A method for detecting power failure of new energy vehicles

Abstract

The present invention discloses a method for detecting power failure of new energy vehicles, and belongs to the technical field of power failure of new energy vehicles. The present invention collects the output current signals of the battery area, BMS and BUCK module; then, the current signal is segmented to construct a matrix, and the first and second linear relationship matrices are constructed according to the matrix; at the same time, the steady-state fluctuation value and attenuation change value of the output power signal of the BUCK module are extracted to construct the corresponding vector; finally, the power failure assessment neural network is used to process the above matrix and vector to obtain the fault classification value. This method significantly improves the accuracy of fault diagnosis through systematic data collection and analysis, and ensures the safe operation of new energy vehicles.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy vehicle power failure, and in particular to a new energy vehicle power failure detection method. Background Art

[0002] In the power system of new energy vehicles, batteries, power management systems (BMS) and BUCK modules constitute the core functional units. As the basic unit for energy storage and conversion, batteries directly determine the vehicle's endurance and performance. The power management system BMS is responsible for fine control of each battery and transmits the current and voltage output by the battery to the BUCK module. As a key component for voltage conversion and regulation, the BUCK module achieves stable voltage output through precise voltage reduction, ensuring that the subsequent circuit can obtain the required electrical energy.

[0003] Existing new energy vehicle power supply fault detection methods mainly rely on static indicators such as voltage, temperature, and internal resistance. This single parameter monitoring method cannot dynamically and comprehensively reflect the real-time working status of the battery pack. Due to the lack of in-depth analysis of the complex dynamic relationship between batteries, BMS, and BUCK modules, traditional detection methods are difficult to accurately capture subtle changes and potential fault signals within the power system. This detection limitation directly leads to low fault diagnosis accuracy and the inability to timely detect and prevent possible power system abnormalities. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, a method for detecting power failure of a new energy vehicle provided by the present invention solves the problem of low fault diagnosis accuracy in the prior art.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a new energy vehicle power failure detection method, comprising the following steps:

[0006] S1. Divide the battery pack in the fully charged new energy vehicle power supply into multiple battery areas, discharge each battery area, and collect the output current signal of the battery area, the output current signal of the power management system BMS, and the output current signal of the BUCK module;

[0007] S2, dividing each output current signal into an early current signal, a mid-term current signal and a late current signal, and constructing an output current matrix;

[0008] S3, constructing a first linear relationship matrix according to the output current matrix of the battery area and the power management system BMS, and constructing a second linear relationship matrix according to the output current matrix of the power management system BMS and the BUCK module;

[0009] S4. According to the output power signal of the BUCK module, the steady-state fluctuation value and the attenuation change value are extracted, and the steady-state fluctuation vector and the attenuation change vector are constructed;

[0010] S5. Use a power supply fault assessment neural network to process the first linear relationship matrix, the second linear relationship matrix, the steady-state fluctuation vector and the attenuation change vector to obtain a new energy vehicle power supply fault classification value.

[0011] Further, S2 includes the following sub-steps:

[0012] S21, marking the time point when the output current signal of the battery area rises to the rated current value as the first dividing point, and marking the time point when the output current signal of the battery area drops to the rated current value as the second dividing point;

[0013] S22, dividing each output current signal into three sections according to the positions of the first dividing point and the second dividing point, to obtain an early current signal, a mid-term current signal and a late current signal;

[0014] S23, respectively collect N current values ​​for the early current signal, the mid-term current signal and the late current signal, and construct an output current matrix: , where E is the output current matrix, X1 is a column vector consisting of N current values ​​of the early current signal, X2 is a column vector consisting of N current values ​​of the mid-term current signal, X3 is a column vector consisting of N current values ​​of the late current signal, and N is a positive integer.

[0015] Furthermore, S3 includes the following sub-steps:

[0016] S31, calculating the mean of each column vector in each output current matrix to obtain a column mean;

[0017] S32. Centering the output current matrix according to the column mean to obtain a current centering matrix, wherein the centering formula is: ,in, is the current value of the i-th column and j-th row of the current centering matrix, X i,j is the current value of the i-th column and j-th row of the output current matrix, X i,avg is the column mean of the i-th column, i and j are positive integers;

[0018] S33, constructing a first linear relationship matrix according to the current centering matrix of the battery area and the power management system BMS;

[0019] S34, constructing a second linear relationship matrix according to the current centering matrix of the power management system BMS and the BUCK module.

[0020] Furthermore, the formula for constructing the first linear relationship matrix in S33 is: , where C1 is the first linear relationship matrix, C battery is the current centering matrix of the battery area, C BMS is the current centering matrix of the power management system BMS, and T is the transposition operation;

[0021] The formula for constructing the second linear relationship matrix in S34 is: , where C2 is the second linear relationship matrix, C BUCK This is the current centering matrix of the BUCK module.

[0022] Further, S4 includes the following sub-steps:

[0023] S41. In the output power signal of the BUCK module, a signal segment greater than 0.8 times the rated power is marked as a stable segment, and a power signal after the stable segment is marked as an attenuation segment;

[0024] S42, collecting N power values ​​in the stable segment, calculating the steady-state fluctuation value of each power value, and constructing a steady-state fluctuation vector with a length of N, where N is a positive integer;

[0025] S43. Collect N+1 power values ​​in the attenuation segment, calculate the attenuation change value of each power value, and construct an attenuation change vector with a length of N.

[0026] Furthermore, the formula for calculating the steady-state fluctuation value of each power value in S42 is: , where S n is the steady-state fluctuation value of the nth power value, P S,n is the nth power value in the stable segment, P S,n+1 is the n+1th power value in the stable segment, P S,n-1 is the n-1th power value in the stable segment, n is a positive integer;

[0027] The formula for calculating the attenuation change value of each power value in S43 is: , where R n is the attenuation change value of the nth power value, P R,n is the nth power value in the attenuation section, P R,n-1 is the n-1th power value in the attenuation section, △T is P R,n With P R,n-1 The length of the time interval.

[0028] Further, the power supply fault assessment neural network in S5 includes: a steady-state fluctuation feature matrix construction layer, an attenuation change feature matrix construction layer, a first convolution layer, a second convolution layer, a third convolution layer, a fourth convolution layer, an adder A1, an adder A2, a first feature synthesis layer, a second feature synthesis layer, a first synthesis feature extraction unit, a second synthesis feature extraction unit and a classifier;

[0029] The input end of the steady-state fluctuation characteristic matrix construction layer is used to input the steady-state fluctuation vector;

[0030] The input end of the attenuation change feature matrix construction layer is used to input the attenuation change vector;

[0031] The input end of the first convolutional layer is used to input the first linear relationship matrix;

[0032] The input end of the fourth convolutional layer is used to input the second linear relationship matrix;

[0033] The input end of the second convolutional layer is connected to the output end of the steady-state fluctuation feature matrix construction layer;

[0034] The input end of the third convolutional layer is connected to the output end of the attenuation change feature matrix construction layer;

[0035] The input end of the adder A1 is connected to the output end of the second convolutional layer and the output end of the third convolutional layer respectively, and the output end thereof is connected to the first input end of the first feature synthesis layer and the first input end of the second feature synthesis layer respectively;

[0036] The second input end of the first feature synthesis layer is connected to the output end of the first convolutional layer, and the output end thereof is connected to the input end of the first synthetic feature extraction unit;

[0037] The second input end of the second feature synthesis layer is connected to the output end of the fourth convolutional layer, and the output end thereof is connected to the input end of the second synthetic feature extraction unit;

[0038] The input end of the adder A2 is connected to the output end of the first synthetic feature extraction unit and the output end of the second synthetic feature extraction unit respectively, and the output end thereof is connected to the input end of the classifier;

[0039] The output of the classifier is used as the output of the power failure assessment neural network.

[0040] Furthermore, the expression of the steady-state fluctuation characteristic matrix construction layer is: , where X S is the steady-state fluctuation characteristic matrix, P conv is the point convolution operation, T is the transposition operation, and S is the steady-state fluctuation vector;

[0041] The expression of the decay change feature matrix construction layer is: , where X R is the attenuation change feature matrix, T is the transpose operation, and R is the attenuation change vector.

[0042] Furthermore, the expressions of the feature synthesis layer are: , where H is the feature output by the feature synthesis layer, h1 is the input feature of the first input terminal of the feature synthesis layer, h2 is the input feature of the second input terminal of the feature synthesis layer, W1 is the weight matrix of h1, B1 is the bias matrix of h1, W2 is the weight matrix of h2, and B2 is the bias matrix of h2. is element-wise multiplication.

[0043] Further, the first synthetic feature extraction unit and the second synthetic feature extraction unit both include an average pooling layer and a feature enhancement layer;

[0044] The average pooling layer is used to extract the mean features of the features output by the feature synthesis layer;

[0045] The feature enhancement layer is used to adaptively enhance each feature element in the mean feature to obtain enhanced features. The expression of the feature enhancement layer is: , where G m is the mth enhanced feature value output by the first synthetic feature extraction unit or the second synthetic feature extraction unit, g m is the mth feature element in the mean feature, w m g m The weight of b m g m The bias of , m is a positive integer.

[0046] The beneficial effects of the present invention are:

[0047] 1. Processing the current signal in stages helps analyze the performance changes of the battery at different discharge stages, capture potential fault signals, and enhance the dynamics of fault diagnosis.

[0048] 2. By constructing a linear relationship matrix, the present invention can deeply analyze the relationship between the battery and the power management system BMS, and the power management system BMS and the BUCK module, identify potential fault modes, and improve the comprehensiveness of fault diagnosis.

[0049] 3. Since the output of the BUCK module is the output of the new energy vehicle power supply, extracting steady-state fluctuations and attenuation change information helps to identify performance changes of the power supply during battery discharge and further improve the sensitivity of fault detection.

[0050] 4. The present invention uses a power fault assessment neural network to process the first linear relationship matrix, the second linear relationship matrix, the steady-state fluctuation vector and the attenuation change vector to realize the relationship between the comprehensive battery and the power management system BMS, and the relationship between the power management system BMS and the BUCK module, and then combines the characteristics of the output power signal of the BUCK module to improve the accuracy of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a schematic diagram of the structure of the power supply for new energy vehicles;

[0052] Figure 2 It is a flow chart of a method for detecting power failure of new energy vehicles;

[0053] Figure 3 Schematic diagram of the architecture of the neural network for power failure assessment. DETAILED DESCRIPTION

[0054] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0055] The structural connection relationship of the new energy vehicle power supply, such as Figure 1 shown.

[0056] like Figure 2 As shown, a method for detecting a power failure of a new energy vehicle comprises the following steps:

[0057] S1. Divide the battery pack in the fully charged new energy vehicle power supply into multiple battery areas, discharge each battery area, and collect the output current signal of the battery area, the output current signal of the power management system BMS, and the output current signal of the BUCK module;

[0058] S2, dividing each output current signal into an early current signal, a mid-term current signal and a late current signal, and constructing an output current matrix;

[0059] S3, constructing a first linear relationship matrix according to the output current matrix of the battery area and the power management system BMS, and constructing a second linear relationship matrix according to the output current matrix of the power management system BMS and the BUCK module;

[0060] S4. According to the output power signal of the BUCK module, the steady-state fluctuation value and the attenuation change value are extracted, and the steady-state fluctuation vector and the attenuation change vector are constructed;

[0061] S5. Use a power supply fault assessment neural network to process the first linear relationship matrix, the second linear relationship matrix, the steady-state fluctuation vector and the attenuation change vector to obtain a new energy vehicle power supply fault classification value.

[0062] The battery of a new energy vehicle is composed of multiple single cells connected in series and parallel. The discharge of each single cell can be controlled, and the battery pack can be divided into multiple battery areas to evaluate the discharge condition of each local area.

[0063] The BUCK module is a step-down converter that has the functions of stepping down and stabilizing voltage.

[0064] In this embodiment, S2 includes the following sub-steps:

[0065] S21, marking the time point when the output current signal of the battery area rises to the rated current value as the first dividing point, and marking the time point when the output current signal of the battery area drops to the rated current value as the second dividing point;

[0066] S22, dividing each output current signal into three sections according to the positions of the first dividing point and the second dividing point, to obtain an early current signal, a mid-term current signal and a late current signal;

[0067] S23, respectively collect N current values ​​for the early current signal, the mid-term current signal and the late current signal, and construct an output current matrix: , where E is the output current matrix, X1 is a column vector consisting of N current values ​​of the early current signal, X2 is a column vector consisting of N current values ​​of the mid-term current signal, X3 is a column vector consisting of N current values ​​of the late current signal, and N is a positive integer.

[0068] The present invention divides all output current signals into early stage, middle stage and late stage according to the output current signal conditions of the battery area. The early stage reflects the start-up and rising characteristics, the middle stage reflects the stable working state, and the late stage reflects the attenuation and performance degradation.

[0069] In this embodiment, S3 includes the following sub-steps:

[0070] S31, calculating the mean of each column vector in each output current matrix to obtain a column mean;

[0071] S32. Centering the output current matrix according to the column mean to obtain a current centering matrix, wherein the centering formula is: ,in, is the current value of the i-th column and j-th row of the current centering matrix, X i,j is the current value of the i-th column and j-th row of the output current matrix, X i,avg is the column mean of the i-th column, i and j are positive integers;

[0072] S33, constructing a first linear relationship matrix according to the current centering matrix of the battery area and the power management system BMS;

[0073] S34, constructing a second linear relationship matrix according to the current centering matrix of the power management system BMS and the BUCK module.

[0074] In this embodiment, the formula for constructing the first linear relationship matrix in S33 is: , where C1 is the first linear relationship matrix, C battery is the current centering matrix of the battery area, C BMS is the current centering matrix of the power management system BMS, and T is the transposition operation;

[0075] The formula for constructing the second linear relationship matrix in S34 is: , where C2 is the second linear relationship matrix, C BUCK This is the current centering matrix of the BUCK module.

[0076] The centered matrix is ​​obtained by subtracting the mean of each column from each element in each column, which eliminates the influence of the mean and makes the data more concentrated around zero. In this way, the linear relationship between variables can be analyzed more clearly.

[0077] The present invention calculates the mean of each column vector in each output current matrix, uses the column mean to center the output current matrix, eliminates the difference in output current size of different modules, and then constructs a first linear relationship matrix based on the current centering matrix of the battery area and the power management system BMS, and constructs a second linear relationship matrix based on the current centering matrix of the power management system BMS and the BUCK module. The first linear relationship matrix and the second linear relationship matrix reflect the linear relationship between the two outputs. For example, when the first element in the first linear relationship matrix is ​​positive, it means that there is a linear relationship between the first row of the current centering matrix of the battery area and the first column of the transposition of the current centering matrix of the power management system BMS, which means that when the current of the battery area increases, the current of the BMS also tends to increase. When the power supply is normal, the output current of the power management system BMS and the output current of the BUCK module are linearly related to the output of the battery.

[0078] The current centering matrix in the present invention includes data of the early stage, the middle stage and the late stage, so the linear relationship matrix can reflect the overall linear relationship of the early stage, the middle stage and the late stage.

[0079] In this embodiment, S4 includes the following sub-steps:

[0080] S41. In the output power signal of the BUCK module, a signal segment greater than 0.8 times the rated power is marked as a stable segment, and a power signal after the stable segment is marked as an attenuation segment;

[0081] S42, collecting N power values ​​in the stable segment, calculating the steady-state fluctuation value of each power value, and constructing a steady-state fluctuation vector with a length of N, where N is a positive integer;

[0082] S43. Collect N+1 power values ​​in the attenuation segment, calculate the attenuation change value of each power value, and construct an attenuation change vector with a length of N.

[0083] In this embodiment, the formula for calculating the steady-state fluctuation value of each power value in S42 is: , where S n is the steady-state fluctuation value of the nth power value, P S,n is the nth power value in the stable segment, P S,n+1 is the n+1th power value in the stable segment, P S,n-1 is the n-1th power value in the stable segment, where n is a positive integer.

[0084] For the first power value, there is no P S,n-1 Therefore, the formula for the steady-state fluctuation value S1 of the first power value is: , the Nth power value, which does not exist P S,n+1 Therefore, the steady-state fluctuation value S of the Nth power value is N The formula is: , where P S,1 is the first power value in the stable segment, P S,2 is the second power value in the stable segment, P S,N is the Nth power value in the stable segment, P S,N-1 It is the N-1th power value in the stable segment.

[0085] The formula for calculating the attenuation change value of each power value in S43 is: , where R n is the attenuation change value of the nth power value, P R,n is the nth power value in the attenuation section, P R,n-1 is the n-1th power value in the attenuation section, △T is P R,n With P R,n-1 The length of the time interval.

[0086] When the battery area is fully charged, the output current is stable, so the output power of the BUCK module is stable. When the battery area is insufficient, the output power of the BUCK module enters the attenuation stage. The steady-state fluctuation value of each power value is calculated in the stable stage to reflect the stability of the stable stage. The attenuation change value of each power value is calculated in the attenuation stage to reflect the attenuation decline speed.

[0087] like Figure 3As shown, the power supply fault assessment neural network in S5 includes: a steady-state fluctuation feature matrix construction layer, an attenuation change feature matrix construction layer, a first convolution layer, a second convolution layer, a third convolution layer, a fourth convolution layer, an adder A1, an adder A2, a first feature synthesis layer, a second feature synthesis layer, a first synthesis feature extraction unit, a second synthesis feature extraction unit and a classifier;

[0088] The input end of the steady-state fluctuation feature matrix construction layer is used to input the steady-state fluctuation vector; the input end of the attenuation change feature matrix construction layer is used to input the attenuation change vector; the input end of the first convolution layer is used to input the first linear relationship matrix; the input end of the fourth convolution layer is used to input the second linear relationship matrix;

[0089] The input end of the second convolutional layer is connected to the output end of the steady-state fluctuation feature matrix construction layer; the input end of the third convolutional layer is connected to the output end of the attenuation change feature matrix construction layer; the input end of the adder A1 is respectively connected to the output end of the second convolutional layer and the output end of the third convolutional layer, and its output end is respectively connected to the first input end of the first feature synthesis layer and the first input end of the second feature synthesis layer; the second input end of the first feature synthesis layer is connected to the output end of the first convolutional layer, and its output end is connected to the input end of the first synthetic feature extraction unit; the second input end of the second feature synthesis layer is connected to the output end of the fourth convolutional layer, and its output end is connected to the input end of the second synthetic feature extraction unit; the input end of the adder A2 is respectively connected to the output end of the first synthetic feature extraction unit and the output end of the second synthetic feature extraction unit, and its output end is connected to the input end of the classifier; the output end of the classifier serves as the output end of the power supply fault assessment neural network.

[0090] In this embodiment, the convolution kernel sizes of the first convolution layer, the second convolution layer, the third convolution layer, and the fourth convolution layer are .

[0091] In this embodiment, the expression of the steady-state fluctuation characteristic matrix construction layer is: , where X S is the steady-state fluctuation characteristic matrix, P conv is the point convolution operation, T is the transposition operation, and S is the steady-state fluctuation vector;

[0092] The expression of the decay change feature matrix construction layer is: , where X R is the attenuation change feature matrix, T is the transpose operation, and R is the attenuation change vector.

[0093] The present invention uses point convolution operation to perform point-by-point convolution on each element in the steady-state fluctuation vector and the attenuation change vector to obtain the characteristics of each element, forming a feature vector with a length of N, and multiplying the feature vector with the feature vector after transposition operation to form The steady-state fluctuation characteristic matrix and the attenuation change characteristic matrix.

[0094] In this embodiment, the point convolution operation is The convolutional layer of size .

[0095] The present invention adopts four convolution layers to process a steady-state fluctuation feature matrix, an attenuation change feature matrix, a first linear relationship matrix and a second linear relationship matrix respectively, adopts an adder A1 to add the outputs of the second convolution layer and the third convolution layer element by element, realizes the fusion of the output features of the BUCK module, then synthesizes the corresponding features of the BUCK module with the corresponding features of the first linear relationship matrix through the first feature synthesis layer, synthesizes the corresponding features of the BUCK module with the corresponding features of the second linear relationship matrix through the second feature synthesis layer, realizes the mutual fusion and enhancement of features, forms two-way features, then adopts two synthetic feature extraction units to perform further feature extraction, adopts the adder A2 to realize the fusion of the two-way features, and adopts a classifier to output the classification value of the power failure of the new energy vehicle.

[0096] In this embodiment, the classifier may adopt a BP neural network or a fully connected layer, etc.

[0097] In this embodiment, the expressions of the feature synthesis layer are all: , where H is the feature output by the feature synthesis layer, h1 is the input feature of the first input terminal of the feature synthesis layer, h2 is the input feature of the second input terminal of the feature synthesis layer, W1 is the weight matrix of h1, B1 is the bias matrix of h1, W2 is the weight matrix of h2, and B2 is the bias matrix of h2. is element-wise multiplication.

[0098] The data input into the four convolutional layers of the present invention is Therefore, h1 and h2 are also matrices. After assigning weights and biases to each element in h1 and h2, the output features of the BUCK module are fused with the corresponding features of the linear relationship matrix to present two-way features.

[0099] In this embodiment, the first synthetic feature extraction unit and the second synthetic feature extraction unit both include an average pooling layer and a feature enhancement layer;

[0100] The average pooling layer is used to extract the mean features of the features output by the feature synthesis layer;

[0101] The feature enhancement layer is used to adaptively enhance each feature element in the mean feature to obtain enhanced features. The expression of the feature enhancement layer is: , where G m is the mth enhanced feature value output by the first synthetic feature extraction unit or the second synthetic feature extraction unit, g mis the mth feature element in the mean feature, w m g m The weight of b m g m The bias of , m is a positive integer.

[0102] In this embodiment, the weights and biases in the power failure assessment neural network may be trained using a gradient descent method.

[0103] The present invention sets a feature enhancement layer to adaptively enhance each feature element, thereby realizing flexible and automatic adjustment of the size of the feature.

[0104] The present invention processes the current signal in stages, which is helpful to analyze the performance changes of the battery in different discharge stages, capture potential fault signals, and enhance the dynamics of fault diagnosis.

[0105] By constructing a linear relationship matrix, the present invention can deeply analyze the relationship between the battery and the power management system BMS, and the power management system BMS and the BUCK module, identify potential fault modes, and improve the comprehensiveness of fault diagnosis.

[0106] Since the output of the BUCK module is the output of the new energy vehicle power supply, extracting steady-state fluctuations and attenuation change information helps to identify performance changes of the power supply during battery discharge and further improve the sensitivity of fault detection.

[0107] The present invention uses a power fault assessment neural network to process the first linear relationship matrix, the second linear relationship matrix, the steady-state fluctuation vector and the attenuation change vector, realizes the relationship between the comprehensive battery and the power management system BMS, the relationship between the power management system BMS and the BUCK module, and then combines the characteristics of the output power signal of the BUCK module to improve the accuracy of fault detection.

[0108] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting power failure of a new energy vehicle, characterized in that: The following steps are involved: S1. Divide the battery pack in the fully charged new energy vehicle power supply into multiple battery areas, discharge each battery area, and collect the output current signal of the battery area, the output current signal of the power management system BMS, and the output current signal of the BUCK module; S2, dividing each output current signal into an early current signal, a mid-term current signal and a late current signal, and constructing an output current matrix; S3, constructing a first linear relationship matrix according to the output current matrix of the battery area and the power management system BMS, and constructing a second linear relationship matrix according to the output current matrix of the power management system BMS and the BUCK module; S4. According to the output power signal of the BUCK module, the steady-state fluctuation value and the attenuation change value are extracted, and the steady-state fluctuation vector and the attenuation change vector are constructed; S5. Use a power supply fault assessment neural network to process the first linear relationship matrix, the second linear relationship matrix, the steady-state fluctuation vector and the attenuation change vector to obtain a new energy vehicle power supply fault classification value.

2. The new energy vehicle power supply fault detection method according to claim 1 is characterized in that: The S2 comprises the following sub-steps: S21, marking the time point when the output current signal of the battery area rises to the rated current value as the first dividing point, and marking the time point when the output current signal of the battery area drops to the rated current value as the second dividing point; S22, dividing each output current signal into three sections according to the positions of the first dividing point and the second dividing point, to obtain an early current signal, a mid-term current signal and a late current signal; S23, respectively collect N current values ​​for the early current signal, the mid-term current signal and the late current signal, and construct an output current matrix: , where E is the output current matrix, X1 is a column vector consisting of N current values ​​of the early current signal, X2 is a column vector consisting of N current values ​​of the mid-term current signal, X3 is a column vector consisting of N current values ​​of the late current signal, and N is a positive integer.

3. The new energy vehicle power supply fault detection method according to claim 1 is characterized in that: The S3 comprises the following sub-steps: S31, calculating the mean of each column vector in each output current matrix to obtain a column mean; S32. Centering the output current matrix according to the column mean to obtain a current centering matrix, wherein the centering formula is: ,in, is the current value of the i-th column and j-th row of the current centering matrix, X i,j is the current value of the i-th column and j-th row of the output current matrix, X i,avg is the column mean of the i-th column, i and j are positive integers; S33, constructing a first linear relationship matrix according to the current centering matrix of the battery area and the power management system BMS; S34, constructing a second linear relationship matrix according to the current centering matrix of the power management system BMS and the BUCK module.

4. The new energy vehicle power supply fault detection method according to claim 3 is characterized in that: The formula for constructing the first linear relationship matrix in S33 is: , where C1 is the first linear relationship matrix, C battery is the current centering matrix of the battery area, C BMS is the current centering matrix of the power management system BMS, and T is the transposition operation; The formula for constructing the second linear relationship matrix in S34 is: , where C2 is the second linear relationship matrix, C BUCK This is the current centering matrix for the BUCK module.

5. The new energy vehicle power supply fault detection method according to claim 1 is characterized in that: The S4 comprises the following sub-steps: S41. In the output power signal of the BUCK module, a signal segment greater than 0.8 times the rated power is marked as a stable segment, and a power signal after the stable segment is marked as an attenuation segment; S42, collecting N power values ​​in the stable segment, calculating the steady-state fluctuation value of each power value, and constructing a steady-state fluctuation vector with a length of N, where N is a positive integer; S43. Collect N+1 power values ​​in the attenuation segment, calculate the attenuation change value of each power value, and construct an attenuation change vector with a length of N.

6. The new energy vehicle power supply fault detection method according to claim 5 is characterized in that: The formula for calculating the steady-state fluctuation value of each power value in S42 is: , where S n is the steady-state fluctuation value of the nth power value, P S,n is the nth power value in the stable segment, P S,n+1 is the n+1th power value in the stable segment, P S,n-1 is the n-1th power value in the stable segment, where n is a positive integer; The formula for calculating the attenuation change value of each power value in S43 is: , where R n is the attenuation change value of the nth power value, P R,n is the nth power value in the attenuation section, P R,n-1 is the n-1th power value in the attenuation section, △T is P R,n With P R,n-1 The length of the time interval.

7. The new energy vehicle power supply fault detection method according to claim 1, characterized in that: The power failure assessment neural network in S5 includes: a steady-state fluctuation feature matrix construction layer, an attenuation change feature matrix construction layer, a first convolution layer, a second convolution layer, a third convolution layer, a fourth convolution layer, an adder A1, an adder A2, a first feature synthesis layer, a second feature synthesis layer, a first synthesis feature extraction unit, a second synthesis feature extraction unit and a classifier; The input end of the steady-state fluctuation characteristic matrix construction layer is used to input a steady-state fluctuation vector; The input end of the attenuation change feature matrix construction layer is used to input the attenuation change vector; The input end of the first convolutional layer is used to input a first linear relationship matrix; The input end of the fourth convolutional layer is used to input the second linear relationship matrix; The input end of the second convolutional layer is connected to the output end of the steady-state fluctuation feature matrix construction layer; The input end of the third convolutional layer is connected to the output end of the attenuation change feature matrix construction layer; The input end of the adder A1 is connected to the output end of the second convolutional layer and the output end of the third convolutional layer respectively, and the output end thereof is connected to the first input end of the first feature synthesis layer and the first input end of the second feature synthesis layer respectively; The second input end of the first feature synthesis layer is connected to the output end of the first convolutional layer, and the output end thereof is connected to the input end of the first synthetic feature extraction unit; The second input end of the second feature synthesis layer is connected to the output end of the fourth convolutional layer, and the output end of the second feature synthesis layer is connected to the input end of the second synthesis feature extraction unit; The input end of the adder A2 is connected to the output end of the first synthetic feature extraction unit and the output end of the second synthetic feature extraction unit respectively, and the output end thereof is connected to the input end of the classifier; The output end of the classifier is used as the output end of the power failure assessment neural network.

8. The new energy vehicle power supply fault detection method according to claim 7 is characterized in that: The expression of the steady-state fluctuation characteristic matrix construction layer is: , where X S is the steady-state fluctuation characteristic matrix, P conv is the point convolution operation, T is the transposition operation, and S is the steady-state fluctuation vector; The expression of the attenuation change feature matrix construction layer is: , where X R is the attenuation change feature matrix, T is the transpose operation, and R is the attenuation change vector.

9. The new energy vehicle power supply fault detection method according to claim 7, characterized in that: The expressions of the feature synthesis layer are: , where H is the feature output by the feature synthesis layer, h1 is the input feature of the first input terminal of the feature synthesis layer, h2 is the input feature of the second input terminal of the feature synthesis layer, W1 is the weight matrix of h1, B1 is the bias matrix of h1, W2 is the weight matrix of h2, and B2 is the bias matrix of h2. is element-wise multiplication.

10. The new energy vehicle power supply fault detection method according to claim 7, characterized in that: The first synthetic feature extraction unit and the second synthetic feature extraction unit both include an average pooling layer and a feature enhancement layer; The average pooling layer is used to extract mean features from the features output by the feature synthesis layer; The feature enhancement layer is used to adaptively enhance each feature element in the mean feature to obtain an enhanced feature, wherein the expression of the feature enhancement layer is: , where G m is the mth enhanced feature value output by the first synthetic feature extraction unit or the second synthetic feature extraction unit, g m is the mth feature element in the mean feature, w m g m The weight of b m g m The bias of , m is a positive integer.

Images