Server residual life prediction method and device, equipment and storage medium
Through the blind source separation of server real-time mixed noise and the improvement of pre-trained model, the problem of data accuracy and inefficiency in server remaining life prediction is solved, high adaptability and real-time performance are achieved, and the accuracy of fault prediction is improved. Dynamic adaptability.
Patent Information
- Application Number
- CN202410950387.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-15
- Publication Date
- 2024-11-15
AI Technical Summary
The prior art has problems in the server remaining life prediction of data inefficient data accuracy and efficiency, limited generalization capabilities, insufficient real-time and dynamic adaptability, and limitations in fault diagnosis.
By obtaining real-time mixed noise data at the server runtime, it is separated into real-time independent noise data for each hardware using blind source separation technology, and inputting this data into a pre-trained model to predict the hardware degradation process, ultimately predicting the remaining life of the server. The method includes improving the RP-Trans model and training with historical noise data, using CEEMDAN-SVD and FastICA algorithms for noise separation, and combining MSE and Adam algorithms to optimize the model.
It improves prediction accuracy and adaptability, improves the generalization ability and real-time of the model, ensures dynamic adaptability, can sense faults in advance and maintain them, and overcomes the limitations of fault diagnosis.
Smart Images

Figure CN118964105A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for predicting the remaining life of a server. Background Art
[0002] As an important financial infrastructure for IT operation and maintenance, the computer room is an important carrier of network and information systems. From the demand perspective, with the development of data centers and the rapid growth of cloud services, it is particularly important to master the reliability and stability of servers, which puts higher requirements on the accuracy and efficiency of server fault detection methods; from a technical perspective, it is currently impossible to accurately locate server health detection and maintenance needs, so that the remaining life of the server cannot be well planned and predicted, and it is impossible to understand and intervene in advance before the failure occurs; from a manpower perspective, computer room inspections require a lot of manpower and cannot detect alarms and handle them in time during unmanned inspections.
[0003] At present, existing technologies usually combine machine learning methods to predict disk failures and predict disk failures through historical data. This prediction method has the limitations of singleness and historical data prediction, and the efficiency and ease of use of these prediction schemes in specific application scenarios need to be further improved. Therefore, although the existing solutions have achieved results in disk failure prediction, there are still defects in server remaining life prediction, such as low data accuracy and efficiency, limited generalization ability and applicability, insufficient real-time and dynamic adaptability, and limitations in fault diagnosis. Summary of the invention
[0004] The present application provides a server remaining life prediction method, device, equipment and storage medium, which are used to overcome the problems of low data accuracy and efficiency, limited generalization ability and applicability, insufficient real-time and dynamic adaptability, and limitations in fault diagnosis in the prior art for server remaining life prediction.
[0005] In a first aspect, the present application provides a method for predicting the remaining life of a server, comprising:
[0006] Acquire real-time mixed noise data generated when the server is running, and perform noise data separation on the real-time mixed noise data by using a blind source separation technology to obtain real-time independent noise data corresponding to each hardware included in the server;
[0007] Input each real-time independent noise data to the prediction model corresponding to each of the hardware to obtain the degradation process of each of the hardware, wherein the prediction model corresponding to each of the hardware is obtained by training the pre-trained model using the sample data separated from the noise data;
[0008] The remaining life of the server is predicted according to the respective degradation process of each of the hardware.
[0009] In a possible design, before inputting each real-time independent noise data into the prediction model corresponding to each hardware, the method further includes:
[0010] Improve the RP-Trans prediction model to obtain an improved training model;
[0011] Using historical mixed noise data to train the improved training model to obtain the pre-training model;
[0012] The pre-trained model is trained using the sample data after the noise data is separated to obtain a prediction model corresponding to each of the hardware;
[0013] The sample data after the noise data separation includes the historical independent noise data corresponding to each of the hardware, and the historical independent noise data corresponding to each of the hardware is obtained by performing noise data separation on the historical mixed noise data through the blind source separation technology.
[0014] In a possible design, the improved training model obtained by improving the RP-Trans prediction model includes:
[0015] Adding one-dimensional convolution and pooling operations to each layer of the encoder on the RP-Trans prediction model to obtain the improved training model;
[0016] Among them, the improved training model has local feature capture characteristics, global information processing characteristics and multi-level feature fusion characteristics.
[0017] In a possible design, before the pre-training model is trained using the sample data after the noise data is separated, the method further includes:
[0018] Acquire the monitoring noise data impedance frequency of the historical mixed noise data;
[0019] Calculate source impedance frequency data using the CEEMDAN-SVD algorithm according to the impedance frequency of the monitoring noise data to obtain the number of source signals corresponding to the historical mixed noise;
[0020] Determine the multi-channel mixed impedance frequency according to the IMF component correlation coefficient and the number of source signals;
[0021] A separation matrix is obtained according to the multi-channel impedance frequency using FastICA algorithm;
[0022] The historical independent noise data corresponding to each of the hardware is obtained based on the unmixing model according to the separation matrix.
[0023] In a possible design, obtaining the historical independent noise data corresponding to each of the hardware based on the unmixing model according to the separation matrix includes:
[0024] Calculate the estimated source noise vector after the historical mixed noise is separated according to the separation matrix based on the unmixing model to obtain each load impedance frequency after the historical mixed noise is separated;
[0025] Performing spectrum analysis according to the separated load impedance frequencies to extract features and obtain feature extraction results;
[0026] The harmonic source load type is identified based on the feature extraction results to obtain historical independent noise data corresponding to each of the hardware.
[0027] In one possible design, the server includes hardware including a CPU, a disk, a fan, and a graphics card.
[0028] In a possible design, before obtaining the monitoring noise data impedance frequency of the historical mixed noise data, the method further includes:
[0029] A plurality of sensor data is acquired through a noise monitoring device, wherein the plurality of sensor data includes historical mixed noise data.
[0030] In one possible design, the training model and the pre-training model are trained by a back-propagation algorithm.
[0031] In one possible design, when training the training model and the pre-training model, an MSE function is used to calculate the error between the remaining life prediction result and the remaining life marking result of the training process; and an Adam algorithm is used to optimize parameters in the training process according to the error.
[0032] In a possible design, predicting the remaining life of the server according to the degradation process of each hardware includes:
[0033] Inputting the degradation process of each of the hardware into a preset classifier;
[0034] The output result of the preset classifier is obtained to obtain the remaining life of the server according to the output result.
[0035] In a second aspect, the present application provides a server remaining life prediction device, comprising:
[0036] A preprocessing module is used to obtain real-time mixed noise data generated when the server is running, and to separate the real-time mixed noise data by blind source separation technology to obtain real-time independent noise data corresponding to each hardware included in the server;
[0037] A first prediction module is used to input each real-time independent noise data into a prediction model corresponding to each hardware to obtain a degradation process of each hardware, wherein the prediction model corresponding to each hardware is obtained by training a pre-trained model using sample data separated from the noise data;
[0038] The second prediction module is used to predict the remaining life of the server according to the degradation process of each hardware.
[0039] In a possible design, the device further includes: a model training module; the model training module is used to:
[0040] Improve the RP-Trans prediction model to obtain an improved training model;
[0041] Using historical mixed noise data to train the improved training model to obtain the pre-training model;
[0042] The pre-trained model is trained using the sample data after the noise data is separated to obtain a prediction model corresponding to each of the hardware;
[0043] The sample data after the noise data separation includes the historical independent noise data corresponding to each of the hardware, and the historical independent noise data corresponding to each of the hardware is obtained by performing noise data separation on the historical mixed noise data through the blind source separation technology.
[0044] In a possible design, the model training module is also used to:
[0045] Adding one-dimensional convolution and pooling operations to each layer of the encoder on the RP-Trans prediction model to obtain the improved training model;
[0046] Among them, the improved training model has local feature capture characteristics, global information processing characteristics and multi-level feature fusion characteristics.
[0047] In one possible design, the apparatus further includes: a blind source separation module; the blind source separation module is used to:
[0048] Acquire the monitoring noise data impedance frequency of the historical mixed noise data;
[0049] Calculate source impedance frequency data using the CEEMDAN-SVD algorithm according to the impedance frequency of the monitoring noise data to obtain the number of source signals corresponding to the historical mixed noise;
[0050] Determine the multi-channel mixed impedance frequency according to the IMF component correlation coefficient and the number of source signals;
[0051] A separation matrix is obtained according to the multi-channel impedance frequency using FastICA algorithm;
[0052] The historical independent noise data corresponding to each of the hardware is obtained based on the unmixing model according to the separation matrix.
[0053] In a possible design, the blind source separation module is further used to:
[0054] Calculate the estimated source noise vector after the historical mixed noise is separated according to the separation matrix based on the unmixing model to obtain each load impedance frequency after the historical mixed noise is separated;
[0055] Performing spectrum analysis according to the separated load impedance frequencies to extract features and obtain feature extraction results;
[0056] The harmonic source load type is identified based on the feature extraction results to obtain historical independent noise data corresponding to each of the hardware.
[0057] In one possible design, the server includes hardware including a CPU, a disk, a fan, and a graphics card.
[0058] In a possible design, the blind source separation module is further used to:
[0059] A plurality of sensor data is acquired through a noise monitoring device, wherein the plurality of sensor data includes historical mixed noise data.
[0060] In one possible design, the training model and the pre-training model are trained by a back-propagation algorithm.
[0061] In one possible design, when training the training model and the pre-training model, an MSE function is used to calculate the error between the remaining life prediction result and the remaining life marking result of the training process; and an Adam algorithm is used to optimize parameters in the training process according to the error.
[0062] In a possible design, the second prediction module is specifically used to:
[0063] Inputting the degradation process of each of the hardware into a preset classifier;
[0064] The output result of the preset classifier is obtained to obtain the remaining life of the server according to the output result.
[0065] In a third aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;
[0066] The memory stores computer-executable instructions;
[0067] The processor executes the computer-executable instructions stored in the memory to implement any possible server remaining life prediction method provided in the first aspect.
[0068] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement any possible server remaining life prediction method provided in the first aspect.
[0069] In a fifth aspect, the present application provides a computer program product, comprising computer execution instructions, which, when executed by a processor, are used to implement any possible server remaining life prediction method provided in the first aspect.
[0070] The present application provides a method, device, equipment and storage medium for predicting the remaining life of a server. First, the real-time mixed noise data generated when the server is running is obtained, and the real-time mixed noise data is separated from the noise data by blind source separation technology to obtain the real-time independent noise data corresponding to each hardware included in the server, and then each real-time independent noise data is input into the prediction model corresponding to each hardware to obtain the degradation process of each hardware, wherein the prediction model corresponding to each hardware is obtained by training the pre-trained model with the sample data after the noise data is separated. Finally, the remaining life of the server is predicted according to the degradation process of each hardware. By separating the real-time mixed noise by blind source separation technology to use the real-time independent noise data of different hardware to predict the remaining life of the server, it not only has high adaptability to the noise generated by different servers, but also can improve the prediction accuracy, and enhance the generalization ability of the prediction model and ensure the real-time and dynamic adaptability of the prediction, which is conducive to the early perception of server failures and early maintenance and intervention, and overcomes the limitations of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0072] Figure 1 A schematic diagram of an application scenario provided for an embodiment of the present application;
[0073] Figure 2 A flow chart of a method for predicting the remaining life of a server provided in an embodiment of the present application;
[0074] Figure 3 A schematic diagram of a process for obtaining a prediction model provided in an embodiment of the present application;
[0075] Figure 4 A schematic diagram of a model framework provided for an embodiment of the present application;
[0076] Figure 5 A schematic diagram of a noise data separation process provided in an embodiment of the present application;
[0077] Figure 6 A schematic diagram of the structure of a server remaining life prediction device provided in an embodiment of the present application;
[0078] Figure 7 A schematic diagram of the structure of another device for predicting the remaining service life of a server provided in an embodiment of the present application;
[0079] Figure 8 A schematic diagram of the structure of another device for predicting the remaining service life of a server provided in an embodiment of the present application;
[0080] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0081] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of methods and devices consistent with some aspects of the present application as detailed in the appended claims.
[0082] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0083] Existing technologies usually combine machine learning methods to predict disk failures and use historical data to predict disk failures. This prediction method has the limitations of singleness and historical data prediction, and the efficiency and ease of use of these prediction schemes in specific application scenarios need to be further improved. Therefore, although existing solutions have achieved results in disk failure prediction, they still have defects such as low data accuracy and efficiency, limited generalization ability and applicability, insufficient real-time and dynamic adaptability, and limitations in fault diagnosis for server remaining life prediction.
[0084] In view of the above problems existing in the prior art, the present application provides a method, device, equipment and storage medium for predicting the remaining life of a server. The inventive concept of the method for predicting the remaining life of a server provided by the present application is to separate and process the real-time mixed noise data generated by the collected server during operation, so as to extract the real-time independent noise data generated by each hardware included in the server from the mixed noise, so as to achieve accurate classification of all noises in the server, which can not only improve the prediction accuracy, but also process the noises generated by different servers, so that the method for predicting the remaining life of a server provided by the present application has high adaptability and can improve the generalization ability of the model. Each real-time independent noise data obtained after separation is used to obtain the respective prediction model of each hardware based on the corresponding prediction model of each hardware trained in advance, and the respective degradation process of each hardware is obtained, and finally the remaining life of the server is preset based on the respective degradation process of each hardware. Among them, each real-time independent noise data is input into the prediction model corresponding to each hardware for prediction, which can realize the targeted prediction of each hardware part included in the server, ensure the accuracy of each hardware prediction part in the server, and ensure the real-time and dynamic adaptability of the prediction in the case of data changes, which is conducive to the early perception of server failures and early maintenance and intervention.
[0085] Figure 1A schematic diagram of an application scenario provided in an embodiment of the present application, such as Figure 1 As shown, the server 100 is an infrastructure deployed in the data center of a financial institution, which can be an important carrier of the network and information system. Regardless of demand, technology or manpower, the life cycle of the server 100 needs to be health-monitored. The embodiment of the present application does not limit the type and quantity of the server 100. Figure 1 Only one part is shown as an example. The electronic device 200 is configured to execute the server remaining life prediction method provided in the embodiment of the present application. The real-time mixed noise generated by the server 100 during operation can be separated by blind source separation technology, and the real-time independent noise data of each hardware included in different servers 100 is used to predict the life of the server 100 based on the prediction model corresponding to each hardware trained in advance. It can not only have high adaptability to the noise generated by different servers 100, but also improve the prediction accuracy, enhance the generalization ability of the prediction model, and ensure the real-time and dynamic adaptability of the prediction, which is conducive to the early perception of the server 100 fault and early maintenance and intervention, and overcome the limitations of fault diagnosis.
[0086] In some embodiments, the electronic device 200 may be, for example, a smart phone, a tablet computer, a computer, a server, a server cluster, etc. The embodiment of the present application does not limit the device type of the electronic device 200. Figure 1 The electronic device 200 in the figure is shown by taking a computer as an example.
[0087] In some embodiments, the hardware included in the server 100 includes but is not limited to a CPU, a disk, a fan, and a graphics card, etc., which can be specifically determined according to the actual type of the server 100.
[0088] It should be noted that the above application scenarios are merely illustrative, and the server remaining life prediction method, apparatus, device and storage medium provided in the embodiments of the present application include but are not limited to the above application scenarios.
[0089] Figure 2 A flow chart of a server remaining life prediction method provided in an embodiment of the present application is shown as follows: Figure 2 As shown, the server remaining life prediction method provided in the embodiment of the present application includes:
[0090] S101: Real-time mixed noise data generated when the server is running is obtained, and noise data is separated from the real-time mixed noise data by blind source separation technology to obtain real-time independent noise data corresponding to each hardware included in the server.
[0091] For example, the real-time mixed noise data generated by the server during operation is obtained through noise monitoring equipment. The real-time mixed noise data may come from multiple hardware included in the server, such as the CPU, disk, fan, graphics card, etc. The collected mixed noise data is separated from the noise data by blind source separation technology to obtain the real-time independent noise data corresponding to each of the multiple hardware.
[0092] In some embodiments, real-time noise data collected by multiple sensor data can be obtained through a noise monitoring device, and the real-time noise data collected by multiple sensors is real-time mixed noise data. The multiple sensor data can come from noise monitoring devices deployed at different parts of the server, and the noise monitoring devices can be, for example, noise sensors.
[0093] In addition, the purpose of noise data separation for real-time mixed noise data is to decompose the real-time mixed noise into multiple independent source noises so as to improve the accurate classification of all noise sources during server operation. On the one hand, it can improve the prediction accuracy, and on the other hand, it can enable the embodiment of the present application to realize the prediction of the remaining life of the server according to the noise generated by different servers, and achieve multi-task prediction for various types of noise. While making the embodiment of the present application highly adaptable, it can also achieve a more accurate evaluation of the server health performance and improve the generalization ability of the prediction model.
[0094] S102: Input each real-time independent noise data into the prediction model corresponding to each hardware to obtain the degradation process of each hardware.
[0095] Among them, the prediction model corresponding to each hardware is obtained by training the pre-trained model using the sample data after noise data separation.
[0096] Each real-time independent noise data obtained after noise data separation is input into the prediction model corresponding to each hardware included in the server to obtain the degradation process of each hardware. For example, the real-time independent noise data of the CPU is input into the prediction model corresponding to the CPU to obtain the CPU degradation process, the real-time independent noise data of the disk is input into the prediction model corresponding to the disk to obtain the disk degradation process, the real-time independent noise data of the fan is input into the prediction model corresponding to the fan to obtain the fan degradation process, the real-time independent noise data of the graphics card is input into the prediction model corresponding to the graphics card to obtain the graphics card degradation process, etc. Among them, the degradation process of each hardware refers to the prediction result of the prediction model corresponding to each hardware.
[0097] It should be noted that the hardware included in the server is determined by the actual working conditions, and each hardware is pre-trained with its own prediction model. That is, the prediction model corresponding to each hardware is the corresponding prediction model obtained by training the pre-trained model with the sample data after noise data separation. Among them, the pre-trained model is a model obtained by improving the RP-Trans prediction model and training it with historical mixed noise data. The sample data after noise data separation includes the historical independent noise data corresponding to each hardware obtained by noise data separation of historical mixed noise data through blind source separation technology.
[0098] In some embodiments, each real-time independent noise data can be used as the main input, and the operating condition data in the prediction process can be used as the service input to obtain the degradation process of each hardware. The operating condition data can be, for example, the unprocessed corresponding data output by the sensor when collecting the real-time mixed noise data.
[0099] S103: Predict the remaining life of the server according to the degradation process of each hardware.
[0100] After the degradation process of each hardware included in the server is obtained, a predicted value of the remaining useful life (RUL) of the server is obtained based on the degradation process of each hardware, thereby realizing the prediction of the remaining useful life of the server.
[0101] In a possible design, a possible implementation of step S103 includes:
[0102] Input the degradation process of each hardware to the preset classifier, obtain the output result of the preset classifier, and obtain the remaining life of the server according to the output result. The preset classifier can be, for example, SVM (Support Vector Machines, SVM). The embodiment of the present application does not limit the implementation method of using the preset classifier to output the remaining life, for example, the remaining life of the server can be obtained using a URL, etc.
[0103] It can be seen from the description of the above embodiments that the server remaining life prediction method provided by the embodiment of the present application first obtains the real-time mixed noise data generated when the server is running, and separates the noise data of the real-time mixed noise data by blind source separation technology to obtain the real-time independent noise data corresponding to each hardware included in the server, and then inputs each real-time independent noise data into the prediction model corresponding to each hardware to obtain the degradation process of each hardware, wherein the prediction model corresponding to each hardware is obtained by training the pre-trained model with the sample data after the noise data is separated. Finally, the remaining life of the server is predicted according to the degradation process of each hardware. By separating the real-time mixed noise by blind source separation technology to use the real-time independent noise data of different hardware to predict the remaining life of the server, it not only has high adaptability to the noise generated by different servers, but also can improve the prediction accuracy, and enhance the generalization ability of the prediction model and ensure the real-time and dynamic adaptability of the prediction, which is conducive to the early perception of server failures and early maintenance and intervention, and overcomes the limitations of fault diagnosis.
[0104] In a possible design, before step S102, the following is also included: Figure 3 The steps of training to obtain the prediction model corresponding to each hardware are shown. Figure 3 A schematic diagram of a process for obtaining a prediction model provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the embodiment of the present application includes:
[0105] S201: Improve the RP-Trans prediction model to obtain an improved training model.
[0106] Model improvement can be performed based on the RP-Trans prediction model, and the improved RP-Trans prediction model can be determined as an improved training model.
[0107] In some embodiments, the improved training model obtained by improving the RP-Trans prediction model may include:
[0108] One-dimensional convolution and pooling operations are added to each layer of the encoder in the RP-Trans prediction model to obtain an improved training model. Figure 4 The framework of the improved training model is shown, in which conv1d and Maxpooling are added during the improvement.
[0109] For example, an improved attention mechanism is added, and rotational position encoding is used instead of the original encoding method in the Transformer of the RP-Trans prediction model. The absolute position encoding is combined with the relative position encoding, so that the model can distinguish tokens at different positions and enhance the dependency between different positions in the sequence.
[0110] Among them, adding convolution and pooling operations can combine the local feature capture capability of convolutional neural networks with the global information processing capability of the multi-head ProbSparse attention mechanism. Unlike the multi-head ProbSparse attention mechanism, the convolution operation is good at refining local features, which can significantly improve the efficiency of the model in learning local features, so that the improved training model has feature capture characteristics. In addition, by fusing the output of each Dilated-SKConv convolution module with the output of a convolutional neural network layer, the improved training model can have multi-level feature fusion characteristics, effectively enhancing the ability of the improved training model to capture and integrate local information, thereby improving the accuracy of the prediction.
[0111] Optionally, in order to ensure the consistency of input and output dimensions, a pooling layer is introduced after each convolutional layer, ensuring that the multi-head Probsparse self-attention mechanism can seamlessly receive and process data.
[0112] In some embodiments, the length of the first convolutional layer is, for example, L, and can be operated using a 1×1×3 convolution kernel and a stride of 1. In order to be consistent with the output of the self-attention module, a maximum pooling layer with a stride of 2 and a kernel size of 3 can be used to reduce the dimensionality of the features. The pooled output is combined with the output of the multi-headed Probsparse attention mechanism and the Dilated-SKConv convolution module as the input of the next round of attention mechanism and convolution layer. This process is repeated in the next attention module to achieve complex multi-level feature fusion and further improve model performance.
[0113] S202: Train the improved training model using historical mixed noise data to obtain a pre-trained model.
[0114] The improved training model is trained using the historical mixed noise data as samples, and the training model that has finished training is determined as the pre-training model. The historical mixed noise data is the mixed noise data of the server with the remaining life of the server marked. The historical mixed noise data can be obtained by acquiring multiple sensor data through a noise monitoring device, that is, the multiple sensor data includes the historical mixed noise data, and the collection method is similar to that of the real-time mixed noise data.
[0115] The embodiments of the present application do not limit the model training process. For example, in some embodiments, the training model can be trained by a back propagation algorithm. When training the training model, a mean square error (MSE) function is used to calculate the error between the remaining life prediction result and the remaining life marking result of the training process. In addition, an adaptive momentum (Adam) algorithm can be used to optimize the parameters in the training process according to the error.
[0116] In some embodiments, the collected multiple sensor data can be used as the main input and the operating condition data can be used as the secondary input to train the improved training model to achieve forward transfer. The operating condition data can be, for example, the unprocessed corresponding data output by the sensor.
[0117] S203: Train the pre-trained model using the sample data after noise data separation to obtain a prediction model corresponding to each hardware.
[0118] The sample data after noise data separation includes historical independent noise data corresponding to each hardware, and the historical independent noise data corresponding to each hardware is obtained by performing noise data separation on historical mixed noise data through blind source separation technology.
[0119] For example, the historical independent noise data corresponding to each hardware obtained after the noise data is separated can be used to further train the pre-training model, which can reduce the accuracy and low efficiency of the final prediction model corresponding to each hardware caused by the preprocessing of different data sets of the algorithm. In practical applications, the prediction rate can be greatly improved, and through further training, the data in training can be compared and learned, so that the accuracy of the final prediction result is greatly improved, thereby improving the reliability and availability of the server remaining life prediction.
[0120] In addition, by generating a prediction model corresponding to each hardware through a pre-trained model, the prediction accuracy of each hardware part in the server can be ensured. Secondly, by adding secondary training, the secondary prediction of the prediction model can be realized according to the characteristics of each hardware. When the data changes, the real-time and dynamic adaptability of the prediction model can be ensured.
[0121] The present application embodiment does not limit the training process of the pre-trained model, and the pre-trained model can be trained by the back propagation algorithm. When training the pre-trained model, the mean square error MSE function is used to calculate the error between the remaining life prediction result and the remaining life marking result of the training process. In addition, the Adam algorithm can also be used to optimize the parameters in the training process according to the error.
[0122] The embodiment of the present application provides a server remaining life prediction method for obtaining a prediction model corresponding to each hardware, using a multi-head ProbSparse self-attention module to replace the original self-attention module in the Transformer model, thereby obtaining an improved training model, and training the improved training model through historical mixed noise data to obtain a pre-trained model, and training the pre-trained model with sample data after noise data separation to obtain a prediction model corresponding to each hardware. The model training can be performed according to the data characteristics of each hardware itself, so that the server remaining life prediction method provided by the embodiment of the present application can perform targeted predictions for each hardware, ensuring the accuracy of the prediction part of each hardware, and thus improving the accuracy of the server remaining life prediction, overcoming the limitation of the prior art that only the remaining life prediction is performed for hardware monomers such as disks or fans, and the server remaining life prediction method provided by the embodiment of the present application can be applicable to different servers and has high adaptability.
[0123] In a possible design, before step S202, the following steps are also included: Figure 5 The steps for obtaining historical independent noise data corresponding to each hardware are shown. Figure 5 A noise data separation process diagram provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, the embodiment of the present application includes:
[0124] S301: Acquire monitoring noise data impedance frequency of historical mixed noise data.
[0125] Blind source separation technology can be used to separate noise data, for example, s(t) = [s1(t), s2(t), ..., s n (t)] T are n independent unknown source noise vectors, x(t)=[x1(t),x2(t),...,x n (t)] T is the known noise vector detected by m sensors after the source noise is transmitted and mixed, y(t) = [y1(t), y2(t), ..., y n (t)] T is the estimated source noise vector obtained by separation of mixed noise, A is an m×n dimensional mixing matrix, W is an n×m dimensional separation matrix, and taking the linear blind source noise mixing model shown in formula (1) as an example, the goal of blind source separation technology is to find the separation matrix W to separate each source noise from the mixed noise x(t), where the unmixing model is shown in formula (2):
[0126] x(t)=As(t) (1)
[0127] y(t)=Wx(t)=WAs(t) (2)
[0128] It can be understood that the historical mixed noise data is represented by x(t), and the monitoring noise data impedance frequency R(t) of the historical mixed noise data is obtained. Before obtaining the monitoring noise data impedance frequency of the historical mixed noise data, it also includes obtaining multiple sensor data through a noise monitoring device to obtain the historical mixed noise data.
[0129] S302: Calculate source impedance frequency data using the CEEMDAN-SVD algorithm according to the impedance frequency of the monitored noise data to obtain the number of source signals corresponding to the historical mixed noise.
[0130] S303: Determine the multi-channel mixed impedance frequency according to the IMF component correlation coefficient and the number of source signals.
[0131] The noise signal belongs to a special type of underdetermined blind source separation, so the number of source noise information needs to be obtained as a known condition. To obtain the number of source noise, the SVD is used to solve the eigenvalue of the mixed noise covariance matrix, and the difference between the eigenvalues of the noise and interference noise is used to estimate the number of source noise.
[0132] Specifically, the original noise, i.e., the monitoring noise data impedance frequency, is decomposed by CEEMDAN to obtain the characteristic intrinsic mode function X containing the source noise. IMF =[IMF1,IMF2,...,IMF n ,R n ]. Since the intrinsic mode function IMF obtained after decomposition i contains the characteristic scale information and characteristic frequency information of the noise, so the IMF with similar main frequency is i The reconstructed virtual mixed noise X′ is obtained by superposition IMF =[IMF′1,IMF′2,...,IMF′ k ]. Then X′ IMF After removing the mean value from the original noise x(t), a new multi-channel mixed noise is formed. Make the number of mixed noise greater than the number of source noise, that is, the source signal data. Among them, the IMF component correlation coefficient is the intrinsic mode function IMF i The correlation coefficient in .
[0133] In order to construct a suitable virtual channel noise signal, it is necessary to screen by comparing the correlation coefficients between the source signal and each IMF component. First, calculate x(t) and X IMFThe correlation coefficient of each IMF component in . The larger the correlation coefficient, the more characteristic information of the original signal contained in the component, and vice versa. The IMF component with a correlation coefficient greater than 0.8 is set as a strong correlation component, which can be directly used as a virtual noise channel signal, while the IMF component with a correlation coefficient less than 0.8 should be combined with the strong correlation component as a virtual channel signal to ensure that the constructed virtual channel signal can contain more characteristic information of the source signal. The correlation coefficient γ is obtained by the following formula (3):
[0134]
[0135] In formula (3), Cov() represents the covariance function; Var() represents the variance function. The larger the γ value is, the better it can reflect the mutual matching degree of the two noise signals.
[0136] After calculating the correlation coefficient γ, the virtual channel noise and the original noise are recombined into multi-channel mixed noise z(t) according to the above rules. The number of channels is shown in formula (4):
[0137] z(t)=[x(t),z1(t),z2(t),...,z δ (t)] (4)
[0138] In formula (4): z i (t) represents the noise of the i-th virtual channel, i = 1, 2, …, δ.
[0139] Further obtain the multi-channel mixed impedance frequency Z corresponding to the multi-channel mixed noise z(t) R (t).
[0140] S304: Using the FastICA algorithm to obtain a separation matrix according to the multi-channel impedance frequency.
[0141] After obtaining multi-channel mixed noise, the single-channel blind source separation model can be converted into a positive definite blind source separation model, and the FastICA algorithm is used to solve the separation matrix.
[0142] S305: Obtain historical independent noise data corresponding to each hardware based on the separation matrix and the unmixing model.
[0143] For example, the estimated source noise vector after the separation of the historical mixed noise is calculated based on the unmixing model shown in the above formula (2) according to the separation matrix, and each load impedance frequency after the separation of the historical mixed noise is obtained. Then, spectrum analysis is performed based on each load impedance frequency after separation to extract features, and each feature extraction result is obtained. Then, the harmonic source load type is identified for each feature extraction result, and the identification result obtained represents the historical independent noise data corresponding to each hardware included in the server.
[0144] It should be noted that the possible implementation method, principle and effect of using blind source separation technology to separate the real-time mixed noise data to obtain the real-time independent noise data corresponding to each hardware are similar to Figure 5 The possible implementation method, principle and effect of performing noise data separation on the historical mixed noise data to obtain the historical independent noise data corresponding to each hardware are similar, and the only difference is that Figure 5 In the illustrated embodiment, noise separation processing is performed on historical mixed noise data, while in the server remaining life prediction process, noise separation processing is performed on real-time mixed noise data. Both are generated when the server is running, the difference being that the former is historical data and the latter is real-time data.
[0145] The server remaining life prediction method provided in the embodiment of the present application uses blind source separation technology to separate the data source and extract the noise of each hardware part from the mixed noise, so as to accurately classify all noises in the server and improve the prediction accuracy.
[0146] It can be seen from the description of the above embodiments that the server remaining life prediction method provided by the embodiment of the present application, through the blind source separation technology, separates the real-time mixed noise to use the real-time independent noise data of different hardware to predict the remaining life of the server, which not only has high adaptability to the noise generated by different servers, but also can improve the prediction accuracy, and enhance the generalization ability of the prediction model and ensure the real-time and dynamic adaptability of the prediction. It is conducive to the early perception of the server fault board and early maintenance and intervention, overcomes the limitations of fault diagnosis, avoids emergencies, and ensures continuous stable and effective operation of production. In addition, the server remaining life prediction method provided by the embodiment of the present application can perform self-supervision detection and analysis, thereby reducing labor costs and risks caused by sudden server failures. During the service life of the server, the remaining life can be predicted to effectively allocate server resources reasonably, and the possible failures of the server can be arranged at a reasonable time for maintenance. Predictive maintenance can more effectively plan maintenance work, improve service efficiency, greatly reduce the cost of multiple uncertain failure maintenance, and save resources. At the same time, it can also avoid equipment failures that may be caused by excessive use of the server, thereby extending the service life of the server, and reducing the additional costs caused by emergency repairs or unexpected downtime.
[0147] Figure 6 A schematic diagram of the structure of a server remaining life prediction device provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, the server remaining life prediction device 400 provided in the embodiment of the present application includes:
[0148] The preprocessing module 401 is used to obtain real-time mixed noise data generated when the server is running, and separate the real-time mixed noise data by blind source separation technology to obtain real-time independent noise data corresponding to each hardware included in the server;
[0149] The first prediction module 402 is used to input each real-time independent noise data into the prediction model corresponding to each hardware to obtain the degradation process of each hardware. The prediction model corresponding to each hardware is obtained by training the pre-trained model with the sample data separated from the noise data;
[0150] The second prediction module 403 is used to predict the remaining life of the server according to the degradation process of each hardware.
[0151] exist Figure 6 On the basis of Figure 7 A schematic diagram of the structure of another server remaining life prediction device provided in an embodiment of the present application, such as Figure 7 As shown, the server remaining life prediction device 400 provided in the embodiment of the present application further includes: a model training module 404, the model training module 404 is used to:
[0152] Improve the RP-Trans prediction model to obtain an improved training model;
[0153] The improved training model is trained using historical mixed noise data to obtain a pre-training model;
[0154] The pre-trained model is trained using the sample data after noise data separation to obtain the prediction model corresponding to each hardware;
[0155] The sample data after noise data separation includes historical independent noise data corresponding to each hardware, and the historical independent noise data corresponding to each hardware is obtained by performing noise data separation on historical mixed noise data through blind source separation technology.
[0156] In a possible design, the model training module 404 is further used to:
[0157] In the RP-Trans prediction model, one-dimensional convolution and pooling operations are added to each layer of the encoder to obtain an improved training model;
[0158] Among them, the improved training model has the characteristics of local feature capture, global information processing and multi-level feature fusion.
[0159] exist Figure 7 On the basis of Figure 8 A schematic diagram of the structure of another server remaining life prediction device provided in an embodiment of the present application is shown in FIG. Figure 8As shown, the server remaining life prediction device 400 provided in the embodiment of the present application further includes: a blind source separation module 405, the blind source separation module 405 is used to:
[0160] Obtain monitoring noise data impedance frequency of historical mixed noise data;
[0161] According to the impedance frequency of the monitored noise data, the CEEMDAN-SVD algorithm is used to calculate the source impedance frequency data, and the number of source signals corresponding to the historical mixed noise is obtained;
[0162] Determine the multi-channel mixed impedance frequency according to the IMF component correlation coefficient and the number of source signals;
[0163] The FastICA algorithm is used to obtain the separation matrix according to the multi-channel impedance frequency;
[0164] The historical independent noise data corresponding to each hardware is obtained based on the separation matrix and the unmixing model.
[0165] In one possible design, the blind source separation module 405 is further configured to:
[0166] According to the separation matrix, based on the unmixing model, the estimated source noise vector after the separation of the historical mixed noise is calculated, and each load impedance frequency after the separation of the historical mixed noise is obtained;
[0167] Perform spectrum analysis according to the separated load impedance frequencies to extract features and obtain feature extraction results;
[0168] The harmonic source load type is identified for each feature extraction result to obtain the historical independent noise data corresponding to each hardware.
[0169] In one possible design, the server includes hardware including a CPU, a disk, a fan, and a graphics card.
[0170] In one possible design, the blind source separation module 405 is further configured to:
[0171] A plurality of sensor data is acquired through a noise monitoring device, wherein the plurality of sensor data includes historical mixed noise data.
[0172] In one possible design, the training model and the pre-trained model are trained through a back-propagation algorithm.
[0173] In a possible design, when training the training model and the pre-training model, the MSE function is used to calculate the error between the remaining life prediction result and the remaining life marking result of the training process; and the Adam algorithm is used to optimize the parameters in the training process according to the error.
[0174] In a possible design, the second prediction module 403 is specifically configured to:
[0175] Input the degradation process of each hardware to the preset classifier;
[0176] The output result of the preset classifier is obtained to obtain the remaining life of the server according to the output result.
[0177] The data processing device provided in the embodiment of the present application can execute the corresponding steps of the data processing method in the above method embodiment. Its implementation principle and technical effect are similar and will not be repeated here.
[0178] Fig. 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Fig. 9 As shown, the electronic device 500 may include: a processor 501 , and a memory 502 communicatively connected to the processor 501 .
[0179] The memory 502 is used to store programs. Specifically, the programs may include program codes, and the program codes include computer-executable instructions.
[0180] The memory 502 may include a high-speed RAM memory, and may also include a non-volatile memory (NoN-volatile memory), such as at least one disk memory.
[0181] The processor 501 is used to execute the computer-executable instructions stored in the memory 502 to implement the server remaining life prediction method provided in the above embodiments.
[0182] The processor 501 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.
[0183] Optionally, the memory 502 may be independent or integrated with the processor 501. When the memory 502 is a device independent of the processor 501, the electronic device 500 may further include:
[0184] The bus 503 is used to connect the processor 501 and the memory 502. The bus can be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc., but it does not mean that there is only one bus or one type of bus.
[0185] Optionally, in a specific implementation, if the memory 502 and the processor 501 are integrated on a chip, the memory 502 and the processor 501 can communicate through an internal interface.
[0186] The present application also provides a computer-readable storage medium, which may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, and other media that can store program codes. Specifically, the computer-readable storage medium stores computer execution instructions, and the computer execution instructions are used for the methods in the above embodiments.
[0187] The present application also provides a computer program product, including computer execution instructions, which implement the method in the above embodiment when executed by a processor.
[0188] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.
[0189] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.
Claims
1. A method for predicting the remaining service life of a server, characterized in that: include: Acquire real-time mixed noise data generated when the server is running, and perform noise data separation on the real-time mixed noise data by using a blind source separation technology to obtain real-time independent noise data corresponding to each hardware included in the server; Input each real-time independent noise data to the prediction model corresponding to each of the hardware to obtain the degradation process of each of the hardware, wherein the prediction model corresponding to each of the hardware is obtained by training the pre-trained model using the sample data separated from the noise data; The remaining life of the server is predicted according to the respective degradation process of each of the hardware.
2. The server remaining life prediction method according to claim 1, characterized in that: Before inputting each real-time independent noise data into the prediction model corresponding to each hardware, the method further includes: Improve the RP-Trans prediction model to obtain an improved training model; Using historical mixed noise data to train the improved training model to obtain the pre-training model; The pre-trained model is trained using the sample data after the noise data is separated to obtain a prediction model corresponding to each of the hardware; The sample data after the noise data separation includes the historical independent noise data corresponding to each of the hardware, and the historical independent noise data corresponding to each of the hardware is obtained by performing noise data separation on the historical mixed noise data through the blind source separation technology.
3. The server remaining life prediction method according to claim 2, characterized in that: The improved training model obtained by improving the RP-Trans prediction model includes: Adding one-dimensional convolution and pooling operations to each layer of the encoder on the RP-Trans prediction model to obtain the improved training model; Among them, the improved training model has local feature capture characteristics, global information processing characteristics and multi-level feature fusion characteristics.
4. The server remaining life prediction method according to claim 2, characterized in that: Before the pre-training model is trained using the sample data separated from the noise data, the method further includes: Acquire the monitoring noise data impedance frequency of the historical mixed noise data; Calculate source impedance frequency data using the CEEMDAN-SVD algorithm according to the impedance frequency of the monitoring noise data to obtain the number of source signals corresponding to the historical mixed noise; Determine the multi-channel mixed impedance frequency according to the IMF component correlation coefficient and the number of source signals; A separation matrix is obtained according to the multi-channel impedance frequency using FastICA algorithm; The historical independent noise data corresponding to each of the hardware is obtained based on the unmixing model according to the separation matrix.
5. The server remaining life prediction method according to claim 4, characterized in that: The obtaining the historical independent noise data corresponding to each of the hardware based on the unmixing model according to the separation matrix includes: Calculate the estimated source noise vector after the historical mixed noise is separated according to the separation matrix based on the unmixing model to obtain each load impedance frequency after the historical mixed noise is separated; Performing spectrum analysis according to the separated load impedance frequencies to extract features and obtain feature extraction results; The harmonic source load type is identified based on the feature extraction results to obtain historical independent noise data corresponding to each of the hardware.
6. The server remaining life prediction method according to claim 1, characterized in that: The hardware included in the server includes a CPU, a disk, a fan and a graphics card.
7. The server remaining life prediction method according to claim 4, characterized in that: Before obtaining the monitoring noise data impedance frequency of the historical mixed noise data, the method further includes: A plurality of sensor data is acquired through a noise monitoring device, wherein the plurality of sensor data includes historical mixed noise data.
8. The server remaining life prediction method according to claim 2, characterized in that: The training model and the pre-training model are trained by a back propagation algorithm.
9. The server remaining life prediction method according to claim 2, characterized in that: When training the training model and the pre-training model, the MSE function is used to calculate the error between the remaining life prediction result and the remaining life marking result of the training process; and the Adam algorithm is used to optimize the parameters in the training process according to the error.
10. The server remaining life prediction method according to any one of claims 1 to 9, characterized in that: The predicting the remaining life of the server according to the degradation process of each hardware includes: Inputting the degradation process of each of the hardware into a preset classifier; The output result of the preset classifier is obtained to obtain the remaining life of the server according to the output result.
11. A server remaining life prediction device, characterized in that: include: A preprocessing module is used to obtain real-time mixed noise data generated when the server is running, and to separate the real-time mixed noise data by blind source separation technology to obtain real-time independent noise data corresponding to each hardware included in the server; A first prediction module is used to input each real-time independent noise data into a prediction model corresponding to each hardware to obtain a degradation process of each hardware, wherein the prediction model corresponding to each hardware is obtained by training a pre-trained model using sample data separated from the noise data; The second prediction module is used to predict the remaining life of the server according to the degradation process of each hardware.
12. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the server remaining life prediction method according to any one of claims 1 to 10.
13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the server remaining life prediction method according to any one of claims 1 to 10.
14. A computer program product, comprising computer executable instructions, which are used to implement the server remaining life prediction method according to any one of claims 1 to 10 when executed by a processor.
Citation Information
Patent Citations
Machine equipment fault diagnosis method based on artificial experience and voice recognition
CN110940539A
Server hardware monitoring method, device and equipment and readable medium
CN113204461A
Human body respiration and heartbeat signal accurate extraction method based on millimeter wave radar
CN117179715A
Method of evaluating dielectric breakdown lifetime
JP2009259947A
