application;

FIG. 8 is a schematic diagram of server hardware anomaly simulation early warning according to an embodiment of the present disclosure;

FIG. 9 is a schematic diagram of a hardware anomaly early warning device according to an embodiment of the present application;

Fig. 10 is a schematic diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without any inventive effort, are intended to be within the scope of the application.

In this embodiment, a fuzzy inference model (ANFIS, adaptive Network-based Fuzzy Inference System) is first built through training, and the training steps of the fuzzy inference model include:

s11: and collecting training samples to obtain a training sample data set, wherein the training samples comprise processing information quantity change rate data, end current change rate data and actual measurement temperature values of each piece of hardware on the server in a preset time period.

In this embodiment, the training sample data set is a set of training data used when training the neural network, and the training sample includes data of a change rate of the processing information amount of each piece of hardware on the server, data of a change rate of the terminal current, and a corresponding measured temperature value in a preset period of time. The processing information quantity of the hardware is the quantity of effective information processed by the hardware in unit time, different business data processing modes can generate different information quantities, the processing information quantity change rate data is the change rate data of the processing information quantity in the time period, which is obtained according to the hardware processing information quantity in the time period, the end current value refers to the component end current of each hardware in the server when processing the data, and is related to the information quantity, generally, the more the processed information or the more complex the information, the higher the current value, the end current change rate data is the change rate data of the outage value in the time period, which is obtained according to the hardware processing information quantity in the time period, the more the consumed electric energy is, and the higher the temperature value is.

In this embodiment, after the processing information amount and the terminal current value of each piece of hardware in the preset time period are collected, the collected raw data are input into the data processing unit, and the received data are processed by the data processing unit, so as to obtain the change rate of the processing information amount and the change rate of the terminal current value.

In this embodiment, the raw data is collected and a training sample data set is generated, which plays a vital role in training the neural network, so that the final training sample data set can enable the neural network to be trained better. The amount of processing information, end currents of components, and temperature of the individual hardware in the server are important parameters related to the hardware. When each part of the server normally works, service data information processed in unit time under different service types is monitored in real time, discretization value is carried out, and terminal current and temperature change information of the part in corresponding time are measured in real time. The processing information amount of each hardware and the relation between the terminal current and the corresponding temperature can be determined by analyzing the processing information amount, the terminal current and the corresponding temperature change information of each hardware, and the collected data is subjected to standardized processing, such as normalization processing, so as to obtain a training sample data set under each service type.

For example, the preset time period may be within 1 hour.

S12: and inputting the training sample into a fuzzy inference model to be trained.

In this embodiment, the fuzzy inference model is a model based on a fuzzy membership function, and predicts corresponding output data according to input data.

Referring to fig. 1, fig. 1 is a schematic diagram of a fuzzy inference model structure. In the figure, circles represent fixed nodes, squares identify self-adaptive nodes, a fuzzy reasoning model consists of five layers, wherein the first layer is a fuzzy layer, the second layer is a rule layer, the third layer is a normalization layer, the fourth layer is a deblurring layer, and the fifth layer is an output layer. In the fuzzy reasoning model, input data is deduced through membership functions to obtain a prediction result, parameters of the membership functions are determined through training of a sample data set, and a mode that the membership functions are combined or interacted with each other is called a rule. The rules are described as follows:

rule 1:（1）

rule 2:（2）

rule 3:（3）

in the method, in the process of the invention,、/>and->Is the input of the node; />Is output; />、/>And->Respectively is +.>、/>And->Related fuzzy sets, +_>；/>、/>、/>And->Is a result parameter, commonly referred to as a back-piece parameter.

In this embodiment, the temperature change rate data x and the end current change rate data y in the training set are used as inputs in the first layer of the model.

The first fuzzy layer is used for fuzzifying input data, and the function formula is as follows:

（4）

in the method, in the process of the invention,outputting a value for the layer; />Is the number of input signals; />、/>And->Is a generalized bell membership function (gbellmf), defined as:

（6）

in the method, in the process of the invention,、/>and->Is a front piece parameter.

The second rule layer may determine the excitation intensity of each rule, that is, the rule fitness of the rule to the data, where the function formula is:

（7）

in the method, in the process of the invention,indicate->Excitation intensity of bar rule.

The function formula of the third normalization layer is:

（8）

in the method, in the process of the invention,for the output value of the third layer, the ratio of the degree of utilization of the ith rule to the sum of the degrees of utilization of all rules is calculated for the ith node.

The fourth deblurring layer is a function of:

（9）

in the method, in the process of the invention,、/>、/>and->Is a result parameter, commonly referred to as a back-piece parameter.

The fifth output layer has the following function:

（10）

and summing the results of each output to obtain a final output result.

In this embodiment, the last output dataIs a predicted temperature value obtained from the input data.

In this embodiment, after a training sample data set is obtained, the obtained training sample data set is input into a fuzzy inference model to be trained, and the fuzzy inference model is trained.

S13: and initializing parameters of the fuzzy inference model to be trained according to the training sample by using a fuzzy clustering method to obtain an initial fuzzy inference model.

In the embodiment, the fuzzy inference model processes the input training sample through a fuzzy clustering method, the fuzzy clustering method is an improvement based on the traditional clustering method, the method comprises the steps of randomly selecting a plurality of clustering centers, endowing all data points with a certain fuzzy membership degree to the clustering centers, continuously correcting the clustering centers through an iterative method to obtain an optimal clustering center, finally outputting a list of the clustering centers and membership degree values of each data point to each clustering center, and using the output data to establish the fuzzy inference model. The initial fuzzy inference model is a fuzzy inference model after the parameters are initialized, and the initial fuzzy inference model has prediction capability, but the accuracy of prediction cannot be guaranteed, and the parameters need to be further adjusted.

In this embodiment, according to the training sample, initializing parameters of the fuzzy inference model to be trained, and obtaining an initial fuzzy inference model includes the specific steps of:

s13-1: and randomly selecting a first preset number of initial clustering centers from the training samples.

In this embodiment, the first preset number may be set by itself according to the actual data amount, and the initial clustering center is a plurality of data points randomly designated in the data set of the training sample.

In this embodiment, a first preset number of cluster center points are randomly selected from the training samples, and the selected cluster center points are used as initial cluster centers.

S13-2: and carrying out iterative optimization on the initial clustering center according to a preset optimization target to obtain an optimized clustering center and corresponding clustering parameters.

In this embodiment, the preset optimization target is a parameter update target for performing cluster optimization on the entire network, and the cluster parameters refer to parameters of a fuzzy inference model when data in the training sample data set is clustered.

In this embodiment, according to a preset optimization target, iterative optimization is performed on an initial cluster center, and when the cluster center cannot continue to perform optimization, iteration is stopped to obtain an optimized cluster center and corresponding cluster parameters.

S13-3: and generating a corresponding network structure by clustering according to the optimized cluster center.

In this embodiment, after the optimized cluster center is determined, a plurality of cluster centers are used as center points to divide and generate a corresponding network structure.

S13-4: and assigning the network parameters of the fuzzy inference model according to the clustering parameters to obtain the initial fuzzy inference model.

In this embodiment, according to the clustering parameters when the clustering center is finally obtained, the network parameters of the fuzzy inference model are assigned, so as to obtain an initial fuzzy inference model.

Referring to fig. 2, fig. 2 is a schematic diagram illustrating network result generation according to an embodiment of the present application. As shown in fig. 2, the training data set is input into the fuzzy inference network, the fuzzy inference network clusters the input samples by a fuzzy clustering method in combination with the initial clustering parameters, the input samples are clustered to generate corresponding network structures, and then the network parameters are assigned according to the clustering result, namely, the network parameters when the clustering result is obtained are used as the final network parameters of the fuzzy inference network, and finally the initial fuzzy inference model is output.

In this embodiment, the optimization objective is to minimize the weighted sum of the distance from the data point in the training sample to its corresponding cluster center and the membership.

In this embodiment, when the weighted sum of the distance and membership between each data point in the training sample data set and the cluster medium corresponding thereto is minimized, the iteration is ended, and the optimization goal is achieved.

S14: and setting initial parameters of the mixed element heuristic optimization algorithm according to the parameters of the initial fuzzy inference model.

In this embodiment, the hybrid heuristic optimization algorithm (IQPSO-GA) is an algorithm created by combining the optimized particle swarm algorithm (IQPSO, quantum Particle Swarm Opti-hybridization) with the genetic algorithm (GA, genetic Algorithm).

For the optimized particle swarm algorithm, first, the update of the speed and the position of the general particle swarm algorithm (PSO, particle Swarm Opti-hybridization) algorithm example is determined by the following equation:

（11）

in the method, in the process of the invention,the inertial weight plays a key balance role in global searching and local searching of the algorithm;representing>The individual particles are at->Velocity vector at the time of iteration. />Indicate->The individual particles are at->Position vector at the time of iteration; />Is->Individual optimal positions of individual particles; />Global optimal positions for the whole population; />And->Is an acceleration constant; />And->Is [0,1]Random numbers subject to uniform distribution.

In this embodiment, one particle corresponds to one parameter of the fuzzy inference model, and the number of particles equivalent to the parameter of the inference model is randomly initialized through the above formula. From the above formula it is known that the fuzzy inference model has a total of 7 unknown parameter values, 、/>、/>、/>、/>、/>、/>. The number of ions is set to 7 and for each of these 7 particles, the parameters of the particles are initialized using the above formula.

OPSO (quantum particle algorithm) is an optimizing evolution algorithm established by introducing quantum theory on the basis of a standard PSO algorithm. In QPSO, the position of the particle is determined by the following equation:

（12）

（13）

（14）

in the method, in the process of the invention,is indicated at +.>Average value of all particle optimal positions in the time of iteration; />Is population particle number;is the dimension of the particles; />Is->And->Random positions in between; />Representing population->No. 5 of individual particles>Maintain at->Optimal position at the time of iteration; />The +.f. representing globally optimal solution for the population>A dimensional location; />,/>And->Are all [0,1 ]]Random numbers in between; />Indicate->No. 5 of individual particles>Maintain at->The position at the time of the iteration; />The contraction-expansion coefficient is an important parameter for controlling the convergence rate of the QPSO algorithm, and is generally changed by adopting a linear reduction method.

（15）

In the method, in the process of the invention,representing the current iteration number, +.>Indicating the set maximum number of iterations.

QPSO introducesAnd->The co-operation and global searching capability among particles are improved, but the global searching efficiency of the particles is also influenced. To improve the particle search efficiency, we adjust the particle position from both the optimal position and the contraction-expansion coefficient of the particle. Set up- >A weighted average of the optimal positions of all particles at each iteration.

（16）

In the method, in the process of the invention,the weight coefficient for each particle is set as follows:

（17）

in the method, in the process of the invention,representing a globally optimal solution->Is>Dimension of the corresponding particle fitness function value, +.>Indicate->No. 5 of individual particles>Fitness function value of dimension.

In practical application, the adjustment of the contraction-expansion coefficient cannot be reasonably changed in the execution process of the algorithm, so that in the embodiment, an error function is defined) The distance relation between the particles and the current optimal position in the particle swarm is represented by an error function, the smaller the error function is, the closer the distance between the particles and the current global optimal position is, the smaller the searching range of the particles is, the early convergence is easy, and when the error function is larger, the farther the distance between the marked particles and the current global optimal position is, the larger the searching range of the particles is, and the convergence speed is reduced.

In this embodiment, the error function is defined as:

（18）

in the method, in the process of the invention,representation ofFirst->A fitness function of the individual particles; />Representing a globally optimal solution->And a fitness function corresponding to the particles.

Based on the definition of the error function, the contraction-expansion coefficientCan be adjusted as follows.

（19）

（20）

In the method, in the process of the invention,is indicated at +. >First->Shrinkage-expansion coefficients of individual particles; when->When larger, the +.>Taking a smaller value to accelerate the convergence rate; when->In smaller cases, the +.>ComparingLarge value, expanding search range and avoiding sinking into local optimum.

In this embodiment, the model parameters are adaptively adjusted by using the error function, and the contraction-expansion coefficients are adjusted according to the actual situation, so as to finally obtain the optimized particle swarm algorithm.

In this embodiment, in the optimized particle swarm algorithm, the initialization of the particle swarm is random, and in the early stage of algorithm iteration, a certain dead zone exists in the population search, which not only reduces the searching efficiency of the algorithm, but also affects the stability of the algorithm, and in addition, along with the continuous iterative updating of the optimized particle swarm algorithm, the diversity of the particle swarm is reduced, and thus falls into local optimum.

In this embodiment, in order to ensure diversity of particle swarms, a genetic algorithm is introduced, and the genetic algorithm updates all chromosomes, i.e., particles, in the current population by using its own selection operator, crossover operator and mutation operator, so that the genetic algorithm has high efficiency in searching for a global optimal solution, and is used in combination with an optimized particle swarm algorithm, thereby well compensating for the defect that the optimized particle swarm algorithm is easy to fall into local optimal, and further obtaining a mixed element heuristic optimization algorithm.

Referring to fig. 3, fig. 3 is a flowchart of a hybrid heuristic optimization algorithm according to an embodiment of the present application, where data in a dataset is used as a particle swarm, the particle swarm is randomly initialized, an fitness value of each particle is calculated, when a stopping criterion is met, an optimal particle is output, the algorithm is ended, when the stopping criterion is not met, a particle position of the particle swarm, that is, a particle position in the particle swarm, is updated, the particle swarm is processed through GA operation (genetic algorithm) to generate a new population, a local optimal value and a global optimal value of the population are updated, an updated particle swarm is obtained, and the fitness value of each particle in the updated particle swarm is calculated until the stopping criterion is met, and the iterative process is ended.

In this embodiment, the fitness function in the hybrid heuristic optimization algorithm is used to calculate the fitness value of each particle, and the root mean square error corresponding to the algorithm particle is used as the current fitness value of the particle, where the higher the fitness value is, the more accurate the result of model prediction is represented, and the model parameter at this time is more close to the optimal parameter.

In this embodiment, the fitness function calculation expression of each particle fitness value is:

（21）/>

In the method, in the process of the invention,representing a total number of data vectors in the training dataset; />The representation is based on particles->Predictive +.>Prediction results of the individual sample data; />Representing the +.>And sample data.

In this embodiment, the particle is optimized by setting the above formula, that is, the parameters of the fuzzy inference model are optimized, and when the particle is optimized to be optimal, the parameters of the particle optimized finally are used as the parameters of the inference model. The fuzzy inference model has a total of 7 unknown parameter values,、/>、/>、/>、/>、/>、/>. After the optimization is finished, the optimal parameter values of the particles 1-7 corresponding to the 7 parameters are obtained, and then the parameters of the fuzzy inference model are determined.

In this embodiment, the fitness function is to calculate the mean square error between the predicted result and the real result obtained by the fuzzy inference model under the parameter, and the smaller the mean square error is, the closer the parameter is to the optimal parameter, that is, the closer the parameter of the particle is to the optimal, and finally, the optimized particle uses the vector value of the particle as the parameter of the inference model, so that the obtained predicted result is optimal.

In this embodiment, the specific step of setting the initial parameters of the hybrid heuristic optimization algorithm according to the parameters of the initial fuzzy inference model includes:

S14-1: and determining the particle number of the particle group in the mixed element heuristic optimization algorithm according to the parameter number of the initial fuzzy inference model.

In this embodiment, when the number of particle groups of the particle groups in the hybrid heuristic optimization algorithm is set, the parameter number of the initial fuzzy inference model is used as the number of particle groups of the particle groups in the hybrid heuristic optimization algorithm.

For example, when there are 10 parameters to be adjusted in the initial fuzzy inference model, i.e., the number of parameters is 10, the number of particle groups of the particle group is set to 10, i.e., there are 10 particles in the particle group.

S14-2: and determining a particle parameter value corresponding to each particle in the particle swarm according to the parameter value of the initial fuzzy inference model.

In this embodiment, after determining the parameter values of the initial fuzzy inference model, the parameter values of each parameter are assigned to the particles in the particle swarm, and the particle parameter values corresponding to each particle in the particle swarm are determined.

In another embodiment of the present application, the method further comprises the steps of:

s14-3, determining a numerical range of particle parameter values of each particle in the particle swarm according to the parameter values of the initial fuzzy inference model.

In this embodiment, according to the parameter value of the initial fuzzy inference model, the value range of the particle parameter value of each particle in the particle swarm is determined, and the parameter value in a certain range is selected as the particle parameter value by taking the parameter value of the inference model as the center.

S14-4: and carrying out random assignment on each particle in the particle swarm according to the numerical range.

In this embodiment, after the value range of the particle parameter value is determined, each particle in the particle group is randomly assigned in the value range.

In this embodiment, the parameter values of the particles are not directly determined, but the particles are randomly assigned within a certain range, so that the diversity of the particles in the particle swarm is ensured, and the model does not tend to converge too early.

S15: and carrying out parameter optimization on the initial fuzzy inference model through the mixed element heuristic optimization algorithm to obtain the trained fuzzy inference model.

In this embodiment, after an initial fuzzy inference model is obtained and initial parameters of a hybrid heuristic optimization algorithm are set, parameter optimization is performed on the initial fuzzy inference model through the hybrid heuristic optimization algorithm, so as to obtain a trained fuzzy inference model.

In this embodiment, the specific steps of performing parameter optimization on the initial fuzzy inference model by using the hybrid heuristic optimization algorithm to obtain the trained fuzzy inference model include:

s15-1, determining parameters of each particle in the particle swarm in the mixed element heuristic optimization algorithm.

In this embodiment, the particle swarm is initialized, and parameters of each particle in the particle swarm are determined, that is, each particle in the particle swarm is assigned, and when the current prediction is completed after the prediction of the fuzzy inference model is completed, the parameters of the fuzzy inference model are used as parameters of the particles in the particle swarm.

And S15-2, carrying out iterative updating on the parameters of the particle swarm through a preset fitness function to obtain result particles.

In this embodiment, the result particles are the optimal particles that are output after stopping the iteration when the hybrid heuristic optimization algorithm satisfies the stopping criterion.

In this embodiment, the root mean square error of each particle for the predicted result is determined through a preset fitness function, and then the root mean square error between the predicted result and the actually measured temperature value of the fuzzy inference model under the current parameter is determined, and then the parameter value of the particle swarm is iteratively updated through a meta-heuristic optimization algorithm, so as to obtain the result particle.

In this embodiment, the step of iteratively updating parameters of the particle swarm to obtain the result particles by using a preset fitness function includes:

s15-2-1: and determining the fitness value of each particle in the particle swarm through a preset fitness function.

In this embodiment, the vector corresponding to each particle is brought into the fitness function formula, i.e. formula (21), through a preset fitness function, so as to obtain the fitness value of each particle in the particle swarm. The fitness value characterizes the root mean square error between the result predicted based on the parameter and the actual result.

S15-2-2: and under the condition that the fitness value does not meet the iteration stop criterion, updating the particle positions in the particle swarm by using an optimized particle swarm algorithm.

In this embodiment, the iteration stop criterion is a rule that the iteration stop needs to satisfy.

In this embodiment, when the fitness value does not meet the iteration stop criterion, the particle positions in the particle swarm are updated by using the optimized particle swarm algorithm, that is, the parameter value is updated by using the fuzzy inference model.

For example, the iteration stop criterion is that the fitness value is lower than a preset fitness value threshold, and the fitness value threshold may be set according to the accuracy requirement.

In another embodiment of the present application, the iteration number may be preset, and after the iteration number is reached, the iteration is stopped, and the particles obtained last time are output as the optimal particles.

S15-2-3: and updating the particle swarm through a genetic algorithm to obtain an updated particle swarm.

In this embodiment, the updated particle group is a particle group obtained by updating the particle group by a genetic algorithm.

In this embodiment, the genetic algorithm is an optimization algorithm for simulating a chromosome mutation process, and uses each particle in the particle swarm as a chromosome, so as to perform cross mutation on the chromosome to generate a new population, thereby obtaining an updated particle swarm.

In this embodiment, updating the particle swarm by a genetic algorithm to obtain an updated particle swarm includes:

s15-2-3-1: and updating each particle in the particle swarm by a selection operator, a crossing operator and a mutation operator to obtain a new particle swarm.

In this embodiment, a genetic algorithm is used to update each particle in a particle swarm by selecting an operator, a crossover operator and a mutation operator, so as to obtain a new particle swarm, and the specific steps include:

Selecting a plurality of particles in a particle swarm through a selection operator to obtain a plurality of target particles;

performing pairwise intersection calculation on a plurality of target particles through an intersection operator to obtain crossed particles;

placing the crossed particles into the particle swarm instead of the primary particles to obtain a primary updated particle swarm;

and carrying out mutation on the particle swarm after preliminary updating through a mutation operator to obtain the new particle swarm.

S15-2-3-2: and determining the fitness value of each particle in the new particle swarm through the fitness function.

In this embodiment, after a new particle group is obtained, the fitness value of each particle in the new particle group is calculated using the fitness function.

S15-2-3-3: and iteratively executing the steps, and obtaining the updated particle swarm under the condition that the fitness value meets a preset stopping rule.

In this embodiment, the operations of crossing, mutation and selection of the chromosomes are iteratively performed until a predetermined stopping rule is satisfied, for example, after a predetermined number of iterations is reached, an updated particle swarm is obtained.

S15-2-4: and determining the fitness value of each particle in the updated particle swarm through the fitness function.

In this embodiment, the fitness function is used to calculate the updated particle swarm, so as to obtain the fitness value of each particle in the updated particle swarm.

S15-2-5: and iteratively executing the steps, and obtaining the result particles when the fitness value meets a preset iteration stop criterion.

In this embodiment, the population particle position update is iteratively performed, and the population genetic operation is performed, so that the result particles are obtained when the obtained fitness value meets a preset iteration stop criterion.

And S15-3, taking the parameters of the result particles as the parameters of the fuzzy inference model to obtain the trained fuzzy inference model.

In this embodiment, the particle parameter value of the obtained result particle is used as the final parameter of the fuzzy inference model, so as to obtain a trained fuzzy inference model.

In another embodiment of the present application, the method further comprises:

s21: and setting a corresponding temperature difference threshold according to a preset temperature difference threshold setting rule.

In this embodiment, the preset temperature difference threshold setting rule is a temperature difference threshold setting rule obtained according to an empirical formula. The preset temperature difference threshold is the maximum value of the root mean square error between the predicted temperature and the actually measured temperature, and when the root mean square error between the predicted temperature and the actually measured temperature exceeds the preset temperature difference threshold, hardware abnormality is indicated.

In this embodiment, according to a preset temperature difference threshold setting rule, the specific steps of setting the corresponding temperature difference threshold include:

s21-1: and determining the root mean square error between the predicted temperature value obtained by the fuzzy inference model and the corresponding measured temperature value.

S21-2: and setting the corresponding temperature difference threshold according to the root mean square error.

In this embodiment, a root mean square error between a predicted temperature value obtained by the fuzzy inference model and a corresponding measured temperature value is determined, and then a temperature difference threshold setting rule is obtained according to an empirical formula, so as to set a corresponding temperature difference threshold.

Illustratively, the temperature difference threshold setting rule is to set the temperature difference threshold to 3 by root mean square error.

Referring to fig. 4, fig. 4 is a flowchart of a hardware anomaly early warning method according to an embodiment of the application. As shown in fig. 4, the method comprises the steps of:

s31: and respectively inputting the processing information quantity change rate data and the terminal current change rate data of each piece of hardware on the server in a preset time period into a pre-trained fuzzy inference model, wherein the fuzzy inference model is trained based on a mixed element heuristic optimization algorithm.

In this embodiment, after a trained fuzzy inference model is obtained, the data of the change rate of the processing information quantity and the data of the change rate of the terminal current of each hardware on the server in a preset time period are respectively input into the pre-trained fuzzy inference model.

For example, during the period that the hardware in the server comprises AI accelerator card, ethernet card, RAID card and other circuit units with low repeatability and smaller transistor density set scale, due to the fact that parameters capable of being monitored are fewer, proper early warning rules cannot be easily found, and therefore abnormal early warning needs to be carried out on the hardware devices through a pre-trained fuzzy inference model.

S32: and the fuzzy inference model obtains a predicted temperature value of the hardware according to the processed information quantity change rate data and the terminal current change rate data.

In this embodiment, after the fuzzy inference model obtains the processing information amount and the terminal current value of each hardware, the predicted temperature value of the hardware is predicted according to the obtained data.

S33: and marking the hardware as abnormal hardware under the condition that the difference value between the predicted temperature value and the actual temperature value of the hardware is larger than a preset temperature difference threshold value.

In this embodiment, the actual temperature value is the actual temperature of each hardware measured by the sensing device on the server.

In this embodiment, when the difference between the predicted temperature value and the actual temperature value of the hardware is greater than the preset temperature difference threshold, it is indicated that the temperature of the hardware is abnormal, and therefore the hardware is marked as abnormal hardware.

For example, when the root mean square error of the measured temperature and the predicted temperature of the ethernet card in the server is greater than a preset temperature difference threshold, the ethernet card is marked as abnormal hardware.

S34: and generating early warning information corresponding to the abnormal hardware.

In this embodiment, when a hardware device marked as abnormal hardware exists, device information of the abnormal device is acquired, and corresponding early warning information is generated according to the acquired device information.

Illustratively, the generated early warning information is: "ethernet card 01 temperature anomaly".

In another embodiment of the present application, the method further comprises:

s41: and storing the abnormal information corresponding to the abnormal hardware into a background log.

In this embodiment, when abnormal hardware is detected, the abnormal information corresponding to the abnormal hardware is stored in the background log, and the manager can check the log at any time to grasp the working condition of each hardware.

In another embodiment of the present application, the method further comprises:

s51: and sending the early warning information to a corresponding remote address in a preset early warning information sending mode.

In this embodiment, after the early warning information is generated, the early warning information is sent to the corresponding remote address in a preset early warning information sending manner.

In this embodiment, a remote address may be preset, after early warning information is generated, the early warning information is sent to a corresponding remote address, so that a manager is guaranteed to receive the early warning information in time, and the hardware of the server is repaired by fault, so that the normal operation of the server is guaranteed.

For example, the preset early warning information sending mode may be short message sending, mail sending, and the like.

In another embodiment of the present application, the method further comprises:

s61: and putting the processed information quantity change rate data and the terminal current change rate prime numbers of each hardware into a training set used for training the fuzzy inference model to obtain an updated training set.

In this embodiment, a large amount of sample data is required for training the prediction model, the model obtained by training should be updated continuously along with the updating of the data, after the pre-selected trained fuzzy inference model is obtained, when the fuzzy inference model predicts according to the data monitored in real time, the received data is put into the training set, and the updated training set is obtained.

S62: and carrying out parameter updating on the fuzzy inference model by using the updated training set to obtain a fuzzy inference model with updated parameters.

In this embodiment, after the updated training set is obtained, the model is trained on the basis of the existing fuzzy inference model by following the new training set, and parameters of the fuzzy inference model are updated to obtain a fuzzy inference model with updated parameters.

In this embodiment, the update time of the model may be set, for example, when the new data amount reaches a certain amount, the parameters of the model are updated, or the parameters of the model are updated every other preset period.

In another embodiment of the present application, the method further comprises:

s71: and determining the root mean square error between the predicted temperature value obtained by the model reasoning model and the corresponding measured temperature value.

In this embodiment, after updating the parameters of the fuzzy inference model, the root mean square error between the predicted temperature value obtained by the fuzzy inference model and the corresponding measured temperature value is determined by using the fitness function.

S72: and updating the preset temperature difference threshold according to the root mean square error.

In this embodiment, the preset temperature difference threshold is updated according to the obtained root mean square error.

In this embodiment, after the parameters of the model are changed, the root mean square error between the predicted temperature value and the actually measured temperature value of the model is also changed, and then the temperature difference threshold is also changed, so that the parameters and the temperature difference threshold of the model are continuously adjusted, the fuzzy inference model is ensured to accurately monitor the temperature condition of hardware, and hardware abnormality early warning is performed.

In the embodiment of the application, the fuzzy inference model is trained by the mixed element heuristic optimization algorithm, and the abnormal early warning is carried out on each hardware in the server by the fuzzy inference model, so that the data information which can be measured by a device with low complexity and small transistor density scale around a circuit unit in the healthy state of the component can be analyzed, the mapping relation among each data is analyzed, and the state of the hardware is further monitored. In the mixed element heuristic optimization algorithm, in order to improve the particle search efficiency, on the basis of a particle swarm algorithm, an optimized particle swarm algorithm is designed based on the influence of a contraction-expansion coefficient on the convergence speed and the search capacity, the intersection and mutation operation of a genetic algorithm is introduced, and a fuzzy inference model is trained based on the algorithm to obtain a more accurate fuzzy inference model, so that the accuracy of server hardware abnormality early warning is effectively improved.

In another embodiment of the application, simulation verification is performed on a fuzzy inference model trained based on a mixed element heuristic optimization algorithm in the application.

Referring to fig. 5, fig. 5 is a schematic diagram of a sample data set according to an embodiment of the present application, and as shown in fig. 5, a simulation experiment is performed by using a set of post-processing data sets as the sample data set. The used sample data set comprises three groups of vectors, the first two groups are respectively the information quantity change rate processed by the server component and the component end current change rate as input data, the other group is the component temperature change rate as output data, and the sample data sets are mapped to 0-1 through normalization processing. There are 500 data points in the data set, and each sample data point is an aperiodic sampling result. The data set was divided into two parts, the first part randomly taking 70% of the data points as training data and the second part taking the remaining 30% of the data as test data.

The experiment was chosen as a fuzzy C-means clustering method and Table 1 gives the optimization algorithm parameter settings with iteration number as termination condition, which were chosen based on the experience of the team in the trial-and-error process. The "selection pressure" in table 1 represents the ratio of the probability of optimal individual selection to the average individual selection probability in the GA algorithm; gamma is a numerical range determined when setting a random array during the cross operation of the GA algorithm.

TABLE 1

Fig. 6 is a training test result diagram of a fuzzy inference model according to an embodiment of the present application, and fig. 7 is a training test result diagram of a model according to an embodiment of the present application based on training of a hybrid heuristic optimization algorithm, and it can be clearly seen by comparing fig. 6 and fig. 7 that a conventional fuzzy inference model has a better ability to fit input-output data than a model according to training of a hybrid heuristic optimization algorithm. Meanwhile, the test data are used for respectively checking the generalization capability of the model obtained by the two training methods, so that the result shows that: the root mean square error rmse=0.011 of the model trained based on the mixed-element heuristic optimization algorithm, and compared with each index of the traditional fuzzy inference model, the model fitting accuracy trained based on the mixed-element heuristic optimization algorithm is improved by more than 47%.

To verify the effectiveness of the proposed server anomaly early warning method based on the IQPSO-GA optimized ANFIS model, we add several anomaly errors (including gaussian noise and outlier noise) on the basis of the sample data shown in fig. 5. The setting error value is shown in fig. 8 (a), fig. 8 is a schematic diagram of server hardware abnormality simulation early warning according to an embodiment of the application, wild value mutation noise is set between 450 th to 490 th groups of sample data, and the comparison result between the predicted value of the ANFIS model after the optimization by the established IQPSO-GA and the actual value after the addition of the error, and the prediction error are shown in fig. 8 (b) and fig. 8 (c), respectively. Therefore, the provided abnormal early warning method based on the IQPSO-GA optimized ANFIS model can effectively map the dynamic relationship among the information quantity change rate, the terminal current change rate and the component temperature change rate processed by the component, namely, whether the component temperature change rate is abnormal or not can be observed through input information. Once the end current or temperature of a certain component is continuously higher than the early warning threshold, the function of the component can still be used at the moment, but the following working state of the component needs to be focused, and the hardware abnormality early warning is carried out on the component.

Based on the same inventive concept, an embodiment of the application provides a server hardware abnormality early warning device. Referring to fig. 9, fig. 9 is a schematic diagram of a hardware anomaly early-warning device 900 according to an embodiment of the application. As shown in fig. 9, the apparatus includes:

the data input module 901 is configured to input the processed information amount change rate data and the terminal current change rate data of each piece of hardware on the server in a preset time period into a pre-trained fuzzy inference model, where the fuzzy inference model is obtained by training based on a mixed element heuristic optimization algorithm;

the predicted temperature value obtaining module 902 is configured to obtain a predicted temperature value of the hardware according to the processed information quantity change rate data and the terminal current change rate data by using the fuzzy inference model;

an abnormal hardware marking module 903, configured to mark the hardware as abnormal hardware when a difference between a predicted temperature value and an actual temperature value of the hardware is greater than a preset temperature difference threshold;

and the early warning information generation module 904 is used for generating early warning information corresponding to the abnormal hardware.

The training sample collection sub-module is used for collecting training samples to obtain a training sample data set, wherein the training samples comprise processing information quantity change rate generation data, end current change rate data and actually measured temperature values of each piece of hardware on the server in a preset time period;

Optionally, the fuzzy inference model training module further includes:

Optionally, the temperature difference threshold setting submodule includes:

Optionally, the model initialization submodule includes:

Optionally, the parameter setting submodule includes:

Optionally, the parameter setting sub-module further includes:

Optionally, the parameter optimization submodule includes:

Optionally, the result particle acquisition submodule includes:

a particle position updating sub-module, configured to update a particle position in the particle swarm using an optimized particle swarm algorithm when the fitness value does not meet an iteration stop criterion;

Optionally, the genetic computing submodule includes:

Optionally, the genetic update submodule includes:

Optionally, the apparatus further comprises:

Optionally, the method further comprises:

Optionally, the apparatus further comprises:

Based on the same inventive concept, another embodiment of the present application provides a readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the hardware anomaly early warning method according to any one of the above embodiments of the present application.

Based on the same inventive concept, another embodiment of the present application provides an electronic device 1000, as shown in fig. 10. Fig. 10 is a schematic diagram of an electronic device according to an embodiment of the present application, including a memory 1002, a processor 1001, and a computer program stored in the memory and capable of running on the processor, where the processor executes the steps in the network service providing method according to any of the foregoing embodiments of the present application.

For the device embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of the method embodiments for relevant points.

In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described by differences from other embodiments, and identical and similar parts between the embodiments are all enough to be referred to each other.

It will be apparent to those skilled in the art that embodiments of the present application may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the application may take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Embodiments of the present application are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the scope of the embodiments of the application.

Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.

The hardware abnormality early warning method, device, equipment and storage medium provided by the application are described in detail, and specific examples are applied to the explanation of the principle and implementation of the application, and the explanation of the above examples is only used for helping to understand the method and core ideas of the application; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.

Claims

1. The server hardware abnormality early warning method is characterized by comprising the following steps:

2. The method of claim 1, wherein the training step of the fuzzy inference model comprises:

collecting training samples to obtain a training sample data set, wherein the training samples comprise processing information quantity change rate data, end current value data and actual measurement temperature values of each piece of hardware on the server in a preset time period;

3. The method according to claim 2, wherein the method further comprises:

4. A method according to claim 3, wherein setting the corresponding temperature difference threshold according to a preset temperature difference threshold setting rule comprises:

5. The method of claim 2, wherein the performing parameter initialization on the fuzzy inference model to be trained according to the training sample by using a fuzzy clustering method to obtain an initial fuzzy inference model comprises:

6. The method of claim 5, wherein the optimization objective is to minimize a weighted sum of the distance and membership of data points in the training sample to their corresponding cluster centers.

7. The method of claim 2, wherein setting initial parameters of the hybrid heuristic optimization algorithm based on parameters of the initial fuzzy inference model comprises:

8. The method of claim 7, wherein the method further comprises:

9. The method according to claim 2, wherein the performing parameter optimization on the initial fuzzy inference model by the hybrid heuristic optimization algorithm to obtain the trained fuzzy inference model includes:

10. The method of claim 9, wherein iteratively updating parameters of the population of particles by a preset fitness function to obtain resultant particles comprises:

11. The method of claim 10, wherein updating the population of particles by a genetic algorithm results in an updated population of particles, comprising:

And iteratively executing the steps, and obtaining the updated particle swarm under the condition that the adaptability value meets the preset stopping rule.

12. The method of claim 11, wherein updating each particle in the particle swarm by a selection operator, a crossover operator, and a mutation operator to obtain a new particle swarm comprises:

13. The method according to any one of claims 9 to 12, wherein the fitness function is used to determine a root mean square error for each particle in the population of particles.

14. The method according to claim 1, wherein the method further comprises:

15. The method according to claim 1, wherein the method further comprises:

16. The method according to claim 1, wherein the method further comprises:

17. The method of claim 16, wherein the method further comprises:

18. A server hardware anomaly early warning device, the device comprising:

19. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any of claims 1 to 17.

20. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any one of claims 1 to 17 when executing the computer program.