Switching power supply health degree evaluation method based on data analysis

Abstract

The invention relates to the technical field of switching power supply monitoring, in particular to a switching power supply health degree evaluation method based on data analysis, and the method comprises the steps: collecting the operation data of a switching power supply, and constructing a multivariable time sequence; performing time domain and frequency domain feature analysis on the acquired data; a regression algorithm is adopted to screen the features, redundant information is removed, and feature subsets highly related to the health degree are extracted based on a relevancy threshold value; based on the screening characteristics, constructing a switching power supply health degree evaluation model, and realizing quantitative evaluation of the health state; the real-time health condition of the switching power supply is evaluated through the health degree model, and the residual service life is predicted in combination with an exponential attenuation rule, so that a scientific basis is provided for fault early warning and maintenance optimization of equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of switching power supply monitoring, in particular to a switching power supply health degree evaluation method based on data analysis. BACKGROUND

[0002] As a core component in electronic devices, switching power supplies are widely used in communication, power, industrial control, and consumer electronics fields. The stability and reliability of their operation have a significant impact on the overall performance of the system. However, due to the aging of power electronic components, environmental factors, and load changes during long-term operation, switching power supplies may gradually degrade in performance, even leading to failure, resulting in device downtime or system failure. Therefore, evaluating the health status of switching power supplies and predicting their remaining useful life have become an important research direction to improve device operation reliability and reduce maintenance costs.

[0003] Traditional switching power supply health management methods mainly rely on periodic maintenance and experience-based judgment. This approach is not only inefficient, but also may lead to serious consequences if faults are not detected in time. With the rapid development of data acquisition technology, signal processing technology, and machine learning algorithms, data-driven health status evaluation methods have gradually attracted widespread attention. By collecting and analyzing switching power supply operation data in real time, key features reflecting the health status of the device can be extracted, and advanced machine learning models can be used to quantify health assessment and remaining life prediction, providing a scientific basis for device maintenance.

[0004] Existing methods still have many shortcomings, such as the influence of redundant information in the data on model performance, the need for improvement in feature selection and health degree modeling accuracy, and the need for a data analysis method to improve the accuracy and efficiency of switching power supply health status evaluation.

[0005] In view of this, the present application provides a switching power supply health degree evaluation method based on data analysis. SUMMARY

[0006] The present application provides the following technical solutions:

[0007] A switching power supply health degree evaluation method based on data analysis includes:

[0008] Step 1: Collect switching power supply operation data, including voltage, current, ripple voltage, output power, switching frequency, and temperature data during switching power supply operation;

[0009] Step 2: Perform feature analysis on the collected switching power supply operation data, including time domain and frequency domain feature analysis of the switching power supply operation data;

[0010] Step 3: The regression algorithm is used to screen the analyzed features, and an adaptive correlation degree screening threshold is constructed, and the switching power supply health degree features are screened out through the correlation degree threshold;

[0011] Step 4: According to the screened switching power supply health degree features, a switching power supply health degree evaluation model is constructed based on a machine learning model, and the switching power supply health state is quantitatively evaluated;

[0012] Step 5: The real-time health condition of the switching power supply is evaluated through the switching power supply health degree model, and the remaining service life is calculated according to the life cycle of the switching power supply.

[0013] As a preferred scheme of the present application, the input voltage V in and the input current I in of the switching power supply are collected, the output voltage V out , the output current I out and the output ripple voltage V ripple of the switching power supply are collected, the output power P out of the switching power supply is calculated V out ×I out , the switching frequency f sw of the switching power supply in unit time is collected, the working environment temperature T amb of the switching power supply and the junction temperature T jun of the components are measured.

[0014] The collected various data are converted through A / D conversion and synchronized to be summarized and time-stamped, and a multivariate time series is obtained:

[0015] X(t)=[V in (t),I in (t),V out (t),I out (t),V ripple (t),P out (t),f sw (t),T amb ,T jun ];

[0016] Wherein, T is the sampling length.

[0017] As a preferred scheme of the present application, the collected switching power supply working data are analyzed, and the feature analysis includes time domain and frequency domain feature analysis of the switching power supply working data;

[0018] The time domain statistical features of the switching power supply working data are calculated, including: mean μ j , variance: Peak x j,max , valley x j,min , peak-valley difference Δxj , the root mean square value x j,rms , the skewness s j , and the kurtosis k j ; wherein j = 1, 2, …, 9, corresponding to the number of variables of the switching power supply operating data collected;

[0019] Performing fast Fourier transform on the switching power supply operating data to obtain the frequency spectrum FFT(x j ), j = 1, 2, …, 9; wherein FFT(x j ) represents performing fast Fourier transform on the jth feature;

[0020] Calculating the frequency domain statistical features of the switching power supply operating data, including: the direct current component X DC,j , the fundamental component amplitude |X FC,j |, the total harmonic distortion THD j , and the spectral entropy H s,k ;

[0021] Summarizing the time domain and frequency domain features of the analysis into a feature vector, denoted as f = [f1, f2, …, f M ], wherein M is the total number of features.

[0022] As a preferred scheme of the present application, the historical data of the switching power supply operation is obtained, the time domain and frequency domain analysis is performed on the historical data of the switching power supply operation, and the switching power supply operation feature vector dataset is obtained; the LASSO regression algorithm is used to filter the features, and the objective function of the LASSO regression algorithm is:

[0023]

[0024] Wherein w = [w1, w2, …, w M ] is the regression coefficient vector, w T is the transpose of the regression coefficient vector, f i = [f i1 , f i2 , …, f iM ] is the feature vector summarized from the time domain and frequency domain features of the ith sample, y i is the corresponding switching power supply health degree label, and λ is the regularization parameter, controlling the sparsity of feature selection;

[0025] The optimal regularization parameter λ * of the regularization parameter λ is selected through cross-validation, and the specific steps are as follows:

[0026] Randomly dividing the dataset into K mutually exclusive subsets D1, D2, …, DK K ;

[0027] For each λ value, K-fold cross-validation is performed: each subset D k is taken in turn as the validation set, and the remaining K-1 subsets as the training set;

[0028] The LASSO regression model is trained on the training set, and the performance indicator is calculated on the validation set;

[0029] The performance indicators of K validations are averaged as the generalization performance of the LASSO regression algorithm corresponding to the λ value, and the λ value with the optimal generalization performance is selected as λ * .

[0030] The optimal regularization parameter λ is substituted into the LASSO regression algorithm, and the final sparse regression coefficient w is trained; the feature index with a non-zero regression coefficient is denoted as S Then S is the index of the screened relevant feature subset.

[0031] As a preferred scheme of the present application, an adaptive relevance screening threshold is constructed based on the screened relevant features; for each feature j ∈ S, the Pearson correlation coefficient of the feature and the switching power supply health degree label y = [y1, y2,..., y N ] is calculated:

[0032]

[0033] Where μ j and are the mean values of the feature j and the switching power supply health degree label, respectively; f ij represents the value of the jth feature of the ith sample, N represents the total number of samples, and ρ j represents the Pearson correlation coefficient of the feature j and the switching power supply health degree label y;

[0034] According to the distribution of the Pearson correlation coefficient, an adaptive relevance screening threshold θ is determined in the form of mean plus or minus standard deviation: Where and σ ρ are the mean and standard deviation of the Pearson correlation coefficient, respectively, and β is a regulation factor;

[0035] The screened switching power supply health degree feature index is S * = {j ∈ S | |ρ j | ≥ θ}; the corresponding switching power supply health degree feature subset is denoted as as the input of the switching power supply health degree evaluation model.

[0036] As a preferred scheme of the present application, a switching power supply health degree evaluation model based on a support vector regression SVM model is constructed, and the switching power supply health degree feature subset and the corresponding switch power health degree label y = [y1, y2,..., y N ] as the input of the SVM model, a SVM regression model is trained;

[0037] The target of the SVM model is to find a hyperplane g(f) = δ T + b, which minimizes the distance of all sample points to the hyperplane and the norm of the hyperplane |δ|; the optimization problem of the SVM model is solved to obtain the parameters δ and b of the SVM model;

[0038] The prediction function of the SVM model is: wherein represents the evaluated switch power health degree value, and φ(f) represents the kernel function of mapping the feature f to a high-dimensional space.

[0039] As a preferred scheme of the present application, the switch power health degree value of the switch power is calculated in real time by using the constructed switch power health degree evaluation model; the switch power health degree value at the current time t is obtained by using the switch power health degree evaluation model

[0040] The exponential decay law of the switch power health degree degradation process of the switch power is constructed, i.e.

[0041] H(t) = exp(-ηt);

[0042] wherein η is the attenuation coefficient of the switch power health degree, reflecting the aging rate of the switch power in the actual use environment; the least square method is used to fit the exponential decay curve by using the historical switch power health degree values to obtain the estimated value of the attenuation coefficient

[0043]

[0044] According to the switch power health degree value at the current time t and the estimated value of the switch power health degree attenuation coefficient the remaining service life R(t) of the switch power is calculated:

[0045] As a preferred scheme of the present application, the sliding average of the estimated values of the remaining service life of the switch power at continuous multiple time points is performed, a time window with a fixed length Q is set to store the estimated values of the remaining service life of the switch power at the last Q time points:

[0046]

[0047] The average value of all the estimated values of the remaining service life in the time window is calculated as the calculation result of the remaining service life of the switch power:

[0048]

[0049] wherein represents a smooth estimated value of the remaining service life of the switching power supply predicted at time t.

[0050] As a preferred scheme of the present application, considering that the actual health state of the switching power supply will change, resulting in deviation of the remaining service life prediction, an anomaly detection mechanism is introduced, and at each time of updating the estimated value of the remaining service life of the switching power supply, the change amount between the estimated value of the remaining service life of the switching power supply and the estimated value at the previous time is calculated:

[0051]

[0052] If the change amount exceeds a preset threshold γ, that is, an abnormal situation is considered to occur, and the remaining service life value of the switching power supply at the current time is updated according to the prediction result of the service life of the switching power supply at the previous time; when the abnormal situation disappears, the remaining service life value of the switching power supply at the current time is updated according to the prediction result of the service life of the switching power supply at the current time.

[0053] A switching power supply health degree evaluation system based on data analysis, comprising: a data acquisition module, a feature analysis module, a feature selection module, a health degree evaluation module and a remaining life calculation module;

[0054] The data acquisition module is used to acquire switching power supply working data, including voltage, current, ripple voltage, output power, switching frequency and temperature data when the switching power supply is working.

[0055] The feature analysis module is used to perform feature analysis on the acquired switching power supply working data, and the feature analysis includes time domain and frequency domain feature analysis on the switching power supply working data.

[0056] The feature selection module uses a regression algorithm to screen the analyzed features, constructs an adaptive correlation degree screening threshold, and screens out the switching power supply health degree features through the correlation degree threshold.

[0057] The health degree evaluation module constructs a switching power supply health degree evaluation model based on a machine learning model according to the screened switching power supply health degree features, and quantitatively evaluates the health state of the switching power supply.

[0058] The remaining life calculation module evaluates the real-time health condition of the switching power supply through the switching power supply health degree model, and calculates the remaining service life according to the life cycle of the switching power supply.

[0059] The application has the beneficial effects that the application comprehensively collects key parameters of the operation of the switching power supply, covers voltage, current, ripple voltage, power, frequency and temperature, provides a data basis for subsequent feature analysis, and ensures that the input information of the health degree evaluation model is complete and effective.

[0060] The application extracts time domain and frequency domain features, mines potential health state information in the data, and provides higher dimension and deeper feature expression for subsequent feature screening and health degree evaluation.

[0061] The application screens key health degree features through a regression algorithm and a correlation threshold, removes redundant or irrelevant features, improves the efficiency and accuracy of the health degree evaluation model, and enhances the interpretability of the model.

[0062] The application trains a machine learning model based on the screened health degree features, realizes quantitative evaluation of the health state, and provides a scientific basis and prediction ability for switching power supply health management.

[0063] The application monitors the health status of the switching power supply in real time, predicts the remaining life in combination with the degradation law, helps to find potential faults in advance, optimizes the maintenance plan, prolongs the equipment use cycle and reduces the operating cost. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0065] Figure 1 The application is a switching power supply health degree evaluation method flow chart based on data analysis;

[0066] Figure 2 The application is a switching power supply health degree evaluation system structure diagram based on data analysis. DETAILED DESCRIPTION

[0067] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the application.

[0068] Embodiment 1

[0069] Reference Figure 1The first embodiment of the present application provides a switch power supply health degree evaluation method based on data analysis.

[0070] Step 1: Collect switch power supply working data, including voltage, current, ripple voltage, output power, switching frequency and temperature data when the switch power supply is working.

[0071] Collect the input voltage V in and input current I in of the switch power supply; the sampling frequency f s should meet the Nyquist sampling theorem, i.e. f s ≥ 2f max , where f max is the highest frequency component of the input signal; the sampling bit number n should not be less than 12 bits to ensure sampling accuracy; the sampling time T should cover several working periods of the switch power supply to obtain comprehensive working state data.

[0072] Collect the output voltage V out , output current I out and output ripple voltage V r of the switch power supply; the output ripple voltage reflects the stability of the power supply output voltage, which can be measured using an oscilloscope; calculate the output power P out of the switch power supply = V out × I out .

[0073] Collect the switching frequency f sw of the switch power supply per unit time using a frequency meter or an oscilloscope; the switching frequency is an important indicator for evaluating the performance of the switch power supply, and its change indicates the aging or failure of components.

[0074] Measure the working environment temperature T amb and the junction temperature T jun of key components such as power electronic switching tubes of the switch power supply using a temperature sensor; the temperature data should be collected as close to the heat generating components as possible and avoid the influence of external thermal disturbance; a thermocouple, a thermistor or an infrared thermometer can be used.

[0075] Convert all types of collected data through A / D conversion and synchronize to get a multivariate time series with time stamping:

[0076] X(t) = [V in (t), I in (t), V out (t), I out (t), V ripple (t), P out (t), f sw (t), T amb , T jun ].

[0077] wherein T is a sampling duration.

[0078] This step comprehensively reflects the running state of the switching power supply by collecting key data (voltage, current, frequency, temperature, etc.) of the switching power supply, provides accurate and rich data basis for subsequent feature extraction and health degree analysis, and ensures the accuracy and reliability of the evaluation results.

[0079] Step 2: Perform feature analysis on the collected switching power supply working data, which includes time domain and frequency domain feature analysis of the switching power supply working data.

[0080] The time domain statistical features of the switching power supply working data are calculated, including: mean μ j , variance: peak x j,max , valley x j,min , peak-valley difference Δx j , root mean square value x j,rms , skewness s j and kurtosis k j ; wherein j = 1, 2,..., 9, corresponding to the number of collected switching power supply working data variables.

[0081] The fast Fourier transform is performed on the switching power supply working data to obtain the frequency spectrum FFT(x j ), j = 1, 2,..., 9; FFT(x j ) represents the fast Fourier transform of the jth feature.

[0082] The frequency domain statistical features of the switching power supply working data are calculated, including: DC component X DC,j , fundamental component amplitude |X FC,j |, total harmonic distortion THD j and spectral entropy H s,k .

[0083] The analyzed time domain and frequency domain features are summarized into a feature vector, denoted as f = [f1, f2,..., f M ], wherein M is the total number of features.

[0084] This step extracts key features such as mean, variance, frequency spectrum, etc. through time domain and frequency domain feature analysis, deeply mines the statistical rules and frequency domain information of the data, and constructs a multi-dimensional feature space required for health degree evaluation, providing a scientific basis for model training.

[0085] Step 3: Use a regression algorithm to filter the analyzed features, construct an adaptive correlation degree filtering threshold, and filter out the switching power supply health degree features through the correlation degree threshold.

[0086] The historical data of the switching power supply is acquired, and time domain and frequency domain analysis is performed on the historical data of the switching power supply to obtain a switching power supply working feature vector data set.

[0087] The LASSO regression algorithm is used for feature screening, the LASSO regression algorithm introduces an L1 norm regularization term on the basis of a conventional least square linear regression, and can realize automatic selection and sparsification of features, and a target function of the LASSO regression algorithm is:

[0088]

[0089] Where w = [w1, w2,..., w M ] is a regression coefficient vector, w T is a transpose of the regression coefficient vector, f i = [f i1 , f i2 ,..., f iM ] is a feature vector of time domain and frequency domain features of an i-th sample, y i is a corresponding switching power supply health degree label, the switching power supply health degree label is marked according to switching power supply operation characteristics by combining artificial experience and prior knowledge of a field expert, and λ is a regularization parameter for controlling sparseness of feature selection.

[0090] An optimal regularization parameter λ * of the regularization parameter λ is selected through cross validation, and specific steps are as follows:

[0091] A data set is randomly divided into K mutually exclusive subsets D1, D2,..., DK K .

[0092] For each λ value, K-fold cross validation is performed: each subset D k is sequentially taken as a validation set, and the remaining K-1 subsets are taken as training sets;

[0093] A LASSO regression model is trained on the training sets, and a performance index (such as a mean square error, a classification accuracy, etc.) is calculated on the validation sets;

[0094] The performance indexes of K times of validation are averaged as a generalization performance of the LASSO regression algorithm corresponding to the λ value, and the λ value with the optimal generalization performance is selected as λ * .

[0095] The optimal regularization parameter λ is substituted into the LASSO regression algorithm, and a final sparse regression coefficient w is trained; a feature index with a non-zero regression coefficient is denoted as S , and S is an index of a screened relevant feature subset.

[0096] Based on the screened relevant features, an adaptive correlation screening threshold is constructed; for each feature j∈S, the Pearson correlation coefficient of the feature and the switching power supply health degree label y = [y1, y2,..., y N ] is calculated:

[0097]

[0098] Where μ j and are the mean values of the feature j and the switching power supply health degree label respectively; f ij represents the value of the jth feature of the ith sample, N represents the total number of samples, and ρ j represents the Pearson correlation coefficient of the feature j and the switching power supply health degree label y.

[0099] According to the distribution of the Pearson correlation coefficient, the mean plus or minus the standard deviation is used to adaptively determine the correlation screening threshold θ: Where and σ ρ are the mean and standard deviation of the Pearson correlation coefficient respectively, and β is the adjustment factor, usually taking a value between 1 and 3.

[0100] The screened switching power supply health degree feature index is: S * ={j∈S||ρ j |≥θ}; the corresponding switching power supply health degree feature subset is as the input of the switching power supply health degree evaluation model.

[0101] This step uses LASSO regression to screen key features, remove redundant data, and improve model running efficiency and prediction accuracy. Through the adaptive correlation screening threshold, it ensures that the selected features are highly correlated with the health degree, enhancing the scientificity and interpretability of the model.

[0102] Step 4: Based on the screened switching power supply health degree features, a switching power supply health degree evaluation model is constructed based on a machine learning model to quantitatively evaluate the health status of the switching power supply.

[0103] A switching power supply health degree evaluation model based on a support vector regression SVM model is constructed, with the switching power supply health degree feature subset and the corresponding switching power supply health degree label y = [y1, y2,..., y N ] as the input of the SVM model, and the SVM regression model is trained.

[0104] The goal of the SVM model is to find a hyperplane g(f) = δ Tφ(f) + b, such that the distance of all sample points to the hyperplane is minimized, while the norm of the hyperplane |δ| is also minimized; the optimization objective of the SVM model is represented as:

[0105]

[0106] where φ(·) represents a kernel function that maps features to a high-dimensional space, ξ i and are slack variables, ε is a tolerance error, and C is a balance factor that controls the complexity and fitting degree of the SVM model;

[0107] Solving the optimization problem of the SVM model, the parameters δ and b of the SVM model are obtained;

[0108] The prediction function of the SVM model is: where represents the health degree value of the evaluated switching power supply; the closer to 1 indicates that the device is healthier, and the closer to 0 indicates that the device is more likely to fail.

[0109] This step is based on the selected health degree features to train the SVM regression model. The model can quantify the health status of the switching power supply and accurately distinguish between normal and abnormal conditions of the device, providing reliable support for real-time monitoring and fault prediction.

[0110] Step 5: Evaluate the real-time health status of the switching power supply through the switching power supply health degree model, and calculate the remaining service life according to the life cycle of the switching power supply.

[0111] Using the constructed switching power supply health degree evaluation model, the switching power supply health degree value of the switching power supply is calculated in real time; through the switching power supply health degree evaluation model, the switching power supply health degree value at the current time t is obtained

[0112] An exponential decay law is constructed for the switching power supply health degree degradation process of the switching power supply, i.e.,

[0113] H(t) = exp(-ηt);

[0114] where η is the switching power supply health degree attenuation coefficient, reflecting the aging rate of the switching power supply in the actual use environment.

[0115] Using the least squares method, the historical switching power supply health degree values are fitted to obtain the estimated value of the attenuation coefficient

[0116]

[0117] According to the switching power supply health degree value of the current time t And switching power supply health degree attenuation coefficient estimated value The remaining service life R(t) of the switching power supply is calculated: The remaining service life R(t) represents the time between the current time and the time when the health degree reaches the critical value.

[0118] The sliding average of the switching power supply remaining service life estimates for a plurality of consecutive time points is set to a fixed length of Q time window for storing the switching power supply remaining service life estimates for the last Q time points:

[0119]

[0120] The average of all remaining service life estimates in the time window is calculated as the calculation result of the remaining service life of the switching power supply:

[0121]

[0122] Wherein The smooth estimate of the predicted remaining service life of the switching power supply at time t is represented.

[0123] Considering that the actual health state of the switching power supply will change, resulting in deviation of the remaining service life prediction, an anomaly detection mechanism is introduced, which calculates the change between the switching power supply remaining service life estimate and the previous time estimate at each update of the switching power supply remaining service life estimate:

[0124]

[0125] If the change exceeds the preset threshold γ, that is An abnormal situation is considered to have occurred, and the switching power supply remaining service life value at the current time is updated according to the prediction result of the switching power supply service life at the previous time; when the abnormal situation disappears, the switching power supply remaining service life value at the current time is updated according to the prediction result of the switching power supply service life at the current time.

[0126] This step uses the health degree model to monitor the device state in real time, and combines the exponential decay law to predict the remaining life. The life estimate is smoothed by sliding average to reduce fluctuations and errors, and to provide a scientific basis for preventive maintenance and service life optimization.

[0127] Embodiment 2

[0128] Referring to Figure 2 , the second embodiment of the present application provides a switching power supply health degree evaluation system based on data analysis.

[0129] The system comprises a data acquisition module, a feature analysis module, a feature selection module, a health degree evaluation module and a remaining life calculation module.

[0130] The data acquisition module is configured to acquire switch power supply working data, including voltage, current, ripple voltage, output power, switching frequency and temperature data when the switch power supply is working.

[0131] The feature analysis module is configured to perform feature analysis on the acquired switch power supply working data, including time domain and frequency domain feature analysis on the switch power supply working data.

[0132] The feature selection module uses a regression algorithm to filter the analyzed features, constructs an adaptive correlation degree filtering threshold, and filters out the switch power supply health degree features through the correlation degree threshold.

[0133] The health degree evaluation module constructs a switch power supply health degree evaluation model based on a machine learning model according to the filtered switch power supply health degree features, and quantitatively evaluates the health state of the switch power supply.

[0134] The residual life estimation module estimates the residual service life of the switch power supply according to the life cycle of the switch power supply by evaluating the real-time health condition of the switch power supply through the switch power supply health degree model.

[0135] Embodiment 3

[0136] The third embodiment of the present application is different from the previous embodiment in that:

[0137] If the function is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0138] The logic and / or steps represented in the flow diagrams, or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be for example but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer. In this context, a "computer-readable medium" can be any means that can store the program for use by or in connection with the instruction execution system, apparatus, or device.

[0139] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer.

[0140] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, alone or in any combination, can be used with the necessary logic gates to implement logic functions on data signals: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and so forth.

[0141] Moreover, for the sake of brevity, the description of the exemplary embodiments has been simplified to some extent, in that not all features of actual implementations have been described (i.e., those related to the best mode for carrying out the application have been omitted or not described in detail).

[0142] It should be understood that numerous specific implementation decisions can be made during the development of any actual implementation method, and in any engineering or design project. Such development efforts may be complex and time-consuming, but for those of ordinary skill in the art who benefit from this disclosure, the development effort will be a routine work of design, manufacturing, and production without requiring much experimentation.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data analysis-based method for assessing the health of a switching power supply, characterized in that, include: Step 1: Collect the operating data of the switching power supply, including voltage, current, ripple voltage, output power, switching frequency, and temperature data when the switching power supply is operating. Step 2: Perform feature analysis on the collected switching power supply operating data. The feature analysis includes time-domain and frequency-domain feature analysis of the switching power supply operating data. Step 3: Use a regression algorithm to filter the analyzed features, construct an adaptive relevance filtering threshold, and filter the switching power supply health features through the relevance threshold; Step 4: Based on the selected health characteristics of the switching power supply, construct a health assessment model for the switching power supply using a machine learning model to quantitatively assess the health status of the switching power supply. Step 5: Assess the real-time health status of the switching power supply using the switching power supply health model, and calculate the remaining service life based on the life cycle of the switching power supply.

2. The method for assessing the health of a switching power supply based on data analysis according to claim 1, characterized in that, Collect the input voltage V of the switching power supply in and input current I in Collect the output voltage V of the switching power supply out Output current I out and output ripple voltage V ripple Calculate the output power P of the switching power supply out =V out ×I out Collect the switching frequency f of the switching power supply per unit time. sw ;Measure the operating ambient temperature T of the switching power supply amb and the junction temperature T of the components jun ; The collected data are converted using an analog-to-digital converter and simultaneously summarized and timestamped to obtain a multivariate time series: X(t)=[V in (t),I in (t),V out (t),I out (t),V ripple (t),P out (t),f sw (t),T amb ,T jun ]; Where T is the sampling duration.

3. The method for assessing the health of a switching power supply based on data analysis according to claim 2, characterized in that, The collected switching power supply operating data is subjected to feature analysis, which includes time-domain and frequency-domain feature analysis of the switching power supply operating data; The time-domain statistical characteristics of the switching power supply operating data were calculated, including: mean μ. j ,variance: Peak x j,max Valley value x j,min Peak-to-valley difference Δx j Root mean square value x j,rms skewness s j and kurtosis k j Where j = 1, 2, ..., 9, corresponding to the number of variables in the switching power supply operating data collected; Perform a Fast Fourier Transform (FFT) on the operating data of the switching power supply to obtain the spectrum FFT(x) j ),j=1,2,...,9; where, FFT(x j ) indicates that the j-th feature is subjected to a Fast Fourier Transform; The frequency domain statistical characteristics of the switching power supply operating data were calculated, including: DC component X. DC,j Fundamental component amplitude |X FC,j Total Harmonic Distortion (THD) j and spectral entropy H s,j ; The analyzed time-domain and frequency-domain features are summarized into an eigenvector, denoted as f = [f1, f2, ..., f...]. M ], where M is the total number of features.

4. The method for assessing the health of a switching power supply based on data analysis according to claim 3, characterized in that, Historical operating data of the switching power supply was acquired, and time-domain and frequency-domain analyses were performed on this data to obtain a feature vector dataset. The LASSO regression algorithm was then used to filter the features. The objective function of the LASSO regression algorithm is: Where w = [w1, w2, ..., w M ] represents the regression coefficient vector, w T f is the transpose of the regression coefficient vector. i =[f i1 ,f i2 ,...,f iM ] is the feature vector formed by summing the time-domain and frequency-domain features of the i-th sample, y i For the corresponding switching power supply health label, λ is the regularization parameter that controls the sparsity of feature selection; The optimal regularization parameter λ is selected through cross-validation. * The specific steps are as follows: The dataset is randomly divided into K mutually exclusive subsets D1, D2, ..., D... K ; For each λ value, perform K-fold cross-validation: sequentially cross-validate each subset D k The remaining K-1 subsets serve as the training set, serving as the validation set. Train the LASSO regression model on the training set and compute performance metrics on the validation set; The average performance metrics from K validations are taken as the generalization performance of the LASSO regression algorithm for the corresponding λ value. The λ value with the best generalization performance is selected as λ. * . Substituting the optimal regularization parameter λ into the LASSO regression algorithm, the final sparse regression coefficients w are obtained through training. Let the feature index where the regression coefficient is not zero be denoted as Then S is the index of the selected relevant feature subset.

5. The method for assessing the health of a switching power supply based on data analysis according to claim 4, characterized in that, Based on the selected relevant features, an adaptive relevance screening threshold is constructed; for each feature j∈S, the correlation between the feature and the switching power supply health label y=[y1,y2,...,y N The Pearson correlation coefficient of ] Where μ j and Let f be the mean of feature j and the health label of the switching power supply, respectively. ij Let ρ represent the value of the j-th feature of the i-th sample, N represent the total number of samples, and ρ represent the value of the j-th feature of the i-th sample. j This represents the Pearson correlation coefficient between feature j and the switching power supply health label y; Based on the distribution of Pearson correlation coefficients, the correlation screening threshold θ is adaptively determined using the mean plus or minus the standard deviation: in and σ ρ , respectively, are the mean and standard deviation of the Pearson correlation coefficient, and β is the adjustment factor; The selected switching power supply health feature index is: S * ={j∈S||ρ j |≥θ};Let the corresponding subset of switching power supply health features be denoted as As input to the switching power supply health assessment model.

6. The method for assessing the health of a switching power supply based on data analysis according to claim 5, characterized in that, Construct a health assessment model for switching power supplies based on a support vector regression (SVM) model, and subset the health features of the switching power supplies. and the corresponding switching power supply health label y = [y1, y2, ..., y N As input to the SVM model, train the SVM regression model; The goal of the SVM model is to find a hyperplane g(f) = δ T φ(f)+b minimizes the distance from all sample points to the hyperplane, and also minimizes the norm |δ| of the hyperplane; solve the optimization problem of the SVM model to obtain the parameters δ and b of the SVM model; The prediction function of the SVM model is: in φ(f) represents the evaluated health value of the switching power supply, and φ(f) represents the kernel function that maps feature f to a high-dimensional space.

7. The method for assessing the health of a switching power supply based on data analysis according to claim 6, characterized in that, Using the constructed switching power supply health assessment model, the switching power supply health value is calculated in real time; the current switching power supply health value is obtained through the switching power supply health assessment model. The exponential decay law of the health degradation process of switching power supplies is constructed, namely: H(t) = exp(-ηt); Where η is the health degradation coefficient of the switching power supply, reflecting the aging rate of the switching power supply under actual use conditions; using the least squares method, through historical switching power supply health values... By fitting the exponential decay curve, an estimated value of the decay coefficient can be obtained. Based on the current time t, the health value of the switching power supply Estimated value of the health degradation coefficient of switching power supply Calculate the remaining service life R(t) of the switching power supply:

8. The method for assessing the health of a switching power supply based on data analysis according to claim 7, characterized in that, A moving average is performed on the estimated remaining lifespan of the switching power supply over multiple consecutive time points. A fixed time window of length Q is set to store the estimated remaining lifespan of the switching power supply for the most recent Q time points. The average of all remaining lifetime estimates within the calculation time window is used as the result of the remaining lifetime calculation for the switching power supply. in This represents a smoothed estimate of the remaining lifespan of the switching power supply predicted at time t.

9. The method for assessing the health of a switching power supply based on data analysis according to claim 8, characterized in that, Considering that the actual health status of the switching power supply may change, leading to deviations in the remaining service life prediction, an anomaly detection mechanism is introduced. Each time the estimated remaining service life of the switching power supply is updated, the change between the estimated remaining service life and the previous estimate is calculated. If the change exceeds the preset threshold γ, that is If an abnormal situation is detected, the remaining lifespan of the switching power supply at the current moment is updated according to the predicted lifespan of the switching power supply at the previous moment; when the abnormal situation disappears, the remaining lifespan of the switching power supply at the current moment is updated according to the predicted lifespan of the switching power supply at the current moment.

10. A system employing a data analysis-based health assessment method for switching power supplies as described in any one of claims 1 to 9, characterized in that, include: The module includes a data acquisition module, a feature analysis module, a feature selection module, a health assessment module, and a remaining life expectancy calculation module. The data acquisition module is used to collect the operating data of the switching power supply, including voltage, current, ripple voltage, output power, switching frequency and temperature data when the switching power supply is working. The feature analysis module is used to perform feature analysis on the collected switching power supply operating data. The feature analysis includes time-domain and frequency-domain feature analysis on the switching power supply operating data. The feature selection module uses a regression algorithm to filter the analyzed features, constructs an adaptive relevance filtering threshold, and filters out the health features of the switching power supply through the relevance threshold. The health assessment module constructs a health assessment model for the switching power supply based on a machine learning model, according to the selected health characteristics of the switching power supply, and quantitatively assesses the health status of the switching power supply. The remaining lifespan calculation module assesses the real-time health status of the switching power supply using a switching power supply health model and calculates the remaining lifespan based on the switching power supply's life cycle.

Images