Server power supply detection method and device, storage medium and electronic equipment

Abstract

The invention discloses a server power supply detection method and device, a storage medium and electronic equipment, and relates to the technical field of computers, and the method comprises the steps: obtaining multi-dimensional operation data of a server power supply; feature extraction is conducted on the multi-dimensional operation data, power consumption features of the server are obtained, and the power consumption features are used for indicating the time sequence change relation of the server for the electric energy demand; generating decay characteristics of the server power supply under the power consumption characteristics according to the multi-dimensional operation data; and detecting the residual service life of the server power supply at the target moment after the current moment according to the decay characteristic and the power consumption characteristic, thereby solving the problem of relatively low detection efficiency of the server power supply in related technologies, and achieving the technical effect of improving the detection efficiency of the server power supply.

Description

Technical Field

[0001] This application relates to the field of computer technologies, and in particular, to a method and device for detecting a server power supply, a storage medium, and an electronic device. Background Art

[0002] The server power supply is the basis for the normal operation of the server. The server power supply provides electrical energy for the server to ensure the normal operation of the server power supply. In order to better ensure the operation quality of the server power supply, the power module life prediction technology has been proposed in the current field. The power module life prediction technology aims to predict its remaining useful life (RUL, Remaining Useful Life) by monitoring and analyzing the working state of the power module, so as to provide a basis for the maintenance plan and fault prevention.

[0003] The current power module life prediction technology usually predicts the remaining useful life of the power supply according to the number of times the power supply is used. This prediction method is relatively one-sided, and the number of times the power supply is used cannot truly reflect the real use state of the server power supply, resulting in inaccurate prediction results. Summary of the Invention

[0004] This application provides a method and device for detecting a server power supply, a storage medium, and an electronic device, so as to at least solve the problem of low detection efficiency of the server power supply in related technologies.

[0005] This application provides a method for detecting a server power supply, including: obtaining multi-dimensional operation data of the server power supply, where the multi-dimensional operation data is used to characterize the power supply state of the server by the server power supply during a reference period before the current moment from multiple feature dimensions; performing feature extraction on the multi-dimensional operation data to obtain the power consumption feature of the server, where the power consumption feature is used to indicate the temporal change relationship of the server's demand for electrical energy; generating a decay feature of the server power supply under the power consumption feature according to the multi-dimensional operation data, where the decay feature is used to indicate the influence relationship of the power consumption feature on the decay of the server power supply; and detecting the remaining useful life of the server power supply at a target moment after the current moment according to the decay feature and the power consumption feature.

[0006] The present application also provides a detection device for a server power supply, including: an acquisition module, configured to acquire multi-dimensional operation data of the server power supply, where the multi-dimensional operation data is used to characterize the power supply state of the server by the server power supply within a reference time period before the current moment from multiple feature dimensions; an extraction module, configured to perform feature extraction on the multi-dimensional operation data to obtain the power consumption feature of the server, where the power consumption feature is used to indicate the temporal variation relationship of the server's demand for electric energy; a generation module, configured to generate a decay feature of the server power supply under the power consumption feature according to the multi-dimensional operation data, where the decay feature is used to indicate the influence relationship of the power consumption feature on the decay of the server power supply; a first detection module, configured to detect the remaining service life of the server power supply at a target moment after the current moment according to the decay feature and the power consumption feature.

[0007] The present application also provides an electronic device, including: a memory, configured to store a computer program; a processor, configured to implement the steps of any one of the above server power supply detection methods when executing the computer program.

[0008] The present application also provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and the computer program, when executed by a processor, implements the steps of any one of the above server power supply detection methods.

[0009] The present application also provides a computer program product, including a computer program, and the computer program, when executed by a processor, implements the steps of any one of the above server power supply detection methods.

[0010] Through the present application, by performing feature extraction on the multi-dimensional operation data used to characterize the power supply state of the server by the server power supply from multiple dimensions, the power consumption feature of the server used to indicate the temporal variation relationship of the server's demand for electric energy is obtained, so as to realize reflecting the true electric energy usage demand of the server for the server power supply according to the multi-dimensional operation data, thereby comprehensively reflecting the true usage state of the server power supply. Furthermore, a decay feature of the server power supply under the power consumption feature is generated according to the multi-dimensional operation data, and the remaining service life of the server is detected according to the decay feature and the power consumption feature, so as to realize predicting the remaining service life of the server power supply according to the true usage state of the server power supply and the decay situation of the server power supply under the true usage state. Therefore, the technical problem of low detection efficiency of the server power supply in the related art can be solved, and the technical effect of improving the detection efficiency of the server power supply can be achieved. Description of the Drawings

[0011] To more clearly illustrate the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0012] Figure 1 is the hardware structure block diagram of the detection method for the server power supply in the embodiments of the present application;

[0013] Figure 2 is the flowchart of the detection method for the server power supply according to the embodiments of the present application;

[0014] Figure 3 is the flowchart of the power module life prediction according to the embodiments of the present application;

[0015] Figure 4 is the schematic diagram of an optional power module life prediction system for edge computing according to the embodiments of the present application;

[0016] Figure 5 is the structure block diagram of a detection device for the server power supply according to the embodiments of the present application. Detailed implementation manners

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present application.

[0018] It should be noted that in the description of the present application, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. The terms "first", "second", etc. in the present application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0019] To enable those skilled in the art of this technology to better understand the solution of the present application, the following will further elaborate on the present application in combination with the accompanying drawings and specific implementation manners.

[0020] Combined with the specific application environment architecture or specific hardware architecture on which the execution of the detection method for the server power supply depends, the specific application environment architecture or specific hardware architecture will be described herein.

[0021] The method embodiments provided in the embodiments of the present application can be executed in a server device or a similar computing device. Taking the operation on a server device as an example, Figure 1 is the hardware structure block diagram of the detection method for the server power supply in the embodiments of the present application. As Figure 1 shown, the server device may include one or more ( Figure 1 only one is shown in the figure) processors 102 (the processors 102 may include, but are not limited to, processing devices such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Among them, the above server device may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only illustrative and does not limit the structure of the above server device. For example, the server device may further include more or fewer components than Figure 1 shown in the figure, or have a different configuration from Figure 1 shown in the figure.

[0022] The memory 104 can be used to store computer programs. For example, software programs and modules of application software, such as the computer program corresponding to the startup method of the operating system in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, implements the above method. The memory 104 may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely set relative to the processor 102, and these remote memories can be connected to the server device through a network. Examples of the above networks include, but are not limited to, the Internet, enterprise intranets, local area networks, mobile communication networks, and their combinations.

[0023] The transmission device 106 is used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by the communication provider of the server device. In one instance, the transmission device 106 includes a network adapter (abbreviated as NIC), which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 may be a radio frequency (abbreviated as RF) module, which is used to communicate with the Internet wirelessly.

[0024] The embodiments of the present application provide a method for detecting the server power supply. Combining the execution process of the method for detecting the server power supply, the method is described in detail.

[0025] The following is an explanation of the professional terms that appear in this application:

[0026] RUL: Remaining Useful Life, which refers to the estimated time that a power module can continue to operate normally under its current conditions.

[0027] FFT: Fast Fourier Transform, an algorithm used to convert a time-domain signal into a frequency-domain representation, helps extract the main frequency components of the signal.

[0028] LSTM: Long Short-Term Memory, a special type of recurrent neural network (RNN) suitable for processing and predicting long-term dependencies in time series data.

[0029] ML: Machine Learning is a type of technology that uses algorithms to analyze data, learn from it, and make predictions or decisions about unknown data based on the learned knowledge.

[0030] IoT: Internet of Things, refers to the communication network between various physical devices connected by the Internet.

[0031] Edge computing: A distributed computing framework that brings computation and data storage closer to the devices where data is collected, rather than relying on remote data centers. This reduces latency and improves efficiency.

[0032] Lightweight machine learning models: These are machine learning models that are optimized to reduce computing resource requirements and are suitable for deployment in resource-constrained environments, such as mobile devices or IoT edge nodes.

[0033] Arrhenius equation: An empirical formula that describes the relationship between chemical reaction rate and temperature, often used in the field of material aging or life prediction, especially when considering the effects of thermal stress.

[0034] Federated learning: A privacy-preserving technique that allows multiple participants to jointly train a shared global model without directly exchanging local datasets, thereby protecting data privacy.

[0035] Pruning optimization: A model compression technique that reduces model size and computational complexity by removing unimportant weights or neurons in a neural network while maintaining model performance as much as possible.

[0036] Wavelet transform: A mathematical tool used for signal processing and analysis, particularly suitable for decomposing and reconstructing non-stationary signals and can effectively remove noise.

[0037] Isolation Forest Algorithm: An anomaly detection method based on ensemble learning, especially suitable for high-dimensional datasets, which identifies anomaly points by constructing multiple isolation trees.

[0038] Homomorphic Encryption: A form of encryption that allows specific types of computations to be performed on ciphertext, and the resulting decrypted result is the same as that obtained by directly performing the same operation on the plaintext. It is widely used in privacy-preserving data analysis.

[0039] In this embodiment, a method for detecting a server power supply is provided. Figure 2 It is a flowchart of the method for detecting a server power supply according to an embodiment of the present application, as Figure 2 shown. The method includes the following steps:

[0040] Step S202: Obtain multi-dimensional operation data of the server power supply, where the multi-dimensional operation data is used to characterize the power supply state of the server by the server power supply during a reference time period before the current moment from multiple feature dimensions;

[0041] Step S204: Extract features from the multi-dimensional operation data to obtain the power consumption characteristics of the server, where the power consumption characteristics are used to indicate the temporal change relationship of the server's power demand;

[0042] Step S206: Generate a decay characteristic of the server power supply under the power consumption characteristics according to the multi-dimensional operation data, where the decay characteristic is used to indicate the influence relationship of the power consumption characteristics on the decay of the server power supply;

[0043] Step S208: Detect the remaining service life of the server power supply at a target moment after the current moment according to the decay characteristic and the power consumption characteristic.

[0044] Through the above steps, by extracting features from the multi-dimensional operation data used to characterize the power supply state of the server by the server power supply from multiple dimensions, the power consumption characteristics of the server used to indicate the temporal change relationship of the server's power demand are obtained, so as to realize reflecting the true power consumption demand of the server for the server power supply according to the multi-dimensional operation data, so as to comprehensively reflect the true usage state of the server power supply. Furthermore, a decay characteristic of the server power supply under the power consumption characteristics is generated according to the multi-dimensional operation data, and the remaining service life of the server is detected according to the decay characteristic and the power consumption characteristic, so as to realize predicting the remaining service life of the server power supply according to the true usage state of the server power supply and the decay situation of the server power supply under the true usage state. Therefore, the technical problem of low detection efficiency of the server power supply in the related art can be solved, and the technical effect of improving the detection efficiency of the server power supply can be achieved.

[0045] The detection method of the server power supply provided by this application can be, but is not limited to, applied to devices with the function of detecting the service life of the server power supply, such as servers, or power detection devices configured separately for the server. This solution does not make any limitations in this regard.

[0046] In the embodiment provided in step S202, the server power supply is used to supply power to the server. The multi-dimensional operation data of the server power supply can be, but is not limited to, operation parameters that can reflect the power transmission state of the server power supply to the server. The operation data can be, but is not limited to, including the current output value of the server power supply, the voltage output value of the server power supply, the power temperature of the server power supply, etc. This solution does not make any limitations in this regard.

[0047] Optionally, in the embodiment of this application, the operation data of the server power supply can be, but is not limited to, obtained by detecting the operation state of the server power supply through a data detection device deployed on the server power supply. For example: using a DS18B20 digital temperature sensor to monitor the internal temperature of the server power supply, obtaining the temperature change of the server power supply during operation, and obtaining the original temperature data; using an ACS712 Hall effect current sensor to monitor the current intensity flowing through the server power supply, capturing the current fluctuation when the server power supply is working, and then obtaining the original current data.

[0048] In the embodiment provided in step S204, the power consumption characteristics are used to indicate the actual usage state of the server for the server power supply. The implementation method of extracting the power consumption characteristics of the server from the multi-dimensional operation data can be, but is not limited to, including: extracting the characteristics of the operation data of each characteristic dimension to obtain the operation characteristics corresponding to each characteristic dimension, where the operation characteristics are used to indicate the power operation state of the server power supply in the corresponding characteristic dimension; splicing the characteristics of multiple operation characteristics and inputting the obtained spliced characteristics into the target prediction model, so as to obtain the power consumption characteristics of the server, where the conversion relationship between the operation characteristics and the power consumption characteristics of the server power supply is recorded in the target prediction model.

[0049] Optionally, in the embodiment of this application, the implementation method of extracting the power consumption characteristics of the server from the multi-dimensional operation data can also be: extracting the characteristics of the operation data of each characteristic dimension to obtain the operation characteristics corresponding to each characteristic dimension, where the operation characteristics are used to indicate the power operation state of the server power supply in the corresponding characteristic dimension; splicing the characteristics of multiple operation characteristics and determining the spliced characteristics as the power consumption characteristics.

[0050] In the embodiment provided in step S206, the method for generating the decay characteristics of the server power supply under the power consumption characteristics based on the multi-dimensional operation data may be, but is not limited to: extracting the target operation data of the target characteristic dimension from the multi-dimensional operation data, where the target characteristic dimension is used to characterize the power supply performance of the server power supply, and the target operation parameters of the target characteristic dimension may be, but are not limited to, the maximum energy storage of the power supply, the maximum power supply of the power supply, the peak value of the power output of the power supply, etc.; converting the target operation data into the power decay information of the server power supply, where the power decay information is used to characterize the time-series change relationship of the power decay amount of the server power supply; performing feature extraction on the power decay information to obtain the decay characteristics corresponding to the power consumption characteristics of the server power supply. In this embodiment, the method for converting the target operation data into the power decay information of the server power supply may be: determining the reference power supply life corresponding to the target operation data from the operation data and the power supply life with a corresponding relationship; calculating the power supply life decay amount of the server power supply at the corresponding moment using the reference power supply life and the initial power supply life of the power supply; constructing the power supply life decay curve of the power supply according to the power supply life decay amounts of the server power supply at each moment within the reference time period, where the power decay information includes the power supply life decay curve. Through the above content, by using the operation data in the multi-dimensional operation data that can reflect the power supply performance of the power supply for feature processing, the decay characteristics of the power supply when the server obtains power from the server power supply according to the power consumption method indicated by the power consumption characteristics are obtained, so that the characteristic dimensions of the power consumption characteristics and the decay characteristics are more matched, and the characteristic correlation between the power consumption characteristics and the decay characteristics is improved.

[0051] Optionally, in the embodiment of the present application, the method for generating the decay characteristics of the server power supply under the power consumption characteristics based on the multi-dimensional operation data may be, but is not limited to: extracting the temperature data of the server power supply within the reference time period from the multi-dimensional operation data, and calculating the decay parameter of the server power supply for the temperature data at each moment using the following calculation method: , where represents the ideal gas constant, is the temperature value in degrees Celsius, is the activation energy, is the frequency factor, is the thermal stress coefficient; then sorting the decay parameters according to the time series relationship to obtain the decay sequence of the server power supply, where the decay sequence is used to indicate the change relationship of the decay amount of the server power supply with time within the reference time period; performing feature extraction on the decay sequence of the server power supply to obtain the decay characteristics corresponding to the power consumption characteristics of the server power supply.

[0052] In the embodiment provided in step S208, the method for detecting the remaining service life of the server power supply at a target time after the current time according to the decay characteristic and the power consumption characteristic may be: inputting the decay characteristic and the power consumption characteristic into a target prediction model to obtain the remaining service life output by the target prediction model, where the target prediction model maintains the association relationship between the decay characteristic, the power consumption characteristic, and the remaining service life. This solution is not limited thereto.

[0053] Optionally, in the embodiment of the present application, there is a corresponding relationship between the decay characteristic and the power consumption characteristic, that is, the decay characteristic reflects the decay characteristic of the server power supply when the server power supply obtains electric energy in the power consumption mode indicated by the power consumption characteristic. Therefore, the method for detecting the remaining service life of the server power supply at a target time after the current time according to the decay characteristic and the power consumption characteristic may also be: performing feature fusion on the decay characteristic and the power consumption characteristic to obtain a fusion characteristic of the server power supply, where the fusion characteristic reflects the association relationship between the power supply characteristic and the decay characteristic of the server power supply to the server, and then inputting the fusion characteristic into the target prediction model to obtain the remaining service life output by the target prediction model, where the target prediction model is for the association relationship between the fusion characteristic and the remaining service life.

[0054] Optionally, in the embodiment of the present application, after detecting the remaining service life of the server power supply at a target time after the current time according to the decay characteristic and the power consumption characteristic, a power supply maintenance operation may be performed on the server power supply based on the detected remaining service life, so as to ensure the reliable operation of the server and avoid the impact on business handling caused by the service life temperature of the server power supply during the process of the server running normally to handle business.

[0055] As an optional implementation manner, the extracting features from the multi-dimensional operation data to obtain the power consumption characteristic of the server includes:

[0056] Extracting features from the operation data of each feature dimension in the multi-dimensional operation data to obtain an initial power supply characteristic, where the initial power supply characteristic is used to indicate the operation state of the server power supply during the reference time period;

[0057] Performing feature conversion on the multiple initial power supply characteristics corresponding to the multi-dimensional operation data to obtain the power consumption characteristic.

[0058] Optionally, in the embodiments of the present application, the initial power supply feature characterizes the power supply state of the server power supply to the server. Then, by performing feature transformation on multiple initial power supply features corresponding to multi-dimensional operation data, a power consumption feature characterizing the usage state of the server power supply by the server is obtained, so as to realize characterizing the usage state of the server power supply by the server through the multi-dimensional operation data of the server power supply.

[0059] Optionally, in the embodiments of the present application, the method of performing feature transformation on multiple initial power supply features to obtain a power consumption feature may be: obtaining a reference feature weight for each initial power supply feature, where each reference feature weight is used to characterize the influence degree of the initial power supply feature in the corresponding dimension on characterizing the power consumption requirement of the server. Then, the reference feature weights are used to perform weighted fusion on the multiple initial power supply features to obtain the power consumption feature. Among them, the reference feature weights may be preset fixed weights, or may also be obtained by calculating weights for the multi-dimensional operation data. This solution does not make a limitation on this.

[0060] Through the above content, by performing feature transformation on the initial power supply features characterizing the operation state of the server power supply, it is realized to reflect the true usage demand of the server for the server power supply according to the multi-dimensional operation data of the server power supply, thereby improving the accuracy of the prediction result of the remaining service life of the server power supply.

[0061] As an optional implementation manner, performing feature transformation on the multiple initial power supply features corresponding to the multi-dimensional operation data to obtain the power consumption feature includes:

[0062] Sorting the initial power supply features of each feature dimension within the reference time period in chronological order to obtain a feature sequence of the operation data of each feature dimension;

[0063] Inputting the multiple feature sequences corresponding to the multi-dimensional operation data into a target transformation model to obtain the power consumption feature output by the target transformation model, where the target transformation model is used to identify the time dependence relationship between the initial power supply features included in the multiple feature sequences and transform the corresponding power consumption feature.

[0064] Optionally, in the embodiments of the present application, the target transformation model is used to extract the time dependence relationship between any features in the multiple feature sequences, and then reflect the true power consumption demand of the server. In the embodiments of the present application, the target transformation model is an algorithm model with a function of extracting feature time dependence relationships, which may be but is not limited to LSTM (Long Short-Term Memory).

[0065] Optionally, in the embodiments of the present application, before the multiple feature sequences are input into the target conversion model, the multiple feature sequences can be first subjected to feature splicing, and the spliced features are input into the target conversion model, so that the target conversion model can extract the temporal dependence relationship between any features in the multiple sequence features, ensuring the accuracy of the obtained power consumption features.

[0066] Through the above content, by arranging the initial power supply features in chronological order and using the target conversion model to extract the temporal dependence relationship of multiple feature sequences, the obtained power consumption features are made more accurate and reliable.

[0067] As an alternative implementation manner, the extracting the running data of each feature dimension in the multi-dimensional running data to obtain the initial power supply features includes:

[0068] Using the wavelet transform algorithm to filter the noise signal carried in the running data to obtain the first data;

[0069] Filtering the first data for outliers through the isolation forest algorithm to obtain the second data;

[0070] Performing feature extraction on the second data to obtain the spectral features of the second data, where the spectral features indicate the fluctuation condition of the data signal corresponding to the feature dimension, and the initial power supply features include the spectral features.

[0071] Through the above content, by using the wavelet transform algorithm to filter the noise signal of the running data and using the isolation forest algorithm for outlier filtering, the accuracy of the second data is ensured. Furthermore, by extracting the spectral features of the second data, the data cycle transformation state of the corresponding running data is reflected through the spectral features.

[0072] As an alternative implementation manner, the performing feature extraction on the second data to obtain the spectral features of the second data includes:

[0073] In the case where the second data is the temperature data of the server power supply during operation, performing Fourier transform on the temperature data to obtain the temperature spectral features in the frequency domain space, where the temperature spectral features are used to indicate the periodic fluctuation state of the operating temperature of the server power supply, and the spectral features include the temperature spectral features;

[0074] In the case where the second data is the current data of the output current of the server power supply, perform the Fourier transform on the current data to obtain the current spectrum feature in the frequency domain space, where the current spectrum feature is used to indicate the periodic fluctuation state of the output current value of the server power supply, and the spectrum feature includes the current spectrum feature.

[0075] As an optional implementation manner, the detecting the remaining service life of the server power supply at a target time after the current time according to the decay feature and the power consumption feature includes:

[0076] Perform feature merging on the decay feature and the power consumption feature to obtain a merged feature;

[0077] Input the merged feature into a target prediction model to obtain the remaining service life of the server power supply output by the target prediction model, where the association relationship between the power consumption feature of the server, the decay feature of the server power supply, and the remaining service life of the server power supply is recorded in the target prediction model.

[0078] Optionally, in the embodiments of the present application, the feature merging of the decay feature and the power consumption feature may be performed by directly performing feature splicing, that is, splicing the power consumption feature and the decay feature to obtain a merged feature. Or alternatively, the feature weights of the power consumption feature and the decay feature may be obtained, where the feature weights are used to indicate the influence of the corresponding features on the prediction of the server power supply life, and then the decay feature and the power consumption feature are weighted and fused using the feature weights to obtain a merged feature.

[0079] Optionally, in the embodiments of the present application, the target prediction model may be, but is not limited to, obtained by training an initial prediction model using reference merged features marked with the remaining service life, and this solution is not limited thereto.

[0080] Through the above content, by performing feature merging on the decay feature and the power consumption feature, and using the target prediction model to predict the merged feature obtained by merging, the prediction of the true remaining life of the server power supply in the future time is realized based on the usage demand feature of the server power supply by the server and the decay feature of the server under this usage demand.

[0081] As an optional implementation manner, the performing feature merging on the decay feature and the power consumption feature to obtain a merged feature includes:

[0082] Obtain a first feature weight of the decay feature and a second feature weight of the power consumption feature, where the first feature weight is used to indicate the influence degree of the decay feature on the detection result of the remaining service life of the power supply, and the second feature weight is used to indicate the influence degree of the power consumption feature on the detection result of the remaining service life of the power supply;

[0083] Perform weighted calculation on the decay feature using the first feature weight, and perform weighted calculation on the power consumption feature using the second feature weight, and perform a summation calculation on the weighted calculation results to obtain the combined feature.

[0084] Optionally, in the embodiments of the present application, the first feature weight and the second feature weight may but are not limited to be preset fixed values, or may also be weight values obtained by calculating the importance degrees of the first feature weight and the second feature weight. This solution does not make any limitations in this regard.

[0085] As an optional implementation manner, after inputting the combined feature into the target prediction model to obtain the remaining service life of the server power supply output by the target prediction model, the method further includes:

[0086] Detect the true service life of the server power supply at the target moment;

[0087] Calculate the difference value between the true service life and the remaining service life;

[0088] Update the model parameters obtained by the target prediction model according to the difference value to obtain the updated target prediction model.

[0089] Optionally, in the embodiments of the present application, after using the target prediction model to predict the remaining service life of the server power supply at the target moment, the true service life of the server power supply at the target moment may also be collected, and the model parameters of the target prediction model are self-updated based on the true service life, so as to ensure the accuracy of the target prediction model. In the embodiments of the present application, the update method of the model parameters of the target prediction model may but is not limited to: input the comprehensive feature vector into the preliminarily deployed model, and perform a complete forward propagation process to obtain the initial remaining service life prediction value of the power supply module; based on the difference between the initial prediction result and the real-time monitored true aging state, calculate the model output error, and calculate the gradient of each layer of weights through the backpropagation algorithm. The expression is: ; where is the learning rate, is the partial derivative of the error with respect to the weight; obtain the adjusted model parameters ; generate a new comprehensive feature vector using the latest temperature raw data and current raw data, and combine it with the adjusted model parameters Together, they are used in the model retraining process to obtain an updated model; the new comprehensive feature vector is input into the updated model, and the forward propagation process is executed to obtain a more accurate predicted value of the remaining service life of the power module.

[0090] As an alternative implementation, updating the model parameters of the target prediction model according to the difference value to obtain a reference prediction model includes:

[0091] Calculating the difference value through the backpropagation algorithm to obtain parameter adjustment information for the model parameters of the target prediction model, where the parameter adjustment information is used to indicate the adjustment method for the model parameters;

[0092] Adjusting the model parameters of the target prediction model according to the parameter adjustment method indicated by the parameter adjustment information to obtain the updated target prediction model.

[0093] Through the above content, the model parameters of the target prediction model are adjusted by using the backpropagation algorithm, so as to realize the dynamic update of the model parameters of the target prediction model, thereby ensuring the reliability of the output result of the target prediction model.

[0094] As an alternative implementation, the server system includes multiple such servers, and each server is correspondingly configured with a target prediction model. The method further includes:

[0095] Obtaining the current target model parameters of each target prediction model, where the target model parameters are used to indicate the correlation relationship among the power consumption characteristics of the server, the decay characteristics of the server power supply, and the remaining service life of the server power supply;

[0096] Converting the multiple target model parameters into candidate model parameters of the target prediction model;

[0097] Sending the candidate model parameters to each target prediction model, where the target prediction model is used to update the model parameters currently used by the target prediction model by using the candidate model parameters.

[0098] Optionally, in the embodiments of the present application, the above method for detecting the server power supply can be applied to a server system. The server system includes multiple servers. By deploying a target prediction model for each server, the prediction of the remaining service life of the corresponding server is realized, and the model parameters of the multiple target prediction models are dynamically updated in the server system by means of federated learning.

[0099] Optionally, in the embodiments of the present application, the above-mentioned server power detection method can be applied to a server system. The server system includes multiple servers. By deploying a target prediction model for each server, the remaining useful life of the corresponding server can be predicted. To better execute the above-mentioned server power detection method, an edge device for executing the above-mentioned server power detection method can be deployed for each server, and the target prediction model can be run through this edge device to achieve dynamic prediction of the remaining useful life of the server power deployed on the corresponding server. Further, a cloud server can be configured for the edge nodes of all servers in the server system to update and maintain the model parameters of the target prediction models on each edge node. The target prediction model deployed on the edge node can be, but is not limited to, a lightweight machine learning model optimized by pruning, and the initial prediction result is used for update and correction, and the model parameters are adjusted based on the latest sensing data to obtain an accurate RUL prediction value; the model parameters on multiple edge devices are aggregated to the cloud through the federated learning mechanism, and global model optimization is performed on the cloud and feedback is given to each edge device. The specific steps are as follows: Run the federated learning client program on each edge device; the federated learning client program is responsible for collecting the model parameters of the local lightweight machine learning model after multiple updates to obtain a local model parameter set ; Apply homomorphic encryption technology to the model parameters on each edge device for encryption to generate encrypted model parameters , obtaining an encrypted model parameter set ; The cloud server receives the encrypted model parameters sent by all edge devices , and restores the original model parameters through the corresponding decryption algorithm ; Use the weighted average method to aggregate all model parameters to obtain global model parameters, and the expression is: ; where represents the importance weight of the th edge device, is the aggregated global model parameter; Run the global optimization algorithm on the cloud server to fine-tune the global model parameters to obtain optimized global model parameters ; The cloud server transmits the optimized global model parameters back to each edge device. After the edge device receives , it replaces the current local model parameters , thereby updating the local model to a new local model; Adopt a feedback mechanism to regularly check and update the global model parameters, and continuously provide accurate predictions of the remaining useful life of the power module to obtain the final accurate RUL prediction value.

[0100] Through the above content, by configuring a target prediction model for each of multiple servers in the server system and updating the model parameters of the multiple target prediction models in a summarized manner, it is possible to achieve that on the one hand, the multiple target prediction models can perform self-updating of the model parameters, and after the target prediction model self-updates the model parameters, the model parameters of the multiple target prediction models can be obtained for summarized updating, thereby ensuring the accuracy and reliability of the model parameters of the multiple target prediction models in the server system and avoiding extreme updates of the model parameters of the model.

[0101] As an optional implementation manner, the converting the multiple target model parameters into candidate model parameters of the target prediction model includes:

[0102] Performing weighted summation on the multiple target model parameters using the model weights of each target prediction model to obtain a model parameter sum value, where the model weights are used to indicate the importance degree of the model parameters of the corresponding target prediction model;

[0103] Calculating the quotient of the model parameter sum value and the number of models of the target prediction model to obtain an average model parameter value;

[0104] Sending the average model parameter value to each target prediction model, where the target prediction model is used to update the model parameters of the target prediction model using the average model parameter value.

[0105] Optionally, in the embodiments of the present application, when sending the average model parameter value to the target prediction model, the average model parameter value can be encrypted and transmitted using the homomorphic encryption method. Further, after the target prediction model receives the encrypted average model parameter value, the encrypted model parameters can be restored to obtain the average model parameter value.

[0106] As an optional implementation manner, the generating the decay characteristics of the server power supply under the power consumption characteristics according to the multi-dimensional operation data includes:

[0107] Calculating the temperature data included in the multi-dimensional operation data using the following formula to obtain the decay characteristics:

[0108] ;

[0109] Wherein, is the gas constant, is the temperature value in degrees Celsius, is the activation energy, is the frequency factor, is the thermal stress coefficient, and the decay characteristics include the thermal stress coefficient.

[0110] As an alternative embodiment, the present application provides a method for predicting the life of a power module based on edge computing. Figure 3 It is a flowchart for predicting the life of a power module according to an embodiment of the present application, which includes:

[0111] Use sensors to monitor the working state of the power module to obtain original sensing data;

[0112] Preprocess the original sensing data to obtain preprocessed sensing data;

[0113] Use the Fast Fourier Transform (FFT) and Long Short-Term Memory (LSTM) network to extract features from the preprocessed sensing data, capture the spectral characteristics and time series characteristics of the signal, and obtain key feature vectors;

[0114] Based on the Arrhenius equation and combined with lightweight machine learning, construct a prediction model, input the key feature vectors into the prediction model, and output an initial prediction result;

[0115] Deploy a lightweight machine learning model optimized by pruning on the edge device, and use the initial prediction result for update and correction. Adjust the model parameters based on the latest sensing data to obtain an accurate Remaining Useful Life (RUL) prediction value;

[0116] Through the federated learning mechanism, summarize the model parameters on multiple edge devices to the cloud, and perform global model optimization on the cloud and feedback it to each edge device.

[0117] As a preferred solution of the method for predicting the life of a power module based on edge computing according to the present application, wherein: the step of using sensors to monitor the working state of the power module to obtain original sensing data is specifically as follows:

[0118] Use a DS18B20 digital temperature sensor to monitor the internal temperature of the power module, obtain the temperature change of the power module during operation, and obtain temperature original data;

[0119] Use an ACS712 Hall effect current sensor to monitor the current intensity flowing through the power module, capture the current fluctuations when the power module is working, and then obtain current original data;

[0120] Summarize the temperature original data and the current original data into complete original sensing data.

[0121] As a preferred solution of the method for predicting the life of a power module based on edge computing according to the present application, wherein: the step of preprocessing the original sensing data to obtain preprocessed sensing data is specifically as follows:

[0122] The original sensing data set is denoised by using wavelet transform. The signal is decomposed by selecting the wavelet basis function db4, and the high-frequency noise components are filtered out to obtain the denoised sensing data;

[0123] The isolated forest algorithm is used to perform anomaly detection on the denoised sensing data. An anomaly score threshold is set to identify and remove the abnormal data points to ensure the quality of the data set, and the preprocessed sensing data is obtained.

[0124] As a preferred solution of the power module life prediction method based on edge computing described in this application, wherein: the fast Fourier transform FFT and the long short-term memory network LSTM are used to extract features from the preprocessed sensing data, capture the spectral characteristics and time series characteristics of the signal, and obtain the key feature vector. The specific steps are as follows:

[0125] Apply the fast Fourier transform to the preprocessed temperature data, convert it into a frequency domain representation form, extract the main frequency components as features, and obtain the temperature spectrum features;

[0126] Apply the fast Fourier transform to the preprocessed current data, convert it into a frequency domain representation form, extract the main frequency components as features, and obtain the current spectrum features;

[0127] Use the long short-term memory network LSTM to perform time series analysis on the temperature spectrum features and the current spectrum features, capture the trends and patterns of the data changing over time, and obtain the key feature vector containing time series information.

[0128] As a preferred solution of the power module life prediction method based on edge computing described in this application, wherein: a prediction model is constructed based on the Arrhenius equation and combined with lightweight machine learning. The key feature vector is input into the prediction model, and the initial prediction result is output. The specific steps are as follows:

[0129] The Arrhenius equation is used to calculate the thermal stress coefficient of the preprocessed temperature data. The activation energy and frequency factor are adjusted according to the ambient temperature to obtain the thermal stress coefficient. The expression is:

[0130] ;

[0131] Wherein, represents the ideal gas constant, is the temperature value in degrees Celsius, is the activation energy, is the frequency factor, is the thermal stress coefficient;

[0132] Input the key feature vector into the pre-trained LSTM model;

[0133] The LSTM model is used to identify long-term dependencies in the data, extract patterns related to the aging of the power module, and obtain a time series feature representation.

[0134] Taking the thermal stress coefficient and the time series feature representation as inputs, they are input into a lightweight machine learning model optimized by pruning. The two features are integrated in a weighted average manner to obtain a comprehensive feature vector, and the expression is:

[0135] ;

[0136] Where, is the time series feature, is the thermal stress coefficient, and are the weight coefficients;

[0137] Inputting the comprehensive feature vector into the lightweight machine learning model, through the application and adjustment of the internal parameters of the model, the initial predicted remaining useful life value of the power module is output, and the initial prediction result is obtained.

[0138] As a preferred solution of the power module life prediction method based on edge computing described in the application, wherein: deploying the lightweight machine learning model optimized by pruning on the edge device, and using the initial prediction result for update and correction, adjusting the model parameters based on the latest sensing data to obtain an accurate RUL prediction value. The specific steps are as follows:

[0139] Optimizing the lightweight machine learning model MobileNet through pruning technology and deploying it on the edge device to obtain a preliminary deployment model;

[0140] Inputting the comprehensive feature vector into the preliminarily deployed model, and performing a complete forward propagation process to obtain the initial predicted remaining useful life value of the power module;

[0141] Based on the difference between the initial prediction result and the real aging state monitored in real time, calculating the model output error, and calculating the gradient of each layer of weights through the backpropagation algorithm. The expression is:

[0142] ;

[0143] Where, is the learning rate, is the partial derivative of the error with respect to the weight;

[0144] Obtaining the adjusted model parameters ;

[0145] Generating a new comprehensive feature vector using the latest temperature raw data and current raw data, and combining it with the adjusted model parameters be used together in the model retraining process to obtain an updated model;

[0146] Input the new comprehensive feature vector into the updated model and perform the forward propagation process to obtain a more accurate predicted value of the remaining service life of the power module.

[0147] As a preferred solution of the power module life prediction method based on edge computing according to the present invention, wherein: the model parameters on multiple edge devices are aggregated to the cloud through the federated learning mechanism, and global model optimization is performed on the cloud and feedback is given to each edge device. The specific steps are as follows:

[0148] Run the federated learning client program on each edge device;

[0149] The federated learning client program is responsible for collecting the model parameters after multiple updates of the local lightweight machine learning model to obtain a local model parameter set ;

[0150] Apply the homomorphic encryption technology to the model parameters on each edge device for encryption to generate encrypted model parameters to obtain an encrypted model parameter set ;

[0151] The cloud server receives the encrypted model parameters sent by all edge devices and restores the original model parameters through the corresponding decryption algorithm ;

[0152] Use the weighted average method to aggregate all model parameters to obtain global model parameters. The expression is:

[0153] ;

[0154] wherein, represents the importance weight of the th edge device, is the aggregated global model parameter;

[0155] Run the global optimization algorithm on the cloud server to fine-tune the global model parameters to obtain optimized global model parameters ;

[0156] The cloud server transmits the optimized global model parameters back to each edge device. After the edge device receives it, it replaces the current local model parameters to update the local model to a new local model;

[0157] A feedback mechanism is adopted to regularly check and update the global model parameters, and continuously provide an accurate prediction of the remaining service life of the power module, so as to obtain the final accurate RUL prediction value.

[0158] This application provides a power module life prediction system based on edge computing, including:

[0159] A data acquisition module, a data preprocessing module, a feature extraction module, a model construction module, a model update module, and a federated learning module;

[0160] The data acquisition module is used to monitor the internal temperature change of the power module by using a DS18B20 digital temperature sensor to obtain the original temperature data. At the same time, an ACS712 Hall effect current sensor is used to monitor the current intensity flowing through the power module, capture the current fluctuation, and summarize it into complete original sensing data;

[0161] The data preprocessing module is used to perform denoising processing on the original sensing data set by using wavelet transform. The signal is decomposed by selecting the wavelet basis function db4 to filter out the high-frequency noise components. Then, the isolation forest algorithm is used to identify and remove the abnormal data points to ensure the data quality, and the preprocessed sensing data is obtained;

[0162] The feature extraction module is used to apply the fast Fourier transform to the preprocessed temperature data and current data, convert them into a frequency domain representation form, extract the main frequency components as features, and then use the long short-term memory network to perform time series analysis on the temperature spectrum features and current spectrum features to capture the trends and patterns changing with time, and obtain the key feature vectors containing time series information;

[0163] The model construction module is used to calculate the thermal stress coefficient according to the Arrhenius equation, and combine it with a lightweight machine learning model to construct a prediction model, input the key feature vectors into the model, and output the initial prediction result;

[0164] The model update module is used to deploy the lightweight machine learning model optimized by pruning on the edge device, and use the initial prediction result for update and correction, adjust the model parameters based on the latest sensing data, and execute the forward propagation process to obtain a more accurate prediction value of the remaining service life of the power module;

[0165] The federated learning module is used to run the federated learning client program on each edge device, collect the locally trained model parameters, encrypt them by using the homomorphic encryption technology and send them to the cloud. The cloud server receives the encrypted model parameters, decrypts them, and uses the weighted average method to summarize all the model parameters to obtain the global model parameters.

[0166] Figure 4It is a schematic diagram of an optional power module life prediction system for edge computing according to an embodiment of the present application. As Figure 4 shown, it includes the following steps:

[0167] S1. Use sensors to monitor the working state of the power module to obtain original sensing data;

[0168] Furthermore, use a DS18B20 digital temperature sensor to monitor the internal temperature of the power module, obtain the temperature change of the power module during operation, and get the original temperature data;

[0169] Use an ACS712 Hall effect current sensor to monitor the current intensity flowing through the power module, capture the current fluctuations when the power module is working, and then obtain the original current data;

[0170] Summarize the original temperature data and the original current data into complete original sensing data;

[0171] It should be noted that by selecting high-precision DS18B20 digital temperature sensors and ACS712 Hall effect current sensors, it can ensure that the collected data has high accuracy and reliability. The high-precision data provides a solid foundation for subsequent preprocessing and feature extraction, and helps to improve the accuracy of the final life prediction.

[0172] S2. Preprocess the original sensing data to obtain preprocessed sensing data;

[0173] Furthermore, use wavelet transform to denoise the original sensing data set, decompose the signal by selecting the wavelet basis function db4, filter out the high-frequency noise components, and obtain the denoised sensing data;

[0174] Use the isolation forest algorithm to perform anomaly detection on the denoised sensing data, set an anomaly score threshold to identify and remove abnormal data points, ensure the quality of the data set, and obtain the preprocessed sensing data;

[0175] It should be noted that the application of wavelet transform and isolation forest algorithm not only effectively removes noise and outliers, but also retains the key information in the data. This step is crucial for subsequent feature extraction and model training because it directly affects the quality of the model input, thereby affecting the accuracy of the prediction results.

[0176] S3. Use the fast Fourier transform FFT and the long short-term memory network LSTM to extract features from the preprocessed sensing data, capture the spectral characteristics and time series characteristics of the signal, and obtain the key feature vectors;

[0177] Further, apply the fast Fourier transform to the preprocessed temperature data, convert it into a frequency-domain representation form, extract the main frequency components as features, and obtain the temperature spectrum features;

[0178] Apply the fast Fourier transform to the preprocessed current data, convert it into a frequency-domain representation form, extract the main frequency components as features, and obtain the current spectrum features;

[0179] Use the long short-term memory network LSTM to perform time series analysis on the temperature spectrum features and current spectrum features, capture the trends and patterns of the data changing over time, and obtain the key feature vectors containing time series information;

[0180] It should be noted that the combined use of FFT and LSTM can comprehensively capture the change characteristics of the power module working state, considering both the frequency-domain characteristics of the signal and its timing characteristics. The method not only improves the comprehensiveness of feature extraction but also enhances the model's understanding ability of the power module state under different working conditions, thus improving the accuracy and reliability of the prediction.

[0181] S4. Based on the Arrhenius equation and combined with lightweight machine learning, construct a prediction model, input the key feature vectors into the prediction model, and output the initial prediction result;

[0182] Further, use the Arrhenius equation to calculate the thermal stress coefficient of the preprocessed temperature data, adjust the activation energy and frequency factor according to the ambient temperature, and obtain the thermal stress coefficient. The expression is:

[0183] ;

[0184] Among them, represents the ideal gas constant, is the temperature value in degrees Celsius, is the activation energy, is the frequency factor, is the thermal stress coefficient;

[0185] Input the key feature vectors into the pre-trained LSTM model;

[0186] The LSTM model is used to identify the long-term dependencies in the data, extract the patterns related to the aging of the power module, and obtain the time series feature representation;

[0187] Take the thermal stress coefficient and the time series feature representation as inputs, input them into the lightweight machine learning model optimized by pruning, and use the weighted average method to integrate the two features to obtain the comprehensive feature vector. The expression is:

[0188] ;

[0189] Among them, is the time series feature, is the thermal stress coefficient, and are the weight coefficients;

[0190] Input the comprehensive feature vector into the lightweight machine learning model. By applying and adjusting the internal parameters of the model, output the initial predicted remaining useful life value of the power supply module to obtain the initial prediction result;

[0191] It should be noted that the combination of the Arrhenius equation and the lightweight machine learning model makes full use of the advantages of physical knowledge and data-driven methods. In this way, not only can the influence of environmental temperature on the aging of the power supply module be explained, but also the model parameters can be dynamically adjusted to adapt to different working conditions, so as to provide a more scientific and accurate prediction of the remaining useful life.

[0192] S5. Deploy the pruned and optimized lightweight machine learning model on the edge device, and use the initial prediction result for update and correction. Adjust the model parameters based on the latest sensing data to obtain the accurate RUL prediction value;

[0193] Furthermore, optimize the lightweight machine learning model MobileNet through pruning technology and deploy it on the edge device to obtain the preliminary deployment model;

[0194] Input the comprehensive feature vector into the preliminarily deployed model, and perform a complete forward propagation process to obtain the initial predicted remaining useful life value of the power supply module;

[0195] Based on the difference between the initial prediction result and the real aging state monitored in real time, calculate the model output error, and calculate the gradient of each layer of weights through the backpropagation algorithm. The expression is:

[0196] ;

[0197] Among them, is the learning rate, is the partial derivative of the error with respect to the weight;

[0198] Obtain the adjusted model parameters ;

[0199] Generate a new comprehensive feature vector using the latest temperature raw data and current raw data, and use it together with the adjusted model parameters for the model retraining process to obtain the updated model;

[0200] Input the new comprehensive feature vector into the updated model, and perform the forward propagation process to obtain a more accurate predicted value of the remaining useful life of the power module;

[0201] It should be noted that the lightweight machine learning model optimized by pruning can operate efficiently on resource-constrained edge devices while maintaining high prediction accuracy. By continuously updating and correcting the model parameters, the model can better adapt to the changes in actual working conditions, thereby continuously providing accurate predicted values of the remaining useful life.

[0202] S6. Aggregate the model parameters on multiple edge devices to the cloud through the federated learning mechanism, and perform global model optimization on the cloud and feedback it to each edge device;

[0203] Furthermore, run the federated learning client program on each edge device;

[0204] The federated learning client program is responsible for collecting the model parameters of the local lightweight machine learning model after multiple updates to obtain the local model parameter set ;

[0205] Apply homomorphic encryption technology to the model parameters on each edge device for encryption to generate the encrypted model parameters to obtain the encrypted model parameter set ;

[0206] The cloud server receives the encrypted model parameters sent by all edge devices , and restores the original model parameters through the corresponding decryption algorithm ;

[0207] Use the weighted average method to aggregate all model parameters to obtain the global model parameters. The expression is:

[0208] ;

[0209] Among them, represents the importance weight of the th edge device, and is the aggregated global model parameter;

[0210] Run the global optimization algorithm on the cloud server to fine-tune the global model parameters to obtain the optimized global model parameters ;

[0211] The cloud server transmits the optimized global model parameters back to each edge device. After the edge device receives it, it replaces the current local model parameters , thereby updating the local model to a new local model;

[0212] Adopt a feedback mechanism to regularly check and update the global model parameters, and continuously provide an accurate prediction of the remaining service life of the power module to obtain the final accurate RUL prediction value;

[0213] It should be noted that the federated learning mechanism not only protects the data privacy of each edge device, but also significantly improves the generalization ability and prediction accuracy of the model through global model optimization. The regular model update and feedback mechanism ensure that all edge devices can obtain the latest global model parameters in a timely manner, thereby realizing continuous and accurate prediction of the remaining service life of the power module. The method not only improves the overall performance of the system, but also provides strong support for maintenance and management in practical applications.

[0214] This embodiment also provides a power module life prediction system based on edge computing, including:

[0215] A data acquisition module, a data preprocessing module, a feature extraction module, a model construction module, a model update module, and a federated learning module;

[0216] The data acquisition module is used to monitor the internal temperature change of the power module by using a DS18B20 digital temperature sensor to obtain the original temperature data. At the same time, an ACS712 Hall effect current sensor is used to monitor the current intensity flowing through the power module, capture the current fluctuation, and summarize it into complete original sensing data;

[0217] The data preprocessing module is used to perform denoising processing on the original sensing data set by using wavelet transform, decompose the signal by selecting the wavelet basis function db4, filter out the high-frequency noise components, and then use the isolation forest algorithm to identify and remove the abnormal data points to ensure the data quality and obtain the preprocessed sensing data;

[0218] The feature extraction module is used to apply the fast Fourier transform to the preprocessed temperature data and current data, convert them into a frequency domain representation form, extract the main frequency components as features, and then use a long short-term memory network to perform time series analysis on the temperature spectrum features and current spectrum features to capture the trends and patterns changing with time, and obtain the key feature vectors containing time series information;

[0219] The model construction module is used to calculate the thermal stress coefficient according to the Arrhenius equation, and combine it with a lightweight machine learning model to construct a prediction model, input the key feature vectors into the model, and output the initial prediction result;

[0220] The model update module is used to deploy a lightweight machine learning model optimized by pruning on the edge device, and use the initial prediction results for update and correction, adjust the model parameters based on the latest sensing data, and perform the forward propagation process to obtain a more accurate predicted value of the remaining service life of the power module.

[0221] The federated learning module is used to run the federated learning client program on each edge device, collect the model parameters after local training, encrypt them using homomorphic encryption technology and send them to the cloud. The cloud server receives the encrypted model parameters, decrypts them and uses the weighted average method to aggregate all the model parameters to obtain the global model parameters.

[0222] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0223] An embodiment of the present application also provides a detection device for a server power supply. Figure 5 It is a structural block diagram of a detection device for a server power supply according to an embodiment of the present application, as Figure 5 shown. The device includes:

[0224] The acquisition module is used to acquire multi-dimensional operation data of the server power supply, where the multi-dimensional operation data is used to characterize the power supply state of the server by the server power supply during a reference time period before the current moment from multiple feature dimensions.

[0225] The extraction module is used to extract features from the multi-dimensional operation data to obtain the power consumption characteristics of the server, where the power consumption characteristics are used to indicate the temporal change relationship of the server's power demand.

[0226] The generation module is used to generate a decay characteristic of the server power supply under the power consumption characteristic according to the multi-dimensional operation data, where the decay characteristic is used to indicate the influence relationship of the power consumption characteristic on the decay of the server power supply.

[0227] The first detection module is used to detect the remaining service life of the server power supply at a target moment after the current moment according to the decay characteristic and the power consumption characteristic.

[0228] Through the above device, by extracting features from multi-dimensional operation data used to characterize the power supply state of the server power supply to the server from multiple dimensions, the power consumption characteristics of the server for indicating the temporal change relationship of the power demand of the server are obtained, so as to realize reflecting the true power consumption demand of the server for the server power supply according to the multi-dimensional operation data, thereby comprehensively reflecting the true usage state of the server power supply. Furthermore, according to the multi-dimensional operation data, the decay characteristics of the server power supply under the power consumption characteristics are generated, and the remaining service life of the server is detected according to the decay characteristics and the power consumption characteristics, so as to realize predicting the remaining service life of the server power supply according to the true usage state of the server power supply and the decay situation of the server power supply under the true usage state. Therefore, the technical problem of low detection efficiency of the server power supply in the related art can be solved, and the technical effect of improving the detection efficiency of the server power supply can be achieved.

[0229] Optionally, the extraction module includes:

[0230] An extraction unit, configured to extract features from the operation data of each feature dimension in the multi-dimensional operation data to obtain an initial power supply feature, where the initial power supply feature is used to indicate the operation state of the server power supply within the reference time period;

[0231] A conversion unit, configured to perform feature conversion on the multiple initial power supply features corresponding to the multi-dimensional operation data to obtain the power consumption feature.

[0232] Optionally, the conversion unit is configured to:

[0233] Sort the initial power supply features of each feature dimension within the reference time period in chronological order to obtain a feature sequence of the operation data of each feature dimension;

[0234] Input the multiple feature sequences corresponding to the multi-dimensional operation data into a target conversion model to obtain the power consumption feature output by the target conversion model, where the target conversion model is used to identify the temporal dependence relationship between the initial power supply features included in the multiple feature sequences and convert the corresponding power consumption feature.

[0235] Optionally, the extraction unit is configured to:

[0236] Use the wavelet transform algorithm to filter the noise signals carried in the operation data to obtain first data;

[0237] Perform outlier filtering on the first data through the isolation forest algorithm to obtain second data;

[0238] Extract features from the second data to obtain the spectral features of the second data, where the spectral features indicate the fluctuation of the data signal corresponding to the feature dimension, and the initial power supply features include the spectral features.

[0239] Optionally, the extraction unit is configured to:

[0240] When the second data is the temperature data of the server power supply during operation, perform a Fourier transform on the temperature data to obtain the temperature spectral features in the frequency domain space, where the temperature spectral features are used to indicate the periodic fluctuation state of the operating temperature of the server power supply, and the spectral features include the temperature spectral features;

[0241] When the second data is the current data of the output current of the server power supply, perform the Fourier transform on the current data to obtain the current spectral features in the frequency domain space, where the current spectral features are used to indicate the periodic fluctuation state of the output current value of the server power supply, and the spectral features include the current spectral features.

[0242] Optionally, the first detection module includes:

[0243] A merging unit configured to perform feature merging on the decay feature and the power consumption feature to obtain a merged feature;

[0244] An input unit configured to input the merged feature into a target prediction model to obtain the remaining service life of the server power supply output by the target prediction model, where the association relationship between the power consumption feature of the server, the decay feature of the server power supply, and the remaining service life of the server power supply is recorded in the target prediction model.

[0245] Optionally, the merging unit is configured to:

[0246] Obtain a first feature weight of the decay feature and a second feature weight of the power consumption feature, where the first feature weight is used to indicate the influence degree of the decay feature on the detection result of the remaining service life of the power supply, and the second feature weight is used to indicate the influence degree of the power consumption feature on the detection result of the remaining service life of the power supply;

[0247] Perform weighted calculation on the decay feature using the first feature weight, and perform weighted calculation on the power consumption feature using the second feature weight, and perform a summation calculation on the weighted calculation results to obtain the merged feature.

[0248] Optionally, the device further includes:

[0249] A second detection module, configured to detect the true service life of the server power supply at the target moment after inputting the combined feature into the target prediction model to obtain the remaining service life of the server power supply output by the target prediction model;

[0250] A calculation module, configured to calculate a difference value between the true service life and the remaining service life;

[0251] An update module, configured to update the model parameters obtained by the target prediction model according to the difference value to obtain the updated target prediction model.

[0252] Optionally, the update module includes:

[0253] A calculation unit, configured to calculate the difference value through a backpropagation algorithm to obtain parameter adjustment information for the model parameters of the target prediction model, where the parameter adjustment information is used to indicate the adjustment method for the model parameters;

[0254] An adjustment unit, configured to adjust the model parameters of the target prediction model according to the parameter adjustment method indicated by the parameter adjustment information to obtain the updated target prediction model.

[0255] Optionally, a plurality of the servers are included in the server system, and each server is correspondingly configured with a target prediction model. The apparatus further includes:

[0256] An acquisition module, configured to acquire the current target model parameters of each target prediction model, where the target model parameters are used to indicate the correlation relationship among the power consumption characteristics of the server, the decay characteristics of the server power supply, and the remaining service life of the server power supply;

[0257] A conversion module, configured to convert candidate model parameters of the target prediction model according to the plurality of target model parameters;

[0258] A sending module, configured to send the candidate model parameters to each target prediction model, where the target prediction model is used to update the model parameters currently used by the target prediction model using the candidate model parameters.

[0259] Optionally, the conversion module includes:

[0260] A processing unit, configured to perform weighted summation on the plurality of target model parameters using the model weights of each target prediction model to obtain a model parameter sum value, where the model weights are used to indicate the importance degree of the model parameters of the corresponding target prediction model;

[0261] A calculation unit for calculating the quotient of the model parameters and values and the number of models of the target prediction model to obtain the average model parameter value.

[0262] A sending unit for sending the average model parameter value to each of the target prediction models, where the target prediction model is used to update the model parameters of the target prediction model using the average model parameter value.

[0263] Optionally, the generation module includes:

[0264] A calculation unit for calculating the decay characteristics by calculating the temperature data included in the multi-dimensional operation data using the following formula:

[0265] ;

[0266] Where is the gas constant, is the temperature value in degrees Celsius, is the activation energy, is the frequency factor, is the thermal stress coefficient, and the decay characteristics include the thermal stress coefficient.

[0267] For the description of the features in the corresponding embodiment of the detection device of the server power supply, reference can be made to the relevant description in the corresponding embodiment of the detection method of the server power supply, which will not be elaborated here one by one.

[0268] An embodiment of the present application also provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any of the above-mentioned embodiments of the detection method of the server power supply.

[0269] An embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps in any of the above-mentioned embodiments of the detection method of the server power supply when running.

[0270] In an exemplary embodiment, the above-mentioned computer-readable storage medium may include, but is not limited to: USB flash drives, read-only memories (ROM for short), random access memories (RAM for short), mobile hard disks, magnetic disks, or optical disks and other media that can store computer programs.

[0271] An embodiment of the present application also provides a computer program product. The above-mentioned computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps in any of the above-mentioned embodiments of the detection method of the server power supply.

[0272] An embodiment of the present application further provides another computer program product, including a non-volatile computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in any of the above-described embodiments of the server power supply detection method are implemented.

[0273] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0274] The above has introduced in detail a method and device for detecting a server power supply, a storage medium, and an electronic device provided by the present application. Specific examples are used herein to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application. It should be noted that for those of ordinary skill in the art in the technical field, without departing from the principle of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A detection method for a server power supply, characterized in that, Including: Obtain multi-dimensional operation data of the server power supply, where the multi-dimensional operation data is used to characterize the power supply state of the server by the server power supply within a reference time period before the current moment from multiple feature dimensions; Extract features from the multi-dimensional operation data to obtain the power consumption feature of the server, where the power consumption feature is used to indicate the temporal change relationship of the server's power demand; Generate a decay feature of the server power supply under the power consumption feature according to the multi-dimensional operation data, where the decay feature is used to indicate the influence relationship of the power consumption feature on the decay of the server power supply; Detect the remaining service life of the server power supply at a target moment after the current moment according to the decay feature and the power consumption feature.

2. The method according to claim 1, wherein: The extracting features from the multi-dimensional operation data to obtain the power consumption feature of the server includes: Extract features from the operation data of each feature dimension in the multi-dimensional operation data to obtain an initial power supply feature, where the initial power supply feature is used to indicate the operation state of the server power supply within the reference time period; Perform feature transformation on the multiple initial power supply features corresponding to the multi-dimensional operation data to obtain the power consumption feature.

3. The method according to claim 2, wherein: The performing feature transformation on the multiple initial power supply features corresponding to the multi-dimensional operation data to obtain the power consumption feature includes: Sort the initial power supply features of each feature dimension within the reference time period in chronological order to obtain a feature sequence of the operation data of each feature dimension; Input the multiple feature sequences corresponding to the multi-dimensional operation data into a target transformation model to obtain the power consumption feature output by the target transformation model, where the target transformation model is used to identify the temporal dependence relationship between the initial power supply features included in the multiple feature sequences and transform the corresponding power consumption feature.

4. The method according to claim 2, wherein: The extracting features from the operation data of each feature dimension in the multi-dimensional operation data to obtain an initial power supply feature includes: Use a wavelet transform algorithm to filter the noise signal carried in the operation data to obtain a first data; Filter outliers from the first data through an isolation forest algorithm to obtain a second data; Extract features from the second data to obtain the spectral feature of the second data, where the spectral feature indicates the fluctuation of the data signal of the corresponding feature dimension, and the initial power supply feature includes the spectral feature.

5. The method according to claim 4, wherein: The extracting features from the second data to obtain the spectral feature of the second data includes: When the second data is the temperature data of the server power supply during operation, perform Fourier transform on the temperature data to obtain the temperature spectrum characteristics in the frequency domain space, where the temperature spectrum characteristics are used to indicate the periodic fluctuation state of the operating temperature of the server power supply, and the spectrum characteristics include the temperature spectrum characteristics; When the second data is the current data of the output current of the server power supply, perform the Fourier transform on the current data to obtain the current spectrum characteristics in the frequency domain space, where the current spectrum characteristics are used to indicate the periodic fluctuation state of the output current value of the server power supply, and the spectrum characteristics include the current spectrum characteristics.

6. The method according to claim 1, wherein: The detecting the remaining service life of the server power supply at a target time after the current time according to the decay characteristics and the power consumption characteristics includes: Performing feature merging on the decay characteristics and the power consumption characteristics to obtain a merged feature; Inputting the merged feature into a target prediction model to obtain the remaining service life of the server power supply output by the target prediction model, where the association relationship between the power consumption characteristics of the server, the decay characteristics of the server power supply, and the remaining service life of the server power supply is recorded in the target prediction model.

7. The method according to claim 6, wherein: The performing feature merging on the decay characteristics and the power consumption characteristics to obtain a merged feature includes: Obtaining a first feature weight of the decay characteristics and a second feature weight of the power consumption characteristics, where the first feature weight is used to indicate the influence degree of the decay characteristics on the detection result of the remaining service life of the power supply, and the second feature weight is used to indicate the influence degree of the power consumption characteristics on the detection result of the remaining service life of the power supply; Performing weighted calculation on the decay characteristics using the first feature weight, and performing weighted calculation on the power consumption characteristics using the second feature weight, and performing summation calculation on the weighted calculation results to obtain the merged feature.

8. The method according to claim 6, wherein: After inputting the merged feature into a target prediction model to obtain the remaining service life of the server power supply output by the target prediction model, the method further includes: Detecting the actual service life of the server power supply at the target time; Calculating the difference value between the actual service life and the remaining service life; Updating the model parameters of the target prediction model according to the difference value to obtain the updated target prediction model.

9. The method according to claim 8, wherein: The updating the model parameters of the target prediction model according to the difference value to obtain a reference prediction model includes: Calculating the difference value through a backpropagation algorithm to obtain parameter adjustment information for the model parameters of the target prediction model, where the parameter adjustment information is used to indicate the adjustment method for the model parameters; Adjust the model parameters of the target prediction model according to the parameter adjustment method indicated by the parameter adjustment information, and obtain the updated target prediction model.

10. The method according to claim 8, wherein The server system includes a plurality of the servers, and each of the servers is correspondingly configured with a target prediction model. The method further includes: Obtain the current target model parameters of each of the target prediction models, where the target model parameters are used to indicate the correlation relationship among the power consumption characteristics of the server, the decay characteristics of the server power supply, and the remaining service life of the server power supply; Convert the target model parameters of the plurality of target prediction models into candidate model parameters of the target prediction model; Send the candidate model parameters to each of the target prediction models, where the target prediction model is used to update the model parameters currently used by the target prediction model with the candidate model parameters.

11. The method according to claim 10, wherein: The converting the target model parameters of the plurality of target prediction models into candidate model parameters of the target prediction model includes: Use the model weights of each of the target prediction models to perform weighted summation on the target model parameters of the plurality of target prediction models to obtain a model parameter sum value, where the model weights are used to indicate the importance degree of the model parameters of the corresponding target prediction model; Calculate the quotient of the model parameter sum value and the number of models of the target prediction model to obtain an average model parameter value; Send the average model parameter value to each of the target prediction models, where the target prediction model is used to update the model parameters of the target prediction model with the average model parameter value.

12. The method according to claim 1, wherein: The generating the decay characteristics of the server power supply under the power consumption characteristics according to the multi-dimensional operation data includes: Use the following formula to calculate the temperature data included in the multi-dimensional operation data to obtain the decay characteristics: ； Among them, is the gas constant, is the temperature value in degrees Celsius, is the activation energy, is the frequency factor, is the thermal stress coefficient, and the decay characteristics include the thermal stress coefficient.

13. An electronic device, characterized in that, including: a memory for storing a computer program; a processor for implementing the steps of the detection method of the server power supply according to any one of claims 1 to 12 when executing the computer program.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, where the computer program, when executed by a processor, implements the steps of the detection method of the server power supply according to any one of claims 1 to 12.

15. A computer program product, comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the detection method of the server power supply according to any one of claims 1 to 12.

Images