Capacitor health detection method and system, electronic device, and storage medium

Abstract

The application relates to the technical field of artificial intelligence, in particular to a capacitor health detection method and system, an electronic device and a storage medium, the method comprising the following steps: obtaining target correlation data of a capacitor to be detected, the target correlation data comprising electric parameter data of the capacitor to be detected within a preset time period after starting charging and electric parameter data within a preset time period after starting discharging, the electric parameter data comprising current data and voltage data; generating a target heat map according to the target correlation data; inputting the target heat map into a preset capacitor health detection model to perform feature extraction and capacitor health detection, so as to obtain capacitor capacitance, capacitor aging percentage and capacitor fault state output by the capacitor health detection model; the method can effectively capture local features in the target heat map, thereby helping to improve capacitor health detection accuracy, reduce calculation and processing difficulty, reduce resource consumption, improve detection efficiency and reduce capacitor health detection cost.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, system, electronic device and storage medium for detecting capacitance health. Background Technology

[0002] With the rapid development of electronic technology, capacitors have become an indispensable part of electronic devices. Supercapacitors, for example, are widely used in electric vehicles, renewable energy storage, and mobile devices due to their characteristics such as rapid charging and discharging, long cycle life, and high energy density. However, during long-term operation, supercapacitors undergo irreversible performance degradation, leading to aging and malfunctions, and in severe cases, even capacitor failure, causing a series of safety hazards. Therefore, health monitoring of capacitors is crucial.

[0003] In related technologies, visual inspection or multimeter testing is commonly used to perform health checks on capacitors. For example, visual inspection or a magnifying glass is used to check for physical damage such as bulging or leakage, or a digital multimeter is used to measure the capacitance. However, both of these methods have drawbacks such as low detection accuracy and poor detection efficiency, making it difficult for relevant personnel to accurately control the health status of capacitors. Summary of the Invention

[0004] This application provides a capacitance health detection method, system, electronic device, and storage medium to solve the problems of low detection accuracy and poor detection efficiency in related technologies.

[0005] This application provides a capacitor health detection method, the method comprising: acquiring target correlation data of a capacitor to be tested, the target correlation data including electrical parameter data of the capacitor to be tested within a preset time period after charging begins, and electrical parameter data within a preset time period after discharging begins, the electrical parameter data including current data and voltage data; Based on the target-related data, a target heatmap is generated; The target heatmap is input into a preset capacitor health detection model for feature extraction and capacitor health detection to obtain the capacitor value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model.

[0006] In some embodiments of this application, the electrical parameter data of the capacitor under test are all positive values ​​within a preset time period after charging begins, and the electrical parameter data of the capacitor under test are all negative values ​​within a preset time period after discharging begins; a target heatmap is generated based on the target association data, including: The current data of the capacitor under test during a preset time period after the start of charging and the current data during a preset time period after the start of discharging are arranged in time sequence to obtain the first one-dimensional array. The voltage data of the capacitor under test during a preset time period after the start of charging and the voltage data during a preset time period after the start of discharging are arranged in chronological order to obtain a second one-dimensional array. The target heatmap is generated based on the first one-dimensional array and the second one-dimensional array.

[0007] In some embodiments of this application, generating the target heatmap based on the first one-dimensional array and the second one-dimensional array includes: According to a preset time window, the first one-dimensional array and the second one-dimensional array are respectively divided into multiple target windows, and the length of the target window is the same as the length of the time window. Feature extraction is performed on the data in the target window to obtain the feature data of the target window, and the target window and the feature data have a corresponding relationship; The feature data of multiple target windows are converted into a two-dimensional table structure to obtain the target table; The target heatmap is generated based on the target table.

[0008] In some embodiments of this application, the acquisition of the capacitance health detection model includes: Obtain a training set, which includes multiple heatmap samples and sample labels corresponding to the heatmap samples. The sample labels include: capacitance value label, capacitor aging percentage label, and capacitor fault status label. The heatmap sample is input into a shared network in a preset initial network architecture to obtain the first feature information output by the shared network. The shared network includes a first convolutional layer, a first normalization layer, a first activation function layer, and an average pooling layer connected in sequence. The first feature information is input into the regression branch network and the classification branch network in the initial network architecture, respectively, to obtain the capacitor value prediction result and capacitor aging percentage prediction result output by the regression branch network, and the capacitor fault state prediction result output by the classification branch network. Based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, the initial network architecture is trained, and the trained initial network architecture is determined as the capacitor health detection model.

[0009] In some embodiments of this application, the regression branch network includes: a second convolutional layer, a second normalization layer, a second activation function layer, a global average pooling layer, and a first fully connected layer; the first feature information is input into the regression branch network to obtain the capacitance value prediction result and capacitance aging percentage prediction result output by the regression branch network, including: The first feature information is input into the second convolutional layer for feature extraction to obtain the second feature information, wherein the number of output channels of the second convolutional layer is greater than the number of output channels of the first convolutional layer; the second feature information is input into the second normalization layer for normalization processing to obtain the third feature information; The third feature information is input into the second activation function layer for activation processing to obtain the fourth feature information, the value of which is in the range of 0 to positive infinity; the fourth feature information is input into the global average pooling layer for global average pooling to obtain the fifth feature information. The fifth feature information is input into the first fully connected layer to perform capacitance value prediction and capacitance aging percentage prediction, so as to obtain the capacitance value prediction result and the capacitance aging percentage prediction result.

[0010] In some embodiments of this application, the classification branch network includes: a third convolutional layer, a third normalization layer, a third activation function layer, a max pooling layer, a second fully connected layer, and a Softmax activation function layer; inputting the first feature information into the classification branch network to obtain the capacitor fault state prediction result output by the classification branch network includes: The first feature information is input into the third convolutional layer for feature extraction to obtain the sixth feature information, wherein the number of output channels of the third convolutional layer is greater than the number of output channels of the first convolutional layer; the sixth feature information is input into the third normalization layer for normalization processing to obtain the seventh feature information. The seventh feature information is input into the third activation function layer for activation processing to obtain the eighth feature information, the value of which is in the range of 0 to positive infinity; the eighth feature information is input into the max pooling layer for max pooling processing to obtain the ninth feature information; The ninth feature information is input into the second fully connected layer to predict the capacitor fault state and obtain intermediate prediction results; the intermediate prediction results are input into the Softmax activation function layer for activation processing to obtain the capacitor fault state prediction results, which include the prediction probability of each preset fault category.

[0011] In some embodiments of this application, the initial network architecture is trained based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, including: Based on the capacitor value prediction result, the capacitor value label, and the preset first mean square error loss function, a first loss is obtained. The first mean square error loss function is used to characterize the difference between the capacitor value prediction result and the corresponding capacitor value label. Based on the capacitor aging percentage prediction result, the capacitor aging percentage label, and the preset second mean square error loss function, a second loss is obtained. The second mean square error loss function is used to characterize the difference between the capacitor aging percentage prediction result and the corresponding capacitor aging percentage label. Based on the capacitor fault state prediction result, the capacitor fault state label, and the preset cross-entropy loss function, a third loss is obtained. The cross-entropy loss function is used to characterize the difference between the capacitor fault state prediction result and the corresponding capacitor fault state label. Based on the first loss, the second loss, the third loss, and preset weight coefficients that correspond one-to-one with the first loss, the second loss, and the third loss, the final loss is obtained; The initial network architecture is iterated based on the final loss to complete the training.

[0012] This application also provides a capacitance health detection system, the system comprising: The target association data acquisition module is used to acquire the target association data of the capacitor to be tested. The target association data includes the electrical parameter data of the capacitor to be tested within a preset time period after the start of charging, and the electrical parameter data within a preset time period after the start of discharging. The electrical parameter data includes current data and voltage data. A heatmap generation module is used to generate a target heatmap based on the target-related data; The capacitor health detection module is used to input the target heat map into a preset capacitor health detection model, perform feature extraction and capacitor health detection, so as to obtain the capacitance value, capacitor aging percentage and capacitor fault status output by the capacitor health detection model.

[0013] This application also provides an electronic device, including a processor, a memory, and a communication bus; the communication bus is used to connect the processor and the memory; the processor is used to execute a computer program stored in the memory to implement the capacitance health detection method provided in any of the above embodiments.

[0014] This application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being used to cause a computer to perform the capacitance health detection method as provided in any of the above embodiments.

[0015] The beneficial effects of the embodiments of this application are as follows: The capacitor health detection method, system, electronic device, and storage medium provided in the embodiments of this application acquire target correlation data of the capacitor to be tested. The target correlation data includes electrical parameter data of the capacitor to be tested within a preset time period after the start of charging and electrical parameter data within a preset time period after the start of discharging. The electrical parameter data includes current data and voltage data. Based on the target correlation data, a target heatmap is generated. The target heatmap is input into a preset capacitor health detection model for feature extraction and capacitor health detection to obtain the capacitance value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model. This method obtains a target heatmap based on one-dimensional target correlation data. By extracting features from the target heatmap, local features in the target heatmap can be effectively captured, thereby helping to improve the accuracy of capacitor health detection. Furthermore, compared with the method of directly performing capacitor health detection based on target correlation data, the above method generates a target heatmap based on the target correlation data and performs subsequent detection based on the target heatmap, which can reduce the computational and processing difficulty, effectively reduce resource consumption, improve detection efficiency, and reduce the cost of capacitor health detection. Attached Figure Description

[0016] Figure 1 A schematic flowchart of a capacitance health detection method provided in an embodiment of this application; Figure 2 This is an exemplary schematic diagram of the target heat map in a capacitance health detection method provided in one embodiment of this application; Figure 3 This is an exemplary schematic diagram illustrating the principle of convolution operation in a capacitance health detection method provided in an embodiment of this application; Figure 4 This is a schematic diagram of the process of capacitor health detection based on target heat map in a capacitor health detection method provided in an embodiment of this application; Figure 5 A flowchart illustrating a specific embodiment of the capacitance health detection method provided in this application; Figure 6 This is a schematic diagram of the structure of a capacitance health detection system provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0018] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0019] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the present application. However, it will be apparent to those skilled in the art that embodiments of the present application may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the present application.

[0020] The following is combined with Figures 1 to 7 This application provides an explanation of the capacitor health testing method, system, electronic device, and storage medium provided in this application.

[0021] Please see Figure 1 , Figure 1 This is a schematic flowchart of a capacitance health detection method provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes: S110: Obtain target correlation data of the capacitor to be tested. The target correlation data includes electrical parameter data of the capacitor to be tested within a preset time period after the start of charging, and electrical parameter data within a preset time period after the start of discharging. The electrical parameter data includes current data and voltage data.

[0022] In some examples of this embodiment, the capacitor to be detected can be a supercapacitor or a traditional capacitor. The preset time period can be set according to the actual situation, such as 50s, 60s, etc., which will not be elaborated here.

[0023] In some examples of this embodiment, an ADC (Analog-to-Digital Converter) can be used to collect and acquire the aforementioned target-related data.

[0024] S120: Generate a target heatmap based on the target association data.

[0025] In some examples of this embodiment, the target associated data can be directly converted into a heatmap format to obtain the corresponding target heatmap.

[0026] Understandably, by generating a target heatmap based on target-related data, it is possible to convert one-dimensional data into an image, which helps to reduce the computational difficulty and processing complexity of subsequent capacitance health detection.

[0027] S130: Input the target heat map into a preset capacitor health detection model, perform feature extraction and capacitor health detection, and obtain the capacitor value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model.

[0028] As we understand it, capacitance value refers to the ability of a capacitor under test to store electrical energy, usually measured in farads (F), microfarads (μF), and picofarads (pF). The larger the capacitance value, the more charge the capacitor can store, reflecting its "capacitance." Capacitor aging percentage indicates the percentage of aging of the capacitor under test, which is related to its current and initial capacitance values. Capacitor fault status indicates whether the capacitor under test has a fault, and if so, the type of fault. Fault types include open circuit faults, short circuit faults, and failure faults.

[0029] It is also understandable that, since the target heatmap is generated based on target-related data, using the capacitance health detection model to extract features from the target heatmap is equivalent to using the capacitance health detection model to extract spatial features from the target-related data, thereby effectively capturing local features in the target-related data. Compared to directly performing capacitance health detection based on target-related data, the method in the above embodiment can significantly improve the accuracy of capacitance health detection and effectively reduce the resource consumption of capacitance health detection, resulting in lower costs.

[0030] In some embodiments, the electrical parameter data of the capacitor under test are all positive values ​​within a preset time period after charging begins, and the electrical parameter data of the capacitor under test are all negative values ​​within a preset time period after discharging begins.

[0031] Understandably, the above settings make it easier to distinguish the electrical parameter data of the capacitor under test during charging from that during discharging, which helps to improve the accuracy of subsequent capacitor health testing.

[0032] In some embodiments, generating a target heatmap based on the target association data includes: First, the current data of the capacitor to be tested during a preset time period after the start of charging and the current data during a preset time period after the start of discharging are arranged in chronological order to obtain a first one-dimensional array.

[0033] In some examples of this embodiment, the mathematical expression of the current data of the capacitor under test within a preset time period after the start of charging can be:

[0034] in, This represents the current data of the capacitor under test within a preset time period after charging begins. This represents n current sampling values ​​of the capacitor under test within a preset time period after charging begins, where n represents the number of samples.

[0035] The mathematical expression for the current data of the capacitor under test within a preset time period after the start of discharge can be:

[0036] in, This represents the current data of the capacitor under test within a preset time period after it begins to discharge. This represents n current sampling values ​​within a preset time period after the capacitor under test begins to discharge.

[0037] By and By merging, we can obtain:

[0038] in, This represents the set of current data of the capacitor under test during a preset time period after it starts charging, and the set of current data of the capacitor under test during a preset time period after it starts discharging. This set is the first one-dimensional array.

[0039] Second, the voltage data of the capacitor to be tested during a preset time period after the start of charging and the voltage data during a preset time period after the start of discharging are arranged in chronological order to obtain a second one-dimensional array.

[0040] In some examples of this embodiment, the mathematical expression of the voltage data of the capacitor under test within a preset time period after the start of charging can be:

[0041] in, This represents the voltage data of the capacitor under test within a preset time period after charging begins. This represents n voltage sampling values ​​of the capacitor under test within a preset time period after charging begins.

[0042] The mathematical expression for the voltage data of the capacitor under test within a preset time period after the start of discharge can be:

[0043] in, This represents the voltage data of the capacitor under test within a preset time period after it begins to discharge. This represents n voltage sampling values ​​of the capacitor under test within a preset time period after it begins to discharge.

[0044] By and By merging, we can obtain:

[0045] in, This represents the set of voltage data of the capacitor under test during a preset time period after it starts charging, and the set of voltage data of the capacitor under test during a preset time period after it starts discharging. This set is the second one-dimensional array.

[0046] Understandably, the above operations facilitate the subsequent generation of the target heat map and reduce the difficulty of generating the target heat map.

[0047] 3. Generate the target heatmap based on the first one-dimensional array and the second one-dimensional array.

[0048] In some examples of this embodiment, the target heatmap can be generated directly based on the first one-dimensional array and the second one-dimensional array. For example, the first one-dimensional array can be placed on top of the second one-dimensional array to form a two-dimensional array, and the corresponding target heatmap can be generated based on the two-dimensional array.

[0049] Besides the methods described in the above embodiments of generating target heatmaps directly from target-related data and directly from the first and second one-dimensional arrays, in some embodiments, to further reduce the difficulty of generating target heatmaps and accelerate the efficiency of subsequent capacitance health detection, feature extraction can be performed on the first and second one-dimensional arrays, and the target heatmap can be generated based on the extracted features. Specifically, generating the target heatmap based on the first and second one-dimensional arrays includes: 1. According to a preset time window, the first one-dimensional array and the second one-dimensional array are divided into multiple target windows, and the length of the target window is the same as the length of the time window.

[0050] In some examples of this embodiment, the length of the time window can be set according to actual needs, such as 5 or 6. The length of the time window represents the number of sampled values ​​it contains.

[0051] In some examples of this embodiment, a first one-dimensional array is used. For example, the mathematical expression of the target window obtained after the above window division can be:

[0052] Where m represents the length of the time window, which is also the length of the target window, and also represents the number of sampled values ​​within the target window. The first one-dimensional array is divided into... One target window, It is an integer.

[0053] Understandably, dividing the window facilitates subsequent feature extraction based on the target window.

[0054] Second, feature extraction is performed on the data in the target window to obtain the feature data of the target window, and the target window and the feature data have a corresponding relationship.

[0055] In some examples of this embodiment, feature extraction is performed on each target window to obtain feature data for each target window.

[0056] In some embodiments, the feature data includes: the maximum value, minimum value, average value, standard deviation, and rate of change of the data within the target window.

[0057] In some examples of this embodiment, taking the target window in the first one-dimensional array as an example, the mathematical expression for obtaining the maximum value of the data within the target window can be:

[0058] in, This represents the maximum value of the data within the target window. This indicates finding the maximum value.

[0059] The mathematical expression for obtaining the minimum value of data within the target window can be:

[0060] in, This represents the minimum value of the data within the target window. This indicates that the minimum value is being sought.

[0061] The mathematical expression for obtaining the average value of the data within the target window can be:

[0062] in, This represents the average value of the data within the target window. This represents the i-th sampled value in the target window.

[0063] The mathematical expression for obtaining the standard deviation of the data within the target window can be:

[0064] in, This represents the standard deviation of the data within the target window. This represents the j-th sampled value in the target window.

[0065] The mathematical expression for the rate of change of data within the target window can be:

[0066] in, This indicates the rate of change of the data within the target window. This represents the k-th sampled value in the target window, where k < m. The (k+1)th sampled value in the target window, " indicates cross product.

[0067] Third, the feature data of the multiple target windows are converted into a two-dimensional table structure to obtain the target table.

[0068] The target table is shown as an example in Table 1 below: Table 1 Example of a target table

[0069] IV. Generate the target heatmap based on the target table.

[0070] Understandably, the above embodiments, through methods such as window segmentation and feature extraction, significantly reduce the amount of data required to generate the target heatmap, thereby effectively improving the generation efficiency of the target heatmap and reducing the resource consumption caused by generating the target heatmap. Furthermore, it helps to improve the efficiency of subsequent capacitor health detection.

[0071] Figure 2 An example heatmap of the target is shown below; please refer to it. Figure 2 , Figure 2 The vertical axis represents different categories of feature data, such as maximum current, minimum current, average current, standard deviation of current, maximum voltage, minimum voltage, average voltage, standard deviation of voltage, rate of change of current, and rate of change of voltage. The horizontal axis represents different target windows, such as target window 1, target window 2, ..., target window 10. Figure 2 The color bar on the right is a legend for the target heatmap, used to show the correspondence between values ​​and colors in the target heatmap.

[0072] In some embodiments, obtaining the capacitance health detection model includes: 1. Obtain a training set, which includes multiple heatmap samples and sample labels corresponding to the heatmap samples. The sample labels include: capacitance value label, capacitor aging percentage label, and capacitor fault status label.

[0073] In some examples of this embodiment, target associated data samples, capacitance values, and capacitor fault states of capacitors in different states (such as normal capacitors, aging capacitors, and faulty capacitors) can be measured first. Then, heatmap samples are generated based on the target associated data samples (refer to the method of generating target heatmaps based on target associated data in the above embodiments). Finally, sample labels are added to each heatmap sample. The capacitor aging percentage label can be obtained as follows:

[0074] in, Label indicating the percentage of capacitor aging This indicates the initial capacitance value of the capacitor (which can be a preset value). This indicates the current capacitance value of the capacitor. This indicates the lifespan of the capacitor under normal operating conditions. This indicates the duration the capacitor has been operating. Capacitor fault status labels include open circuit fault, short circuit fault, and failure fault, which can be represented by a three-digit binary integer. Each digit in the three-digit number represents a different fault category, and the value of each digit indicates whether the corresponding fault category exists. Specifically, for any digit in the three-digit number, if the corresponding fault category exists, the value is 1; otherwise, the value is 0. The remaining two digits are valued in the same way. For example, if the current capacitor has an open circuit fault but no short circuit or failure fault, its capacitor fault status label can be 100. If the current capacitor does not have an open circuit fault but has both a short circuit and failure fault, its corresponding capacitor fault status label is 011.

[0075] Understandably, the above method can be used to obtain a dataset containing multiple heatmap samples and corresponding sample labels. In some embodiments, 70% of the data in the dataset can be used as the training set, 15% as the validation set, and 15% as the test set.

[0076] 2. Input the heatmap sample into a shared network in a preset initial network architecture to obtain the first feature information output by the shared network. The shared network includes a first convolutional layer, a first normalization layer, a first activation function layer, and an average pooling layer connected in sequence.

[0077] In some examples of this embodiment, assuming the heatmap sample of the input shared network is I, and the size of the convolution kernel K of the first convolutional layer is f×f, then the mathematical expression of the convolution operation of the first convolutional layer can be:

[0078] in, This represents the feature map output by the first convolutional layer at position (p, q) and the value at channel d, where d represents the channel number of the first convolutional layer, and s represents the stride, which can be 1. u and v represent the indices within the convolution kernel, which can also be understood as the number of rows and columns of the convolution kernel during convolution, or as the sliding offset used to traverse all elements during sliding convolution. This represents the bias term for the d-th channel. " indicates dot product.

[0079] Figure 3 The principle of convolution operation is illustrated by an example, such as... Figure 3 As shown, assuming the size of the convolution kernel K is 3*3, then Figure 3 Currently, the value of the convolution kernel in row 0, column 0 is being calculated with the 3*3 region of the input image (such as a heatmap sample). At this time, u=0, v=0. Figure 3 The 2x2 graph at the top represents the output feature map, and the 4x4 graph at the bottom represents the input image. Each square in the 2x2 graph is the result of the convolution kernel sliding across the input image.

[0080] After the convolution operation, the feature map output from the first convolutional layer is fed into the first normalization layer for batch normalization (BN) processing. Assuming the dimension of the feature map output from the first convolutional layer becomes h, the input to the first normalization layer is B = ... The first normalization layer performs batch normalization on its input, including the following steps: First, calculate the mean and variance of each feature dimension in this batch:

[0081]

[0082] in, This represents the mean. Represents variance. This represents the i-th input.

[0083] Next, each input is normalized to have a mean of 0 and a variance of 1. The mathematical expression for normalization can be:

[0084] in, This represents the i-th eigenvalue after normalization. To protect constants and avoid denominators of 0, multiple eigenvalue distributions constitute a feature map.

[0085] Then, the normalized feature map is input into the first activation function layer for activation processing. In some embodiments, the first activation function layer may employ the ReLU (Rectified Linear Unit) activation function.

[0086] Finally, the output data of the first activation function layer is input into the average pooling layer for average pooling processing to obtain the first feature information.

[0087] Understandably, since the subsequent branch tasks include both regression tasks (capacitor value prediction and capacitor aging percentage prediction tasks corresponding to the regression branch network) and classification tasks (capacitor fault state prediction tasks corresponding to the classification branch network), the use of average pooling can better preserve global information and retain the spatial hierarchy structure, which helps to improve the processing accuracy of subsequent branch tasks.

[0088] In some examples of this embodiment, for the input feature map X of the average pooling layer, assuming its pooling window size is a×a and its stride is s, then the mathematical expression of the first feature information output by the average pooling layer can be:

[0089] in, The first feature information is represented by z and r, which represent the offsets of the pooling window when it slides across the feature map for calculation, and g and l represent the position indices of the pooling window.

[0090] Third, input the first feature information into the regression branch network and the classification branch network in the initial network architecture respectively to obtain the capacitor value prediction result and capacitor aging percentage prediction result output by the regression branch network, and the capacitor fault state prediction result output by the classification branch network.

[0091] In some examples of this embodiment, the regression branch network is used to handle the capacitor value prediction task and the capacitor aging percentage prediction task, and its output capacitor value prediction result and capacitor aging percentage prediction result are specific numerical values. The classification branch network is used to handle the capacitor fault state prediction task, and its output capacitor fault state prediction result is a probability distribution of all preset fault categories.

[0092] Understandably, by setting up two branch networks, namely the regression branch network and the classification branch network, the processing needs of the above-mentioned capacitor health detection process, such as capacitor value prediction, capacitor aging percentage prediction, and capacitor fault state prediction, can be well met, making it easier for relevant personnel to grasp the health status of the capacitor from multiple dimensions.

[0093] Fourth, based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, the initial network architecture is trained, and the trained initial network architecture is determined as the capacitor health detection model.

[0094] Understandably, the above method can yield a highly accurate capacitance health detection model.

[0095] In some embodiments, the regression branch network includes: a second convolutional layer, a second normalization layer, a second activation function layer, a global average pooling layer, and a first fully connected layer.

[0096] In some embodiments, the first feature information is input into a regression branch network to obtain the capacitance value prediction result and capacitance aging percentage prediction result output by the regression branch network, including: First, the first feature information is input into the second convolutional layer for feature extraction to obtain the second feature information, wherein the number of output channels of the second convolutional layer is greater than the number of output channels of the first convolutional layer; the second feature information is input into the second normalization layer for normalization processing to obtain the third feature information.

[0097] In some examples of this embodiment, the number of output channels of the second convolutional layer can be an integer multiple of the number of output channels of the first convolutional layer, such as 64 output channels for the second convolutional layer and 32 output channels for the first convolutional layer. By increasing the number of output channels, richer features can be captured, thereby helping to improve the accuracy of subsequent predictions.

[0098] Second, the third feature information is input into the second activation function layer for activation processing to obtain the fourth feature information, the value of which is in the range of 0 to positive infinity; the fourth feature information is input into the global average pooling layer for global average pooling to obtain the fifth feature information.

[0099] In some examples of this embodiment, the second activation function layer may employ the ReLU activation function.

[0100] In some examples of this embodiment, the mathematical expression for global average pooling can be:

[0101] in, This represents the output of the global average pooling layer. , This represents the height and width of the feature map corresponding to the fourth feature information input to the global average pooling layer. Indicates the position in the feature map of the d-th channel. The value at that location.

[0102] Third, input the fifth feature information into the first fully connected layer to perform capacitance value prediction and capacitance aging percentage prediction, so as to obtain the capacitance value prediction result and the capacitance aging percentage prediction result.

[0103] In some embodiments, the regression branch network further includes a linear activation function layer located after the first fully connected layer to linearly activate the output of the first fully connected layer, thereby obtaining the capacitor value prediction result and the capacitor aging percentage prediction result.

[0104] Understandably, after global average pooling, the feature values ​​of each channel are mapped to two regression values, capacitance value and capacitance aging percentage, through the first fully connected layer and the linear activation function layer, thereby obtaining the capacitance value prediction result and the capacitance aging percentage prediction result.

[0105] It is understandable that the above regression branch network can obtain highly accurate prediction results for capacitor value and capacitor aging percentage.

[0106] In some embodiments, the classification branch network includes: a third convolutional layer, a third normalization layer, a third activation function layer, a max pooling layer, a second fully connected layer, and a Softmax activation function layer.

[0107] In some embodiments, the first feature information is input into a classification branch network to obtain the capacitor fault state prediction result output by the classification branch network, including: First, the first feature information is input into the third convolutional layer for feature extraction to obtain the sixth feature information, wherein the number of output channels of the third convolutional layer is greater than the number of output channels of the first convolutional layer; the sixth feature information is input into the third normalization layer for normalization processing to obtain the seventh feature information.

[0108] Second, the seventh feature information is input into the third activation function layer for activation processing to obtain the eighth feature information, the value of which is in the range of 0 to positive infinity; the eighth feature information is input into the max pooling layer for max pooling processing to obtain the ninth feature information.

[0109] In some examples of this embodiment, assuming the pooling window size for max pooling is b*b and the stride is s, the mathematical expression for max pooling is:

[0110] in, This represents the ninth feature information.

[0111] Third, the ninth feature information is input into the second fully connected layer to predict the capacitor fault state and obtain intermediate prediction results; the intermediate prediction results are input into the Softmax activation function layer for activation processing to obtain the capacitor fault state prediction results, which include the prediction probability of each preset fault category, i.e., the probability distribution.

[0112] In some examples of this embodiment, the mathematical expression of this probability distribution is as follows:

[0113] in, This represents the predicted probability of the A-th fault category. This represents the raw score for the A-th fault category, i.e., the output value of the second fully connected layer, where R represents the number of fault categories. Represents the natural constant.

[0114] Understandably, the classification branch network in the above embodiments can obtain capacitor fault state prediction results with high accuracy.

[0115] In some embodiments, the initial network architecture is trained based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, including: First, based on the predicted capacitance value, the capacitance value label, and a preset first mean square error loss function, a first loss is obtained. The first mean square error loss function is used to characterize the difference between the predicted capacitance value and the corresponding capacitance value label.

[0116] In some examples of this embodiment, the mathematical expression of the first mean squared error loss function is:

[0117] in, Indicates the first loss. This label represents the capacitance value of the i-th heatmap sample. This represents the predicted capacitance value for i heatmap samples. This indicates the number of samples in the heatmap.

[0118] Second, based on the predicted capacitor aging percentage, the capacitor aging percentage label, and the preset second mean square error loss function, a second loss is obtained. The second mean square error loss function is used to characterize the difference between the predicted capacitor aging percentage and the corresponding capacitor aging percentage label.

[0119] In some examples of this embodiment, the mathematical expression of the second mean squared error loss function is:

[0120] in, Indicates the second loss. This label represents the percentage of capacitance aging for the i-th heatmap sample. This represents the predicted percentage of capacitor aging for i heatmap samples.

[0121] Third, based on the capacitor fault state prediction result, the capacitor fault state label, and the preset cross-entropy loss function, a third loss is obtained. The cross-entropy loss function is used to characterize the difference between the capacitor fault state prediction result and the corresponding capacitor fault state label.

[0122] In some examples of this embodiment, the mathematical expression of the cross-entropy loss function can be:

[0123] in, Indicates the third loss. This represents logarithmic function operations.

[0124] Fourth, based on the first loss, the second loss, the third loss, and preset weight coefficients that correspond one-to-one with the first loss, the second loss, and the third loss, the final loss is obtained.

[0125] In some examples of this embodiment, the mathematical expression of the final loss can be:

[0126] in, Indicates the final loss. , , These are preset weighting coefficients, which can be dynamically adjusted according to actual conditions.

[0127] In some examples of this embodiment, in addition to iteratively training the entire capacitance health detection model based on the final loss, the regression branch network can also be iteratively trained based on the first loss and the second loss, and the classification branch network can be trained based on the third loss, so as to further improve the accuracy of the regression branch network and the classification branch network.

[0128] 5. Based on the final loss, iterate the initial network architecture to complete the training.

[0129] Figure 4 This is a schematic flowchart illustrating the capacitance health detection based on a target heatmap in a capacitance health detection method provided in an embodiment of this application. Please refer to [link / reference]. Figure 4 First, the target heatmap is input into a shared network, which consists of a first convolutional layer (with a kernel size of 3*3 and 32 output channels, etc.), a first normalization layer (used for batch normalization (BN) processing), a first activation function layer (such as ReLU activation function), and an average pooling layer, all connected in sequence. Second, the first feature information output by the shared network is input into a regression branch network and a classification branch network, respectively. The regression branch network consists of a second convolutional layer (with a kernel size of 3*3 and 64 output channels, etc.), a second normalization layer (used for batch normalization (BN) processing), a second activation function layer (such as ReLU activation function), a global average pooling layer, a first fully connected layer, and a linear activation function layer, all connected in sequence. The classification branch network consists of a third convolutional layer (with a kernel size of 3*3 and 64 output channels, etc.), a third normalization layer (used for batch normalization (BN) processing), a third activation function layer (such as ReLU activation function), a max pooling layer, a second fully connected layer, and a Softmax activation function layer, connected in sequence. Subsequently, the regression branch network outputs the capacitor value prediction and capacitor aging percentage prediction, while the classification branch network outputs the capacitor fault state prediction.

[0130] Furthermore, during the model training process described above, a first loss for the capacitor value prediction result can be obtained using a preset first mean squared error loss function; a second loss for the capacitor aging percentage prediction result can be obtained using a preset second mean squared error loss function; and a third loss for the capacitor fault state prediction result can be obtained using a preset cross-entropy loss function. Then, based on these first, second, and third losses, and preset weight coefficients, the model is trained to improve its accuracy.

[0131] The capacitance health detection method in the above embodiments will be explained and illustrated below with a specific example. Please refer to [the example provided]. Figure 5 : First, target correlation data of the capacitor to be tested is obtained. The target correlation data includes electrical parameter data of the capacitor to be tested within a preset time period after the start of charging, and electrical parameter data within a preset time period after the start of discharging. The electrical parameter data includes current data and voltage data.

[0132] Secondly, the charging and discharging data are merged. Specifically, the current data of the capacitor under test during a preset time period after the start of charging and the current data during a preset time period after the start of discharging are arranged in chronological order to obtain a first one-dimensional array; the voltage data of the capacitor under test during a preset time period after the start of charging and the voltage data during a preset time period after the start of discharging are arranged in chronological order to obtain a second one-dimensional array.

[0133] Next, the target windows are divided. Specifically, according to the preset time window, the first one-dimensional array and the second one-dimensional array are divided into multiple target windows, and the length of the target window is the same as the length of the time window.

[0134] Next, feature extraction is performed on the data within the time window.

[0135] Next, the feature data of multiple target windows are converted into a two-dimensional table structure to obtain the target table.

[0136] Furthermore, a target heatmap is generated based on the target table.

[0137] Finally, the target heatmap is input into the preset capacitor health detection model for feature extraction and capacitor health detection to obtain the capacitor value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model.

[0138] The capacitor health detection system provided in this application is described below. The capacitor health detection system described below can be referred to in correspondence with the capacitor health detection method described above.

[0139] Please refer to Figure 6 The capacitance health detection system provided in this embodiment includes: The target association data acquisition module 610 is used to acquire the target association data of the capacitor to be tested. The target association data includes the electrical parameter data of the capacitor to be tested within a preset time period after the start of charging, and the electrical parameter data within a preset time period after the start of discharging. The electrical parameter data includes current data and voltage data. The heat map generation module 620 is used to generate a target heat map based on the target-related data; The capacitor health detection module 630 is used to input the target heatmap into a preset capacitor health detection model, perform feature extraction and capacitor health detection, and obtain the capacitance value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model. The target associated data acquisition module 610, the heatmap generation module 620, and the capacitor health detection module 630 are connected. The capacitor health detection system in this embodiment can achieve the technical effects achieved by the capacitor health detection method in the above embodiments, which will not be elaborated here.

[0140] It should be noted that the capacitance health detection method and capacitance health detection system provided in the above embodiments belong to the same concept. The specific operation methods of each module have been described in detail in the method embodiments and will not be repeated here. In practical applications, the capacitance health detection system provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0141] In some embodiments, the heat map generation module 620 is specifically used to arrange the current data of the capacitor to be tested within a preset time period after the start of charging and the current data within a preset time period after the start of discharging in chronological order to obtain a first one-dimensional array. The voltage data of the capacitor under test during a preset time period after the start of charging and the voltage data during a preset time period after the start of discharging are arranged in chronological order to obtain a second one-dimensional array. The target heatmap is generated based on the first one-dimensional array and the second one-dimensional array.

[0142] In some embodiments, the heatmap generation module 620 is further configured to divide the first one-dimensional array and the second one-dimensional array into multiple target windows according to a preset time window, wherein the length of the target window is the same as the length of the time window; Feature extraction is performed on the data in the target window to obtain the feature data of the target window, and the target window and the feature data have a corresponding relationship; The feature data of multiple target windows are converted into a two-dimensional table structure to obtain the target table; The target heatmap is generated based on the target table.

[0143] In some embodiments, the system further includes: a model training module for acquiring a training set, the training set including multiple heat map samples and sample labels corresponding to the heat map samples, the sample labels including: capacitance value label, capacitor aging percentage label, and capacitor fault status label; The heatmap sample is input into a shared network in a preset initial network architecture to obtain the first feature information output by the shared network. The shared network includes a first convolutional layer, a first normalization layer, a first activation function layer, and an average pooling layer connected in sequence. The first feature information is input into the regression branch network and the classification branch network in the initial network architecture, respectively, to obtain the capacitor value prediction result and capacitor aging percentage prediction result output by the regression branch network, and the capacitor fault state prediction result output by the classification branch network. Based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, the initial network architecture is trained, and the trained initial network architecture is determined as the capacitor health detection model.

[0144] In some embodiments, the model training module is specifically used to input the first feature information into the second convolutional layer for feature extraction to obtain the second feature information, wherein the number of output channels of the second convolutional layer is greater than the number of output channels of the first convolutional layer; and to input the second feature information into the second normalization layer for normalization processing to obtain the third feature information. The third feature information is input into the second activation function layer for activation processing to obtain the fourth feature information, the value of which is in the range of 0 to positive infinity; the fourth feature information is input into the global average pooling layer for global average pooling to obtain the fifth feature information. The fifth feature information is input into the first fully connected layer to perform capacitance value prediction and capacitance aging percentage prediction, so as to obtain the capacitance value prediction result and the capacitance aging percentage prediction result.

[0145] In some embodiments, the model training module is specifically used to input the first feature information into the third convolutional layer for feature extraction to obtain the sixth feature information, wherein the number of output channels of the third convolutional layer is greater than the number of output channels of the first convolutional layer; and to input the sixth feature information into the third normalization layer for normalization processing to obtain the seventh feature information. The seventh feature information is input into the third activation function layer for activation processing to obtain the eighth feature information, the value of which is in the range of 0 to positive infinity; the eighth feature information is input into the max pooling layer for max pooling processing to obtain the ninth feature information; The ninth feature information is input into the second fully connected layer to predict the capacitor fault state and obtain intermediate prediction results; the intermediate prediction results are input into the Softmax activation function layer for activation processing to obtain the capacitor fault state prediction results, which include the prediction probability of each preset fault category.

[0146] In some embodiments, the model training module is specifically used to obtain a first loss based on the capacitor value prediction result, the capacitor value label, and a preset first mean square error loss function, wherein the first mean square error loss function is used to characterize the difference between the capacitor value prediction result and the corresponding capacitor value label. Based on the capacitor aging percentage prediction result, the capacitor aging percentage label, and the preset second mean square error loss function, a second loss is obtained. The second mean square error loss function is used to characterize the difference between the capacitor aging percentage prediction result and the corresponding capacitor aging percentage label. Based on the capacitor fault state prediction result, the capacitor fault state label, and the preset cross-entropy loss function, a third loss is obtained. The cross-entropy loss function is used to characterize the difference between the capacitor fault state prediction result and the corresponding capacitor fault state label. Based on the first loss, the second loss, the third loss, and preset weight coefficients that correspond one-to-one with the first loss, the second loss, and the third loss, the final loss is obtained; The initial network architecture is iterated based on the final loss to complete the training.

[0147] In some embodiments, an electronic device is also provided, which may be a server, and its internal structure diagram is shown below. Figure 7 As shown, the electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of the server-side method described above.

[0148] In some embodiments, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps: acquiring target correlation data of a capacitor to be tested, the target correlation data including electrical parameter data of the capacitor to be tested within a preset time period after charging begins, and electrical parameter data within a preset time period after discharging begins, the electrical parameter data including current data and voltage data; generating a target heatmap based on the target correlation data; inputting the target heatmap into a preset capacitor health detection model, performing feature extraction and capacitor health detection, to obtain the capacitance value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model.

[0149] In some embodiments, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following steps: acquiring target correlation data of a capacitor to be tested, the target correlation data including electrical parameter data of the capacitor to be tested within a preset time period after the start of charging, and electrical parameter data within a preset time period after the start of discharging, the electrical parameter data including current data and voltage data; generating a target heatmap based on the target correlation data; inputting the target heatmap into a preset capacitor health detection model, performing feature extraction and capacitor health detection, to obtain the capacitance value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model.

[0150] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or electronic device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0151] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0152] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for detecting capacitance health, characterized in that, include: Obtain target correlation data of the capacitor to be tested. The target correlation data includes electrical parameter data of the capacitor to be tested within a preset time period after the start of charging, and electrical parameter data within a preset time period after the start of discharging. The electrical parameter data includes current data and voltage data. Based on the target-related data, a target heatmap is generated; The target heatmap is input into a preset capacitor health detection model for feature extraction and capacitor health detection to obtain the capacitor value, capacitor aging percentage, and capacitor fault status output by the capacitor health detection model. The acquisition of the capacitance health detection model includes: Obtain a training set, which includes multiple heatmap samples and sample labels corresponding to the heatmap samples. The sample labels include: capacitance value label, capacitor aging percentage label, and capacitor fault status label. The heatmap sample is input into a shared network in a preset initial network architecture to obtain the first feature information output by the shared network. The shared network includes a first convolutional layer, a first normalization layer, a first activation function layer, and an average pooling layer connected in sequence. The first feature information is input into the regression branch network and the classification branch network in the initial network architecture, respectively, to obtain the capacitor value prediction result and capacitor aging percentage prediction result output by the regression branch network, and the capacitor fault state prediction result output by the classification branch network. Based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, the initial network architecture is trained, and the trained initial network architecture is determined as the capacitor health detection model.

2. The capacitance health detection method according to claim 1, characterized in that, The electrical parameter data of the capacitor under test are all positive values ​​within a preset time period after it starts charging, and the electrical parameter data of the capacitor under test are all negative values ​​within a preset time period after it starts discharging. Based on the target-related data, a target heatmap is generated, including: The current data of the capacitor under test during a preset time period after the start of charging and the current data during a preset time period after the start of discharging are arranged in time sequence to obtain the first one-dimensional array. The voltage data of the capacitor under test during a preset time period after the start of charging and the voltage data during a preset time period after the start of discharging are arranged in chronological order to obtain a second one-dimensional array. The target heatmap is generated based on the first one-dimensional array and the second one-dimensional array.

3. The capacitance health detection method according to claim 2, characterized in that, Generating the target heatmap based on the first one-dimensional array and the second one-dimensional array includes: According to a preset time window, the first one-dimensional array and the second one-dimensional array are respectively divided into multiple target windows, and the length of the target window is the same as the length of the time window. Feature extraction is performed on the data in the target window to obtain the feature data of the target window, and the target window and the feature data have a corresponding relationship; The feature data of multiple target windows are converted into a two-dimensional table structure to obtain the target table; The target heatmap is generated based on the target table.

4. The capacitance health detection method according to claim 1, characterized in that, The regression branch network includes: a second convolutional layer, a second normalization layer, a second activation function layer, a global average pooling layer, and a first fully connected layer; the first feature information is input into the regression branch network to obtain the capacitance value prediction result and capacitance aging percentage prediction result output by the regression branch network, including: The first feature information is input into the second convolutional layer for feature extraction to obtain the second feature information, wherein the number of output channels of the second convolutional layer is greater than the number of output channels of the first convolutional layer; the second feature information is input into the second normalization layer for normalization processing to obtain the third feature information; The third feature information is input into the second activation function layer for activation processing to obtain the fourth feature information, the value of which is in the range of 0 to positive infinity; the fourth feature information is input into the global average pooling layer for global average pooling to obtain the fifth feature information. The fifth feature information is input into the first fully connected layer to perform capacitance value prediction and capacitance aging percentage prediction, so as to obtain the capacitance value prediction result and the capacitance aging percentage prediction result.

5. The capacitance health detection method according to claim 1, characterized in that, The classification branch network includes: a third convolutional layer, a third normalization layer, a third activation function layer, a max pooling layer, a second fully connected layer, and a Softmax activation function layer; the first feature information is input into the classification branch network to obtain the capacitor fault state prediction result output by the classification branch network, including: The first feature information is input into the third convolutional layer for feature extraction to obtain the sixth feature information, wherein the number of output channels of the third convolutional layer is greater than the number of output channels of the first convolutional layer; the sixth feature information is input into the third normalization layer for normalization processing to obtain the seventh feature information. The seventh feature information is input into the third activation function layer for activation processing to obtain the eighth feature information, the value of which is in the range of 0 to positive infinity; the eighth feature information is input into the max pooling layer for max pooling processing to obtain the ninth feature information; The ninth feature information is input into the second fully connected layer to predict the capacitor fault state and obtain intermediate prediction results; the intermediate prediction results are input into the Softmax activation function layer for activation processing to obtain the capacitor fault state prediction results, which include the prediction probability of each preset fault category.

6. The capacitance health detection method according to claim 1, 4, or 5, characterized in that, Based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, the initial network architecture is trained, including: Based on the capacitor value prediction result, the capacitor value label, and the preset first mean square error loss function, a first loss is obtained. The first mean square error loss function is used to characterize the difference between the capacitor value prediction result and the corresponding capacitor value label. Based on the capacitor aging percentage prediction result, the capacitor aging percentage label, and the preset second mean square error loss function, a second loss is obtained. The second mean square error loss function is used to characterize the difference between the capacitor aging percentage prediction result and the corresponding capacitor aging percentage label. Based on the capacitor fault state prediction result, the capacitor fault state label, and the preset cross-entropy loss function, a third loss is obtained. The cross-entropy loss function is used to characterize the difference between the capacitor fault state prediction result and the corresponding capacitor fault state label. Based on the first loss, the second loss, the third loss, and preset weight coefficients that correspond one-to-one with the first loss, the second loss, and the third loss, the final loss is obtained; The initial network architecture is iterated based on the final loss to complete the training.

7. A capacitance health detection system, characterized in that, include: The target association data acquisition module is used to acquire the target association data of the capacitor to be tested. The target association data includes the electrical parameter data of the capacitor to be tested within a preset time period after the start of charging, and the electrical parameter data within a preset time period after the start of discharging. The electrical parameter data includes current data and voltage data. A heatmap generation module is used to generate a target heatmap based on the target-related data; The capacitor health detection module is used to input the target heat map into a preset capacitor health detection model, perform feature extraction and capacitor health detection, so as to obtain the capacitance value, capacitor aging percentage and capacitor fault status output by the capacitor health detection model. The model training module is used to acquire a training set, which includes multiple heat map samples and sample labels corresponding to the heat map samples. The sample labels include: capacitance value label, capacitor aging percentage label, and capacitor fault status label. The heatmap samples are input into a shared network in a preset initial network architecture to obtain the first feature information output by the shared network. The shared network includes a first convolutional layer, a first normalization layer, a first activation function layer, and an average pooling layer connected in sequence. The first feature information is then input into a regression branch network and a classification branch network in the initial network architecture to obtain the capacitor value prediction result and capacitor aging percentage prediction result output by the regression branch network, and the capacitor fault state prediction result output by the classification branch network. Based on the capacitor value prediction result, the capacitor value label, the capacitor aging percentage prediction result, the capacitor aging percentage label, the capacitor fault state prediction result, and the capacitor fault state label, the initial network architecture is trained, and the trained initial network architecture is determined as the capacitor health detection model.

8. An electronic device, characterized in that, It includes a processor, a memory, and a communication bus; the communication bus is used to connect the processor and the memory; the processor is used to execute a computer program stored in the memory to implement the capacitance health detection method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores a computer program that enables the computer to perform the capacitance health detection method as described in any one of claims 1 to 6.

Images