Health feature extraction method and device for front-end isolated switching power supply of aging table

Abstract

The embodiment of the invention provides a health feature extraction method and device for a front-end isolated switching power supply of an aging table, and the method comprises the steps: obtaining a plurality of historical signal values of an analog signal of the front-end isolated switching power supply of the aging table; a plurality of historical signal values are encoded through a Transform model encoder in a multi-layer model encoder, corresponding self-attention features are generated, the multi-layer model encoder is of a neural network structure which is constructed through the Transform model encoder and a stack type automatic encoder and is fused with context features, the multi-layer model encoder comprises an input layer, a middle layer and an output layer, and the input layer and the middle layer are connected with the stack type automatic encoder. The middle layer comprises a Transform model encoder structure and a stack type automatic encoder which are connected in sequence; extracting frequency domain features of the plurality of historical signal values, and fusing the self-attention features with the frequency domain features; and encoding the fused features through a stack type automatic encoder, and generating health features corresponding to the isolated switching power supply at the front end of the aging table.

Description

Technical Field

[0001] The embodiments of this specification relate to the technical field of lossless reliability screening of integrated circuits, and more particularly to a method for extracting health characteristics of an isolated switching power supply at the front end of a burn-in station. Background Art

[0002] The integrated circuit high temperature aging test bench accelerates various physical and chemical reaction processes inside the components by continuously applying a certain amount of electrical stress to the components over a long period of time, prompting various potential faults inside the components to be exposed early, thereby eliminating early failure products and making the electronic components enter a period of low failure rate and relative stability from the beginning of use. The current integrated circuit high temperature aging test bench can achieve long-term monitoring of the test environment by adding long-term monitoring in the machine and overstress protection mechanism, and can respond in time when the machine fails to achieve post-maintenance based on fault data. However, due to the lack of active protection technology for integrated circuit high temperature aging test benches, it is difficult to achieve the integrity of the test process and the consistency of the test environment stress, and it is very easy to cause the test process to be interrupted due to machine failure, resulting in major property losses such as damage to millions of test devices, or the aging test is identified as a failure test due to adverse effects such as the introduction of additional stress during the test due to machine performance degradation, resulting in ineffective waste of resources. At the same time, ensuring the quality of high-temperature aging tests can avoid over-aging and under-aging failures of integrated circuits, significantly reduce the failure rate of integrated circuits, and help avoid large-scale electronic system failures and shutdowns due to integrated circuit failures in new energy vehicles, civil airliners, energy storage and transmission stations, etc. that use the same integrated circuits.

[0003] The front-end isolated switching power supply provides the required power for the entire aging test bench to ensure that all parts of the test bench can work normally. As the core key equipment of the aging test bench, the front-end isolated switching power supply has a great impact on the overall reliability of the aging test bench. Once circuit degradation, failure or sudden failure occurs, the aging test bench test will stop and damage the object under test. In severe cases, the voltage will be overloaded, causing a fire, resulting in major property losses and safety hazards. Therefore, it is very necessary to perform fault prediction and health assessment on the front-end isolated switching power supply of the aging test bench. The fault prediction and health assessment of the front-end isolated switching power supply of the aging test bench must be based on the health characteristics of the front-end isolated switching power supply of the aging test bench. Therefore, how to extract the health characteristics of the front-end isolated switching power supply of the aging test bench has become an urgent problem to be solved. Summary of the invention

[0004] In view of this, an embodiment of this specification provides a method for extracting the health characteristics of an isolated switching power supply at the front end of a burn-in station. One or more embodiments of this specification also relate to a device for extracting the health characteristics of an isolated switching power supply at the front end of a burn-in station, a computing device, a computer-readable storage medium, and a computer program to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a method for extracting health features of a front-end isolated switching power supply of a burn-in station is provided, comprising: Acquire multiple historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station; The plurality of historical signal values ​​are encoded by a Transformer model encoder in a multi-layer model encoder to obtain self-attention features in the plurality of historical signal values, wherein the multi-layer model encoder is a neural network structure for integrating context features constructed by the Transformer model encoder and a stacked autoencoder, including an input layer, an intermediate layer, and an output layer, wherein the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder; Extracting frequency domain features of the multiple historical signal values, and fusing the self-attention features with the frequency domain features to generate target fusion features; The target fusion features are encoded by the stacked autoencoder in the multi-layer model encoder to generate health features corresponding to the isolated switching power supply at the front end of the aging station.

[0006] Optionally, the multi-layer model encoder further includes a multi-head attention layer, a frequency domain feature extraction layer and a fusion layer, wherein the frequency domain feature extraction layer is used to extract frequency domain features, and the fusion layer is used to fuse the data features extracted by the Transformer model encoder and the frequency domain feature extraction layer; Accordingly, the method further comprises: Obtain a transformation result generated by the multi-head attention layer performing a linear transformation on the multiple historical signal values; Encoding the transformation result through the Transformer model encoder to generate self-attention features in the multiple historical signal values; Performing frequency domain feature extraction on the transformation result through the frequency domain feature extraction layer to obtain frequency domain features in the multiple historical signal values ​​output by the frequency domain feature extraction layer; The self-attention feature and the frequency domain feature are fused through the fusion layer to obtain the target fusion feature output by the fusion layer.

[0007] Optionally, after obtaining a plurality of historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station, the method further includes: Performing sliding window cutting on the plurality of historical signal values ​​according to a first window width and a first step length to generate a first sliding window cutting result, and determining an initial sample data set based on the first sliding window cutting result; Normalizing the initial sample data set to generate a target sample data set; Accordingly, encoding the multiple historical signal values ​​by the Transformer model encoder in the multi-layer model encoder includes: The target sample data set is encoded by a Transformer model encoder in a multi-layer model encoder.

[0008] Optionally, extracting frequency domain features of the multiple historical signal values ​​includes: The target sample data set is subjected to sliding window cutting according to a second window width and a second step length to generate a second sliding window cutting result, and frequency domain features of the plurality of historical signal values ​​are extracted based on the second sliding window cutting result.

[0009] Optionally, extracting frequency domain features of the plurality of historical signal values ​​based on the second sliding window cutting result includes: Performing Fourier transform on each to-be-processed sample data set contained in the second sliding window cutting result to generate a time series energy spectrum corresponding to each to-be-processed sample data set; Respectively extracting eigenvalues ​​corresponding to at least two initial frequency domain features of the time series energy spectrum; The characteristic values ​​are normalized using a maximum and minimum normalization algorithm to generate frequency domain features of the multiple historical signal values.

[0010] Optionally, the at least two initial frequency domain features include: peak value, bandwidth, center frequency, shape factor, and power spectral density.

[0011] Optionally, fusing the self-attention feature and the frequency domain feature through the fusion layer to obtain a target fusion feature output by the fusion layer includes: The self-attention feature and the frequency domain feature are concatenated through the fusion layer to obtain the target fusion feature output by the fusion layer.

[0012] According to a second aspect of the embodiments of this specification, a health feature extraction device for a front-end isolated switching power supply of a burn-in station is provided, comprising: An acquisition module is configured to acquire multiple historical signal values ​​of an analog signal of an isolated switching power supply at the front end of the aging station; A first encoding module is configured to encode the multiple historical signal values ​​through a Transformer model encoder in a multi-layer model encoder to obtain self-attention features in the multiple historical signal values, wherein the multi-layer model encoder is a neural network structure that integrates context features and is constructed by the Transformer model encoder and a stacked autoencoder, including an input layer, an intermediate layer, and an output layer, wherein the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder; A feature fusion module is configured to extract frequency domain features of the multiple historical signal values, and fuse the self-attention feature with the frequency domain feature to generate a target fusion feature; The second encoding module is configured to encode the target fusion features through the stacked autoencoder in the multi-layer model encoder to generate health features corresponding to the isolated switching power supply at the front end of the aging station.

[0013] According to a third aspect of an embodiment of this specification, a computing device is provided, including: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement any step of the method for extracting health characteristics of the isolated switching power supply at the front end of the aging station.

[0014] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of any one of the methods for extracting health characteristics of the isolated switching power supply at the front end of the aging station are implemented.

[0015] According to a fifth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned method for extracting health characteristics of the isolated switching power supply at the front end of the aging station.

[0016] The embodiment of this specification obtains multiple historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station, encodes the multiple historical signal values ​​through the Transformer model encoder in the multi-layer model encoder, and obtains the self-attention features in the multiple historical signal values, wherein the multi-layer model encoder is a neural network structure that integrates context features constructed by the Transformer model encoder and the stacked autoencoder, including an input layer, an intermediate layer and an output layer, and the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder; extracts the frequency domain features of the multiple historical signal values, and fuses the self-attention features with the frequency domain features to generate target fusion features; encodes the target fusion features through the stacked autoencoder in the multi-layer model encoder to generate the health features corresponding to the isolated switching power supply at the front end of the aging station. The embodiment of this specification encodes the historical signal values ​​of the analog signal once through the Transformer model encoder to obtain the self-attention features, then extracts the frequency domain features of the historical signal values ​​of the analog signal, and then encodes the splicing results of the self-attention features and the frequency domain features through the stacked autoencoder to achieve deep feature fusion. Through this processing method, the timing dependency and change trend between multiple historical signal values ​​can be directly mapped into the hidden layer deep features, which is helpful to ensure the accuracy of the health feature output results of the isolated switching power supply at the front end of the aging station. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a method for extracting health characteristics of a front-end isolated switching power supply of a burn-in station provided by an embodiment of this specification; Figure 2 is a flow chart of a feature extraction process provided by an embodiment of this specification; Figure 3 is a schematic diagram of the structure of a Transformer model encoder provided by an embodiment of this specification; Figure 4 is a flowchart of another feature extraction process provided by an embodiment of this specification; Figure 5 is a schematic diagram of the structure of a stacked autoencoder provided by an embodiment of the present specification; Figure 6 It is a structural schematic diagram of a health feature extraction device for a front-end isolated switching power supply of a aging station provided by an embodiment of this specification; Figure 7 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0018] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0019] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0020] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0021] First, the terms involved in one or more embodiments of this specification are explained.

[0022] Stacked Auto-Encoders (SAE): A deep learning model that originates from the concept of traditional autoencoders and forms a deep neural network structure by stacking multiple simple autoencoders layer by layer.

[0023] Current high-temperature aging products can achieve long-term monitoring of the test environment by adding long-term in-machine monitoring and overstress protection mechanisms, and can respond promptly when a product fails, achieving post-maintenance based on fault data. However, due to the lack of active protection technology for high-temperature aging products, existing aging products are difficult to achieve the integrity of the test process and the consistency of the test environment stress. It is very easy to cause the test process to be interrupted due to product failure, resulting in major property losses such as damage to millions of test devices, or the aging test is identified as a failure test due to adverse effects such as additional stress introduced during the test due to product performance degradation, resulting in ineffective waste of resources. At the same time, ensuring the quality of high-temperature aging tests can avoid over-aging and under-aging failures in integrated circuits, greatly reduce the failure rate of integrated circuits, and avoid large-scale electronic system failures such as new energy vehicles, civil airliners, and energy storage transmission stations due to integrated circuit failures.

[0024] Therefore, developing an intelligent test quality assurance system not only has important economic value, but also can help my country's high-end test equipment move to the forefront of the world.

[0025] Currently, active protection technology is difficult to adapt to high-temperature aging products. The key difficulties include: (1) inconsistent test environment makes it difficult to determine the health benchmark; (2) the multiple structural levels of the product make it difficult to calculate the performance degradation trends of different levels; (3) the complex composition of fault sources makes it difficult to construct fault self-healing strategies.

[0026] In addition, traditional time series extrapolation prediction methods usually adopt the strategy of time series decomposition, by decomposing the time series into trend terms, seasonal terms, residual terms, etc., and predicting them separately, and finally integrating the prediction results to obtain the time series extrapolation prediction sequence of the parameters. Although the output signal can reflect the degradation process of the device or even the module, this change is relatively weak. At the same time, considering the training and reaction time of the later model, the original signal is obviously not suitable for the prediction model. Therefore, it is necessary to extract data that can represent the fault characteristics from a large number of original signals, that is, to reduce the dimension and remove noise of the original data. For switching power supply circuits, due to the existence of tolerance, nonlinearity, etc., the operating frequency is high, so both high and low frequency signals are rich, and there are relatively high requirements for feature extraction.

[0027] In response to the above problems and the urgent need to independently ensure the performance of high-end test equipment in key areas, accurate monitoring of the health characteristics of high-temperature aging products can be carried out to ensure the quality of the long-term operating environment of the test based on the health characteristics, meet the needs of large-scale integrated circuit batch aging, and the relevant technical system can be effectively expanded to the same type of aging system to enhance the key scientific and technological strength of the integrated circuit testing industry.

[0028] Based on this, the embodiment of this specification proposes a feature extraction method based on dual self-encoding time-frequency feature fusion. This method combines frequency domain features with self-attention features, realizes feature fusion through a secondary self-encoding mechanism, and can directly map the temporal dependency and change trend of the original parameters to the hidden layer depth characteristics, avoiding the problem of sequence decomposition in traditional methods, and providing a more practical method for the extrapolation prediction of the degradation timing of key parameters of the isolated switching power supply at the front end of the aging station.

[0029] In this specification, a method for extracting the health characteristics of an isolated switching power supply at the front end of a aging station is provided. This specification also relates to a device for extracting the health characteristics of an isolated switching power supply at the front end of a aging station, a computing device, a computer-readable storage medium, and a computer program, which are described in detail one by one in the following embodiments.

[0030] Figure 1 A flow chart of a method for extracting health characteristics of a front-end isolated switching power supply of a burn-in station provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0031] Step 102: Acquire multiple historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station.

[0032] Specifically, the aging bench is an integrated circuit high-temperature aging test bench, which accelerates various physical and chemical reaction processes inside the components by continuously applying a certain amount of electrical stress to the components for a long time, prompting various potential faults inside the components to be exposed early, thereby eliminating early failure products and making the electronic components enter a period of low failure rate and relative stability from the beginning of use. The front-end isolated switching power supply is used to provide the required power for the entire aging test bench.

[0033] Since the isolated switching power supply at the front end of the aging test bench will generate monitoring signals such as current and voltage during operation, the analog signal described in the embodiment of this specification may include voltage signals and current signals, and the current signal may include capacitor current signals and inductor current signals, and the voltage signal may include average output voltage signals and ripple voltage signals.

[0034] When acquiring the analog signal of the isolated switching power supply at the front end of the aging station, high-precision sensors are used to collect the output voltage and current signals to ensure that the signal accuracy error is less than ±0.5%. For sensitive signals (such as capacitor current and inductor current), pre-processing is performed through a digital signal amplifier, and the gain coefficient is dynamically adjusted to enhance the characteristics of weak signals and avoid feature loss caused by noise interference. In addition, according to the preset accuracy indicators (such as power supply output stability requirements), the sampling frequency and signal amplification factor are adaptively adjusted to ensure that the signal-to-noise ratio of key signals in the feature extraction stage meets the adaptive threshold.

[0035] In addition, the multiple historical signal values ​​described in the embodiments of this specification may be historical time series data of analog signals.

[0036] Step 104: Encode the multiple historical signal values ​​through the Transformer model encoder in the multi-layer model encoder to obtain self-attention features in the multiple historical signal values, wherein the multi-layer model encoder is a neural network structure that integrates context features and is constructed by the Transformer model encoder and the stacked autoencoder, including an input layer, an intermediate layer, and an output layer, and the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder.

[0037] In an optional embodiment, the multi-layer model encoder further includes a multi-head attention layer, a frequency domain feature extraction layer and a fusion layer, wherein the frequency domain feature extraction layer is used to extract frequency domain features, and the fusion layer is used to fuse the data features extracted by the Transformer model encoder and the frequency domain feature extraction layer; Accordingly, the method further comprises: Obtain a transformation result generated by the multi-head attention layer performing a linear transformation on the multiple historical signal values; Encoding the transformation result through the Transformer model encoder to generate self-attention features in the multiple historical signal values; Performing frequency domain feature extraction on the transformation result through the frequency domain feature extraction layer to obtain frequency domain features in the multiple historical signal values ​​output by the frequency domain feature extraction layer; The self-attention feature and the frequency domain feature are fused through the fusion layer to obtain the target fusion feature output by the fusion layer.

[0038] The step of fusing the self-attention feature and the frequency domain feature through the fusion layer to obtain the target fusion feature output by the fusion layer includes: The self-attention feature and the frequency domain feature are concatenated through the fusion layer to obtain the target fusion feature output by the fusion layer.

[0039] Specifically, the multi-layer model encoder is a multi-layer Transformer model encoder, which includes an input layer, an intermediate layer and an output layer. The intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder, wherein the Transformer model encoder includes a multi-head attention sublayer, a fully connected layer and a feedforward neural network; in addition, the multi-layer Transformer model encoder also includes a frequency domain feature extraction layer and a fusion layer. The multi-head attention sublayer is used to extract multi-head attention features in historical time series data; the frequency domain feature extraction layer is used to extract frequency domain features in historical time series data; the fusion layer is used to fuse the multi-head attention features extracted by the multi-head attention sublayer with the frequency domain features extracted by the frequency domain feature extraction layer, so as to achieve the purpose of fully extracting features.

[0040] In an optional implementation, after obtaining multiple historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station, the method further includes: Performing sliding window cutting on the plurality of historical signal values ​​according to a first window width and a first step length to generate a first sliding window cutting result, and determining an initial sample data set based on the first sliding window cutting result; Normalizing the initial sample data set to generate a target sample data set; Accordingly, encoding the multiple historical signal values ​​by the Transformer model encoder in the multi-layer model encoder includes: The target sample data set is encoded by a Transformer model encoder in a multi-layer model encoder.

[0041] A flow chart of a feature extraction process provided in the embodiment of this specification is as follows: Figure 2 It includes the following steps.

[0042] Step 202: Obtain fault prediction data of the isolated switching power supply at the front end of the aging station.

[0043] Step 204: Perform comprehensive preprocessing on the fault data.

[0044] Step 206: Perform sliding window cutting on the preprocessing result according to the first window width and the first step length to generate a first sliding window cutting result, and determine an initial sample data set based on the first sliding window cutting result.

[0045] Step 208: normalize the initial sample data set to generate a target sample data set.

[0046] Step 210: construct a training data set and a test data set based on the target sample data set.

[0047] Step 212: Perform multi-head attention feature extraction on the training dataset through the Transformer model encoder.

[0048] Step 214: extract frequency domain features from the training data set.

[0049] In the process of frequency domain feature extraction, for sensitive frequency bands (such as the resonant frequency region of the isolated switching power supply at the front end of the aging station), the energy spectrum density of the target frequency band is enhanced through a bandpass filter, and the feature significance is evaluated in combination with the power spectrum entropy algorithm. The frequency domain feature weights are adaptively selected. For example, in scenarios with high accuracy requirements, the weight ratio of the center frequency and shape factor is increased to improve the discrimination of degradation features.

[0050] Step 216: Perform deep feature fusion on the spliced ​​features through a stacked autoencoder to obtain the health features corresponding to the isolated switching power supply at the front end of the aging station.

[0051] Step 218: Output health characteristics.

[0052] The feature extraction process of the embodiment of this specification is described in detail below.

[0053] To obtain the fault prediction data of the front-end isolated switching power supply of the aging station, the voltage and current signals of the front-end isolated switching power supply in the high-temperature dynamic aging detection system of the integrated circuit can be monitored by sensors to obtain the original fault prediction data of the front-end isolated switching power supply of the aging station, that is, the historical time series data of the voltage and current signals.

[0054] For comprehensive preprocessing of fault data, the collected original fault prediction data can be sent to the fault data preprocessing unit for comprehensive processing to obtain a training data set and a test data set. This can be achieved through the following steps: Step a: Perform sliding window cutting on the historical time series data to construct the initial sample data set.

[0055] Specifically, the historical time series data of any front-end isolated switching power supply collected by the sensor is X, , perform sliding window cutting on X to generate the corresponding initial sample data set. When the first window width is W and the first step length is S, the number of samples generated by cutting is:

[0056] The corresponding initial sample data set is generated as ,right Each sample in Take the length The data is used as training data, and W- The data of length is used as the prediction data corresponding to this section of training data.

[0057] Step b: Perform maximum and minimum normalization on the training data set.

[0058] In order to improve the data expression ability and speed up the convergence of subsequent model training, it is necessary to normalize the training data set, mainly by scaling the amplitude of the original parameters through the maximum and minimum normalization method to complete the linear transformation of the data. , the normalization process is realized through the following formula to obtain the normalized sample data set .

[0059]

[0060] Step c: Construct training data set and test data set.

[0061] The first r% of the data are selected from all the data as the training data set, and the remaining data are used as the test data set to verify the model prediction performance. Generally speaking, r is generally 60-80, and r in the embodiment of this specification is 70.

[0062] For multi-head attention feature extraction of processed fault data based on the Transformer model encoder, the training data set obtained after processing by the comprehensive data preprocessing module can be sent to the Transformer model encoder and the frequency domain feature extraction layer respectively to obtain self-attention features and frequency domain features.

[0063] A schematic diagram of the structure of a Transformer model encoder provided in the embodiments of this specification is as follows Figure 3 shown.

[0064] Another feature extraction process flow chart provided in the embodiment of this specification is as follows: Figure 4 The specific steps include the following steps.

[0065] Step 402: Obtain a training data set.

[0066] Step 404: Construct a Transformer model encoder.

[0067] Step 406: Train and construct the Transformer model encoder.

[0068] Step 408: Perform self-attention feature extraction on the training data set based on the trained Transformer model encoder.

[0069] Step 410: Perform sliding window cutting on the normalized target sample data set.

[0070] Step 412: extract at least two initial frequency domain features based on the sliding window cutting result.

[0071] Step 414: normalize at least two initial frequency domain features.

[0072] Step 416: perform feature concatenation on the self-attention features and the frequency domain features.

[0073] When performing multi-head attention feature extraction on processed fault data based on the Transformer model encoder, the Transformer model encoder can be first constructed using the training data set. Specifically, the training data set can be used to pre-train the model. The Transformer model usually requires that the input data is three-dimensional, so a training data set needs to be constructed. The purpose of constructing a training data set is to meet the input requirements of the model. The method is to convert the data format of the training data set into ,in, is the sample size, is the data length of each sample, and 3 is the number of features, which represents the three features of output voltage signal, capacitor current signal and inductor current signal at each time point. After the training data set is constructed, the constructed training sample data set can be input into Figure 3 The Transformer model encoder shown.

[0074] The Transformer model is a model built on the Seq-to-Seq framework. Compared with the classic deep learning model, the most prominent advantage of the Transformer model is the use of a multi-head attention mechanism. The purpose of the multi-head attention sublayer is to give different importance to words / tokens in the sequence from multiple aspects. The Transformer model is mainly composed of input, encoder, decoder and output.

[0075] The position encoding layer in the Transformer model is to determine the position information of the sequence. Since there is no recursive layer and convolution layer in RNN and CNN, the input order information cannot be obtained by relying solely on the self-attention mechanism. The order information of the sequence needs to be actively passed to the model. The Transformer model uses a combination of sine and cosine functions to encode the position of the sequence. The calculation method is as follows:

[0076]

[0077] Among them, pos is the position of the current sequence; i is the dimension; is the dimension of the input features.

[0078] The multi-head attention mechanism in the Transformer model uses multiple attention mechanisms to perform operations in parallel, and then concatenates the operation results through linear transformation. The core technology in the Transformer model is the multi-head attention mechanism, which is used to extract the dependency features between data, capture the correlation between data and establish a context prediction model. The calculation method is as follows:

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] Among them, Q is the query matrix; K is the key matrix; V is the value matrix; , , is a trainable parameter matrix; X is the processed input; is the dimension of the key matrix; , , , is the learnable parameter matrix.

[0085] In the Transformer model, the encoding and decoding parts also include feedforward networks, summation, and normalization. The calculation formula of the feedforward neural network is as follows:

[0086] Where x is the input; , , , are parameters that can be obtained through training.

[0087] The calculation formula for summation and normalization is as follows:

[0088] Where x is the input; This is the result after module processing.

[0089] The model processing flow is as follows: the input data is sent to the encoding part after position encoding, the output of the encoding part is flattened and sent to the decoder to map the feature parameters in the high-dimensional hidden layer to the original input data, thereby training the model's feature extraction ability.

[0090] Secondly, select the appropriate number of iterations and loss function, input the constructed data set into the feature extraction model to repeatedly perform forward propagation and back propagation iterative calculation processes; in this process, the model parameters of the embedding dimension, number of attention heads, encoder and decoder layers are continuously adjusted to complete the pre-training of the model.

[0091] Again, take out the encoding layer and decoding layer of the pre-trained model and retain their weight parameters to construct them as the trained Transformer model encoder.

[0092] Finally, based on the pre-trained Transformer model encoder, the training dataset Perform self-attention feature extraction to obtain the self-attention feature set .

[0093] Step 106: extract the frequency domain features of the multiple historical signal values, and fuse the self-attention features with the frequency domain features to generate target fusion features.

[0094] In an optional implementation, extracting frequency domain features of the multiple historical signal values ​​includes: The target sample data set is subjected to sliding window cutting according to a second window width and a second step length to generate a second sliding window cutting result, and frequency domain features of the plurality of historical signal values ​​are extracted based on the second sliding window cutting result.

[0095] Further, the extracting frequency domain features of the plurality of historical signal values ​​based on the second sliding window cutting result includes: Performing Fourier transform on each to-be-processed sample data set contained in the second sliding window cutting result to generate a time series energy spectrum corresponding to each to-be-processed sample data set; Respectively extracting eigenvalues ​​corresponding to at least two initial frequency domain features of the time series energy spectrum; The characteristic values ​​are normalized using a maximum and minimum normalization algorithm to generate frequency domain features of the multiple historical signal values.

[0096] The at least two initial frequency domain features include: peak value, bandwidth, center frequency, shape factor, and power spectral density.

[0097] Specifically, the frequency domain features of the multiple historical signal values ​​are extracted, and the training data set generated by sliding window cutting can be specifically Perform frequency domain feature extraction based on expert knowledge.

[0098] Furthermore, the target sample data set generated by the sliding window cutting mentioned above can be subjected to sliding window cutting according to the second window width and the second step size to generate a second sliding window cutting result, and frequency domain features of multiple historical signal values ​​can be extracted based on the second sliding window cutting result.

[0099] Among them, when the target sample data set is cut by sliding window, if the second window width is , the second step length is 1, for the sample Can be cut out samples, each sample length is , that is, .

[0100] After the sliding window cutting is completed, Fourier transform can be performed on each sample data set to be processed contained in the second sliding window cutting result to convert it into the frequency domain to obtain the corresponding frequency domain representation F′ (time series energy spectrum).

[0101] For each frequency domain representation F′ of the sample data set to be processed, five initial frequency domain features are extracted, namely, the peak value, bandwidth, center frequency, shape factor, and power spectrum density of the energy spectrum. Each initial frequency domain feature includes a frequency domain feature and the eigenvalue corresponding to the frequency domain feature. The initial frequency domain features extracted are , so for the sample The initial frequency domain features extracted are After the extraction is completed, the eigenvalues ​​of each initial frequency domain feature can be normalized using the maximum and minimum normalization algorithm to generate frequency domain features of multiple historical signal values.

[0102] After the frequency domain features and self-attention features (high-dimensional hidden layer features) are extracted, the extracted frequency domain features and self-attention features can be concatenated.

[0103] For the training data set Each sample Perform self-attention feature extraction and frequency domain feature extraction and feature fusion. Let the dimension of the fused feature be , the training data set can be reorganized into The two-dimensional fusion feature matrix of .

[0104] Step 108: Encode the target fusion features through the stacked autoencoder in the multi-layer model encoder to generate health features corresponding to the front-end isolated switching power supply of the aging station.

[0105] Specifically, an autoencoder is an unsupervised learning method that aims to learn effective representations or encodings of data and achieve self-learning by reconstructing input data. Stacked autoencoders (SAEs) are an extension of this method. Through multiple layers of nonlinear transformations, the original data is mapped to an increasingly abstract and compact feature space, thereby achieving deeper feature learning and representation learning, which is of great significance for dimensionality reduction, feature extraction, and classification tasks of complex data.

[0106] In the embodiments of the present specification, the target fusion features are encoded by a stacked autoencoder in the multi-layer model encoder, specifically including constructing a stacked autoencoder and a decoder, training the stacked autoencoder and the decoder, and using the stacked autoencoder to encode the target fusion features.

[0107] A schematic diagram of the structure of a stacked autoencoder provided in one embodiment of this specification is as follows Figure 5 shown.

[0108] First, for building a stacked autoencoder and decoder, the model structure of the stacked autoencoder is as follows Figure 5 As shown in the figure, the number of encoding layers is the same as the number of decoding layers, which enables the model to have better secondary encoding capabilities for deep features.

[0109] Secondly, the two-dimensional fusion feature matrix obtained in the previous steps is used to pre-train the stacked autoencoder model. The two-dimensional fusion feature matrix is ​​used as the input and output of the stacked autoencoder model. An appropriate loss function and number of iterations are selected to complete the forward propagation and back propagation iterative calculation process, so that the model continuously reconstructs its own input, and finally extracts the encoding layer from the pre-trained stacked autoencoder model as a usable stacked autoencoder.

[0110] Finally, the fusion features are secondary self-encoded based on the pre-trained stacked autoencoder to obtain the secondary encoding feature set , the secondary coding feature set is the health feature corresponding to the isolated switching power supply at the front end of the aging station.

[0111] After obtaining a plurality of health features, the health features may also be binned, and based on at least two binned data sets included in the binning results, feature vectors of the isolated switching power supply at the front end of the aging station are respectively constructed; A training sample set is constructed based on the feature vector and the state label corresponding to the feature vector, and a support vector machine diagnostic model to be trained is trained based on the training sample set, wherein the trained support vector machine diagnostic model is used to perform fault detection and diagnosis on the isolated switching power supply at the front end of the aging station.

[0112] The training sample set includes a training sample subset and a test sample subset, and the state label includes a healthy state label and a fault state label.

[0113] In an optional implementation, the health features may be binned using a K-means binning algorithm.

[0114] In another optional implementation, value range reference information and change rule reference information corresponding to the health feature may also be obtained; Determine the abnormal value interval corresponding to the health feature according to the fault information of the isolated switching power supply at the front end of the aging station; Determine the binning interval corresponding to the health feature based on the value range reference information, the change rule reference information and the abnormal value interval; The health features are binned according to the binning intervals.

[0115] Specifically, based on the parameter distribution range of the isolated switching power supply at the front end of the aging station under normal operating conditions (healthy conditions), the normal interval of the health characteristics is determined; combined with the impact of possible fault scenarios on the health characteristics in actual scenarios, the abnormal interval of the health characteristics is delineated to facilitate the distinction of fault types.

[0116] Through the above binning and discretization processing, continuous performance parameters can be converted into discrete performance parameters, which improves the comparability of data and reduces the complexity of model training. The binning interval setting and discretization results provide efficient and accurate input features for the subsequent support vector machine diagnostic model to be trained.

[0117] In an optional implementation, the step of constructing the feature vectors of the front-end isolated switching power supply of the aging station based on at least two binned data sets included in the binning processing result includes: Based on the target health features contained in the target binned data set, feature extraction is performed on the target binned data set, and a feature vector of the front-end isolated switching power supply of the aging station is constructed based on the feature extraction result, wherein the target binned data set is each of the at least two binned data sets.

[0118] Furthermore, the step of extracting features from the target binned data set based on the target health features contained in the target binned data set includes: Based on the target health features contained in the target binned data set, feature extraction is performed on the target binned data set to obtain the corresponding average value, standard deviation and peak value of the target health features.

[0119] Specifically, the target binned data set is each of the at least two binned data sets, and the health features included in the target binned data sets are the target health features.

[0120] Alternatively, the similarity between different degradation paths of the front-end isolated switching power supply of the aging station may also be calculated, wherein different degradation paths are composed of health features corresponding to different historical time series; Clustering different degradation paths based on the similarity to generate at least two clustering results; The health features corresponding to the degradation paths contained in different clustering results are respectively used as training data to input the corresponding fault prediction models to be trained to obtain the trained fault prediction models; The number of the fault prediction models to be trained is equal to the number of the at least two clustering results, and each fault prediction model to be trained corresponds to each clustering result one by one.

[0121] In practical applications, the fault prediction model to be trained is a multidimensional recurrent neural network model, including an input layer, an intermediate layer and an output layer. The intermediate layer includes a sequentially connected neighbor component analysis model structure and a multidimensional recurrent neural network model structure. The multidimensional recurrent neural network model structure includes a fully connected layer, a bidirectional long short-term memory network layer, a bidirectional gated recurrent unit layer and a splicing layer.

[0122] Furthermore, the health characteristics corresponding to different degradation paths of the isolated switching power supply at the front end of the aging station can be processed through a neighbor component analysis model structure to obtain the similarity between different degradation paths, wherein the neighbor component analysis model structure processes the health characteristics corresponding to different degradation paths of the isolated switching power supply at the front end of the aging station through a neighbor component analysis algorithm.

[0123] Alternatively, the middle layer includes a neighbor component analysis model structure, a Gaussian mixture model structure, and a multi-dimensional recurrent neural network model structure connected sequentially; Based on this, different degradation paths may be clustered based on the similarity through the Gaussian mixture model structure, wherein the Gaussian mixture model structure clusters different degradation paths based on the similarity through a Gaussian mixture algorithm.

[0124] The embodiments of this specification provide active protection technology for high-temperature aging test benches for integrated circuits to achieve the integrity of the test process and the consistency of the test environment stress, and can reduce the significant property losses caused by the damage of millions of test devices due to forced interruption of the test process due to machine failure, or can reduce the adverse effects such as the introduction of additional stress during the test due to machine performance degradation, so as to avoid the aging test being identified as a failure test, thereby helping to avoid waste of resources. At the same time, ensuring the quality of high-temperature aging tests can avoid over-aging and under-aging failures of integrated circuits, greatly reduce the failure rate of integrated circuits, and help avoid the failure and shutdown of large-scale electronic systems such as new energy vehicles, civil airliners, energy storage and transmission stations using the same integrated circuits due to integrated circuit failures.

[0125] The embodiment of this specification obtains multiple historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station, encodes the multiple historical signal values ​​through the Transformer model encoder in the multi-layer model encoder, and obtains the self-attention features in the multiple historical signal values, wherein the multi-layer model encoder is a neural network structure that integrates context features constructed by the Transformer model encoder and the stacked autoencoder, including an input layer, an intermediate layer and an output layer, and the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder; extracts the frequency domain features of the multiple historical signal values, and fuses the self-attention features with the frequency domain features to generate target fusion features; encodes the target fusion features through the stacked autoencoder in the multi-layer model encoder to generate the health features corresponding to the isolated switching power supply at the front end of the aging station. The embodiment of this specification encodes the historical signal values ​​of the analog signal once through the Transformer model encoder to obtain the self-attention features, then extracts the frequency domain features of the historical signal values ​​of the analog signal, and then encodes the splicing results of the self-attention features and the frequency domain features through the stacked autoencoder to achieve deep feature fusion. Through this processing method, the timing dependency and change trend between multiple historical signal values ​​can be directly mapped into the hidden layer deep features, which is helpful to ensure the accuracy of the health feature output results of the isolated switching power supply at the front end of the aging station.

[0126] Corresponding to the above method embodiment, this specification also provides an embodiment of a health feature extraction device for an isolated switching power supply at the front end of a burn-in station, Figure 6 FIG. 1 is a schematic diagram showing a healthy feature extraction device for a front-end isolated switching power supply of a burn-in station provided by an embodiment of the present specification. Figure 6 As shown, the device comprises: The acquisition module 602 is configured to acquire multiple historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station; A first encoding module 604 is configured to encode the multiple historical signal values ​​through a Transformer model encoder in a multi-layer model encoder to obtain self-attention features in the multiple historical signal values, wherein the multi-layer model encoder is a neural network structure that integrates context features and is constructed by the Transformer model encoder and a stacked autoencoder, including an input layer, an intermediate layer, and an output layer, wherein the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder; A feature fusion module 606 is configured to extract frequency domain features of the multiple historical signal values, and fuse the self-attention feature with the frequency domain feature to generate a target fusion feature; The second encoding module 608 is configured to encode the target fusion features through the stacked autoencoder in the multi-layer model encoder to generate health features corresponding to the front-end isolated switching power supply of the aging station.

[0127] Optionally, the multi-layer model encoder further includes a multi-head attention layer, a frequency domain feature extraction layer and a fusion layer, wherein the frequency domain feature extraction layer is used to extract frequency domain features, and the fusion layer is used to fuse the data features extracted by the Transformer model encoder and the frequency domain feature extraction layer; Accordingly, the apparatus further includes a processing module 610 configured to: Obtain a transformation result generated by the multi-head attention layer performing a linear transformation on the multiple historical signal values; Encoding the transformation result through the Transformer model encoder to generate self-attention features in the multiple historical signal values; Performing frequency domain feature extraction on the transformation result through the frequency domain feature extraction layer to obtain frequency domain features in the multiple historical signal values ​​output by the frequency domain feature extraction layer; The self-attention feature and the frequency domain feature are fused through the fusion layer to obtain the target fusion feature output by the fusion layer.

[0128] Optionally, the processing module 610 is further configured to: Performing sliding window cutting on the plurality of historical signal values ​​according to a first window width and a first step length to generate a first sliding window cutting result, and determining an initial sample data set based on the first sliding window cutting result; Normalizing the initial sample data set to generate a target sample data set; Accordingly, the first encoding module 604 is further configured to: The target sample data set is encoded by a Transformer model encoder in a multi-layer model encoder.

[0129] Optionally, the feature fusion module 606 is further configured to: The target sample data set is subjected to sliding window cutting according to a second window width and a second step length to generate a second sliding window cutting result, and frequency domain features of the plurality of historical signal values ​​are extracted based on the second sliding window cutting result.

[0130] Optionally, the feature fusion module 606 is further configured to: Performing Fourier transform on each to-be-processed sample data set contained in the second sliding window cutting result to generate a time series energy spectrum corresponding to each to-be-processed sample data set; Respectively extracting eigenvalues ​​corresponding to at least two initial frequency domain features of the time series energy spectrum; The characteristic values ​​are normalized using a maximum and minimum normalization algorithm to generate frequency domain features of the multiple historical signal values.

[0131] Optionally, the at least two initial frequency domain features include: peak value, bandwidth, center frequency, shape factor, and power spectral density.

[0132] Optionally, the processing module 610 is further configured to: The self-attention feature and the frequency domain feature are concatenated through the fusion layer to obtain the target fusion feature output by the fusion layer.

[0133] The above is a schematic scheme of a health feature extraction device for an isolated switching power supply at the front end of an aging station of this embodiment. It should be noted that the technical scheme of the health feature extraction device for the isolated switching power supply at the front end of the aging station and the technical scheme of the health feature extraction method for the isolated switching power supply at the front end of the aging station belong to the same concept. For details not described in detail in the technical scheme of the health feature extraction device for the isolated switching power supply at the front end of the aging station, please refer to the description of the technical scheme of the health feature extraction method for the isolated switching power supply at the front end of the aging station.

[0134] Figure 7 The block diagram of a computing device 700 according to an embodiment of the present specification is shown. The components of the computing device 700 include but are not limited to a memory 710 and a processor 720. The processor 720 is connected to the memory 710 via a bus 730, and the database 750 is used to store data.

[0135] The computing device 700 also includes an access device 740 that enables the computing device 700 to communicate via one or more networks 760. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 740 may include one or more of any type of network interface (e.g., a network interface card (NIC)) that is wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a World Wide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0136] In one embodiment of the present specification, the above components of the computing device 700 and Figure 7 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 7 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0137] The computing device 700 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or PC. The computing device 700 may also be a mobile or stationary server.

[0138] The processor 720 is used to execute the following computer executable instructions, which, when executed by the processor, implement the steps of the above-mentioned method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station.

[0139] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station mentioned above belong to the same concept. For the details not described in detail in the technical scheme of the computing device, please refer to the description of the technical scheme of the method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station mentioned above.

[0140] An embodiment of the present specification also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned method for extracting health characteristics of the isolated switching power supply at the front end of the aging station.

[0141] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station mentioned above belong to the same concept. For details not described in detail in the technical scheme of the storage medium, please refer to the description of the technical scheme of the method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station mentioned above.

[0142] An embodiment of the present specification further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station.

[0143] The above is a schematic scheme of a computer program of this embodiment. It should be noted that the technical scheme of the computer program and the technical scheme of the method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station mentioned above belong to the same concept. For details not described in detail in the technical scheme of the computer program, please refer to the description of the technical scheme of the method for extracting the health characteristics of the isolated switching power supply at the front end of the aging station mentioned above.

[0144] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0145] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0146] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0147] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0148] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A method for extracting health features of a front-end isolated switching power supply of a burn-in station, comprising: Obtain multiple historical signal values ​​of the analog signal of the isolated switching power supply at the front end of the aging station; The plurality of historical signal values ​​are encoded by a Transformer model encoder in a multi-layer model encoder to obtain self-attention features in the plurality of historical signal values, wherein the multi-layer model encoder is a neural network structure for integrating context features constructed by the Transformer model encoder and a stacked autoencoder, including an input layer, an intermediate layer, and an output layer, wherein the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder; Extracting frequency domain features of the multiple historical signal values, and fusing the self-attention features with the frequency domain features to generate target fusion features; The target fusion features are encoded by the stacked autoencoder in the multi-layer model encoder to generate health features corresponding to the isolated switching power supply at the front end of the aging station.

2. According to the method for extracting healthy features of the front-end isolated switching power supply of the burn-in station in claim 1, the multi-layer model encoder further comprises a multi-head attention layer, a frequency domain feature extraction layer and a fusion layer, the frequency domain feature extraction layer is used to extract frequency domain features, and the fusion layer is used to fuse the data features extracted by the Transformer model encoder and the frequency domain feature extraction layer; Accordingly, the method further comprises: Obtain a transformation result generated by the multi-head attention layer performing a linear transformation on the multiple historical signal values; Encoding the transformation result through the Transformer model encoder to generate self-attention features in the multiple historical signal values; Performing frequency domain feature extraction on the transformation result through the frequency domain feature extraction layer to obtain frequency domain features in the multiple historical signal values ​​output by the frequency domain feature extraction layer; The self-attention feature and the frequency domain feature are fused through the fusion layer to obtain the target fusion feature output by the fusion layer.

3. The method for extracting health features of the front-end isolated switching power supply of the aging station according to claim 1, after obtaining multiple historical signal values ​​of the analog signal of the front-end isolated switching power supply of the aging station, further comprising: Performing sliding window cutting on the plurality of historical signal values ​​according to a first window width and a first step length to generate a first sliding window cutting result, and determining an initial sample data set based on the first sliding window cutting result; Normalizing the initial sample data set to generate a target sample data set; Accordingly, encoding the multiple historical signal values ​​by the Transformer model encoder in the multi-layer model encoder includes: The target sample data set is encoded by a Transformer model encoder in a multi-layer model encoder.

4. The method for extracting healthy features of a front-end isolated switching power supply of a burn-in station according to claim 3, wherein the step of extracting frequency domain features of the plurality of historical signal values ​​comprises: The target sample data set is subjected to sliding window cutting according to a second window width and a second step length to generate a second sliding window cutting result, and frequency domain features of the plurality of historical signal values ​​are extracted based on the second sliding window cutting result.

5. The method for extracting healthy features of the front-end isolated switching power supply of the burn-in station according to claim 4, wherein the step of extracting the frequency domain features of the plurality of historical signal values ​​based on the second sliding window cutting result comprises: Performing Fourier transform on each to-be-processed sample data set contained in the second sliding window cutting result to generate a time series energy spectrum corresponding to each to-be-processed sample data set; Respectively extracting eigenvalues ​​corresponding to at least two initial frequency domain features of the time series energy spectrum; The characteristic values ​​are normalized using a maximum and minimum normalization algorithm to generate frequency domain features of the multiple historical signal values.

6. The method for extracting healthy features of a front-end isolated switching power supply of a burn-in station according to claim 5, wherein the at least two initial frequency domain features include: Peak value, bandwidth, center frequency, shape factor, power spectral density.

7. According to the method for extracting healthy features of the front-end isolated switching power supply of the burn-in station in claim 2, the self-attention feature and the frequency domain feature are fused through the fusion layer to obtain the target fusion feature output by the fusion layer, comprising: The self-attention feature and the frequency domain feature are concatenated through the fusion layer to obtain the target fusion feature output by the fusion layer.

8. A health feature extraction device for a front-end isolated switching power supply of a burn-in station, comprising: An acquisition module is configured to acquire multiple historical signal values ​​of an analog signal of an isolated switching power supply at the front end of the aging station; A first encoding module is configured to encode the multiple historical signal values ​​through a Transformer model encoder in a multi-layer model encoder to obtain self-attention features in the multiple historical signal values, wherein the multi-layer model encoder is a neural network structure that integrates context features and is constructed by the Transformer model encoder and a stacked autoencoder, including an input layer, an intermediate layer, and an output layer, wherein the intermediate layer includes a sequentially connected Transformer model encoder structure and a stacked autoencoder; A feature fusion module is configured to extract frequency domain features of the multiple historical signal values, and fuse the self-attention feature with the frequency domain feature to generate a target fusion feature; The second encoding module is configured to encode the target fusion features through the stacked autoencoder in the multi-layer model encoder to generate health features corresponding to the isolated switching power supply at the front end of the aging station.

9. A computing device comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the method for extracting health characteristics of the front-end isolated switching power supply of the aging station described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for extracting health characteristics of a front-end isolated switching power supply of a burn-in station as claimed in any one of claims 1 to 7.

Images