A direct current arc fault detection method based on image analysis and domain adaptation

Abstract

The application relates to a direct-current arc fault detection method based on image analysis and field adaptation, which comprises the following steps: building a direct-current micro-grid arc fault test experiment platform, simulating normal working conditions and arc fault working conditions of changing loads, and recording currents at various positions under various working conditions; analyzing time-frequency domain characteristics of the currents, recording time-frequency domain images, and marking working condition states; adopting multiple methods to extract features of the time-frequency domain characteristic images and performing merging to obtain a joint feature vector; taking the joint feature vectors of the images at different positions and state labels as source domain data and target domain data respectively, and performing dimension reduction on the feature vectors; constructing a classifier, adjusting parameters, performing data distribution adaptive processing, training the classifier by minimizing edge distribution distance and conditional distribution distance of the source domain and the target domain, and using the trained classifier to realize arc fault detection. Compared with the prior art, the application has the advantages of higher algorithm accuracy, smaller calculation amount and the like.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electrical engineering, in particular to a direct-current arc fault detection method based on image analysis and field adaptation. BACKGROUND

[0002] In the operation of a direct-current micro-grid system, problems such as cable joint loosening, insulation layer damage and poor contact may cause a direct-current arc fault, and if the arc is not extinguished in time, an electrical fire may be caused. Among them, the direct-current series arc fault occurs frequently, which may cause the resistance to increase, and according to the different positions of the series arc fault, the current amplitude may increase, decrease or even remain basically unchanged. The complex fault phenomenon increases the difficulty of detecting the series arc fault. Selecting effective fault features is conducive to the accurate detection of the arc fault. In addition, when the probability density functions of the training data and the test data conform to different distributions, the prediction effect of the classifier under the traditional machine learning method is usually poor. That is, the prior art has the problems that the arc fault features cannot be effectively extracted and the prediction algorithm is difficult to adapt to different data distributions. SUMMARY

[0003] The purpose of the present application is to provide a direct-current arc fault detection method based on image analysis and field adaptation, which can effectively extract arc fault features and has field adaptation capability, and realize rapid and accurate detection of direct-current arc faults.

[0004] The purpose of the present application can be achieved by the following technical solutions:

[0005] A direct-current arc fault detection method based on image analysis and field adaptation, comprising the following steps:

[0006] S1: A direct-current micro-grid arc fault test experiment platform is built, platform parameters are adjusted according to the actual working conditions in the direct-current micro-grid by using the control variable method to perform experiments, an external resistance load is connected in parallel to the load of the direct-current micro-grid system through a circuit breaker switch, normal working conditions of the load are simulated, arc fault working conditions are simulated under different arc lengths and different positions, and the current at each position under each working condition is measured and recorded;

[0007] S2: Time-frequency domain features of the current under each working condition at different positions are analyzed, time-frequency domain images are recorded and the corresponding working condition states are marked;

[0008] S3: A plurality of image feature extraction methods are used to extract features from the time-frequency domain feature images, and the multi-feature vectors are combined to obtain a joint feature vector of the image;

[0009] S4: the joint feature vectors of images at different positions and the state labels are taken as source domain data and target domain data respectively, and the feature vectors of the source domain and the target domain are reduced in dimension, wherein the source domain data has working condition state labels, and the target domain data does not contain labels;

[0010] S5: arc fault detection based on domain adaptation: constructing a classifier and adjusting parameters, performing data distribution adaptive processing on the source domain and target domain data, training the classifier by minimizing the edge distribution distance and conditional distribution distance between the source domain and the target domain, and realizing arc fault detection based on the trained classifier.

[0011] Further, the direct current microgrid arc fault test experiment platform comprises a direct current microgrid system and an arc generating device, wherein the direct current microgrid system comprises a direct current source and a constant power load, wherein the direct current source is used for power supply, and the constant power load is connected with the direct current source through a DC-DC converter; the arc generating device comprises a lead screw sliding table, copper rods, a driver and a controller, a stepping motor is controlled by the driver to separate the two copper rods to generate an arc, and the arc generating device is connected in series in different positions of branches of the direct current microgrid to simulate an actual arc fault.

[0012] Further, the positions during simulation of the arc fault working condition include a first position and a second position, the first position is an output end of the DC-DC converter, and the second position is an input end of the DC-DC converter.

[0013] Further, the S1 comprises the following steps:

[0014] S11: setting a direct current source output voltage, adjusting an arc generating device controller so that the two copper rods are in complete contact, turning on a circuit breaker switch so that the entire experiment circuit is closed, and starting the experiment;

[0015] S12: recording current waveforms and data at two ends of the arc generating device under normal working conditions through an oscilloscope;

[0016] S13: switching the circuit breaker switch of the external resistance load, and recording current waveforms and data at the load end and the power supply end under the changed load condition through the oscilloscope;

[0017] S14: placing the arc generating device at the first position, adjusting the controller to separate the two copper rods to generate an arc, and recording current waveforms and data at two ends of the arc generating device during the entire process through the oscilloscope;

[0018] S15: placing the arc generating device at the second position, adjusting the controller to separate the two copper rods to generate an arc, and recording current waveforms and data at two ends of the arc generating device during the entire process through the oscilloscope;

[0019] S16: change the direct current micro-grid system current and voltage, repeat steps S12 and S13, and record the current and data at different positions under different working conditions;

[0020] S17: change the arc length and the direct current micro-grid system current, repeat steps S14 and S15, and record the arc fault current and data under different arc lengths and arc fault positions.

[0021] Further, the S2 is specifically: according to the measured current data, time-frequency domain analysis is carried out on the current to obtain the image of the current about time domain information and frequency domain information, and the function relationship is as follows:

[0022] I=F1(t)+F2(f)

[0023] Wherein, I is the current data recorded by the oscilloscope, F1(t) represents the time domain information of the current, and F2(f) represents the frequency domain information of the current. The decomposed time-frequency domain information is drawn into an image and saved, and the corresponding working condition is marked.

[0024] Further, the S3 is specifically: two kinds of image feature extraction methods are used to extract features of the time-frequency domain information images of different positions under various working conditions, the obtained multi-dimensional image features are converted into one-dimensional feature vectors, and the one-dimensional features obtained by the two kinds of image feature extraction methods are combined to form a new feature vector:

[0025] vector=[vector1 vector2]

[0026] Wherein, vector represents the combined feature vector obtained by the two kinds of image feature extraction methods, and vector1 and vector2 represent the feature vectors obtained by the two different image feature extraction methods respectively.

[0027] Further, in the S4, the feature vector of the current at the first position and the state label are taken as source domain data, and the feature vector of the current at the second position is taken as target domain data.

[0028] Further, the constant power load is one or more. When the constant power load is one, the source domain and the target domain data come from different positions of the same load branch. When the constant power load is multiple, the source domain and the target domain data come from different positions of the same load branch or different positions of different load branches.

[0029] Further, the S4 includes the following steps:

[0030] S41, the current data and the state label at a certain position are selected to constitute source domain data:

[0031]

[0032] wherein D s represents source domain data, x i represents the feature vector of the i-th group of current data, y i represents the category vector corresponding to the i-th group of current data, and n represents a total of n groups of source domain current data;

[0033] S42, selecting the feature vectors of the currents at different positions from the source domain data to constitute target domain data:

[0034]

[0035] wherein D t represents target domain data, x j represents the feature vector of the j-th group of current data, and m represents a total of m groups of target domain current data;

[0036] S43, using a principal component analysis method to reduce the dimensionality of the feature vectors in the source domain and the target domain.

[0037] Further, the S5 comprises the following steps:

[0038] S51: mapping the source domain and target domain data to a Hilbert space;

[0039] S52: using maximum mean difference to measure the distribution distance between the data of the source domain and the target domain:

[0040]

[0041] MMD represents the distance of maximum mean difference, X1 represents the feature space of the source domain, X2 represents the feature space of the target domain, n1 and n2 respectively represent the number of feature vectors of the source domain and the target domain, and Φ(·) represents mapping, which is used to map the original variable to a reproducing Hilbert space;

[0042] S53: using the Lagrange multiplier method to minimize the marginal distribution distance between the source domain and the target domain;

[0043] S54: substituting the target domain data into the classifier trained by the source domain data to obtain target domain pseudo-labels;

[0044] S55: using the Lagrange multiplier to minimize the conditional distribution distance between the source domain and the target domain;

[0045] S56: updating the distribution distance between the source domain and the target domain, repeating the above steps S52-S55, and through multiple iterations, the pseudo-labels are close to the real labels until convergence or a preset number of iterations is reached, and the classifier training is completed;

[0046] S57: realizing arc fault detection based on the trained classifier.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] (1) Compared with the existing direct current series arc fault detection method, the present application adopts a variety of image feature extraction methods to analyze the current signal containing current time domain and frequency domain information, which can effectively extract arc fault features.

[0049] (2) The present application adopts data distribution self-adaptation, has the effect of data dimension reduction, and has smaller calculation amount and faster calculation speed than the classifier under the traditional machine learning method.

[0050] (3) Based on the idea of transfer learning, the present application can still realize reliable detection of direct current series arc fault in actual application scenes with different data probability distributions, and improve the detection accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a method flowchart of the present application;

[0052] Figure 2 is a DC microgrid fault arc system circuit diagram for testing in an embodiment;

[0053] Figure 3 is a specific implementation process diagram of the present application;

[0054] Figure 4 is a classification error change graph with iteration number in an embodiment. DETAILED DESCRIPTION

[0055] The present application will be described in detail below in combination with the drawings and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present application, and gives a detailed implementation manner and specific operation process, but the protection scope of the present application is not limited to the following embodiments.

[0056] Since most existing arc fault detection methods assume that the application scene data and the experimental data conform to a unified probability distribution, which is different from the actual situation, the accuracy of the existing arc fault detection algorithm is greatly reduced in actual application. The present embodiment proposes an arc fault detection method based on data distribution self-adaptation for the case that the arc fault data probability distributions are inconsistent, obtains the direct current arc fault features through time-frequency domain analysis method and image feature extraction method, and aligns the data distribution by using the field self-adaptation method. Finally, the arc fault detection algorithm is verified by MATLAB.

[0057] The present embodiment provides a direct current arc fault detection method based on image analysis and field self-adaptation, as shown in Figure 1As shown, it mainly includes data acquisition, time-frequency domain analysis of current, image feature extraction, arc fault detection based on domain adaptation, and specifically includes the following steps:

[0058] S1: Build a DC microgrid arc fault test platform.

[0059] According to the actual working conditions in the DC microgrid, the control variable method is used to adjust the platform parameters for experiments. The DC microgrid arc fault test platform includes a DC microgrid system and an arc generation device. As shown in Figure 2 , the DC microgrid system includes a 110V DC source and a constant power load, and the DC source is used for power supply, and the constant power load is connected with the DC source through a DC-DC converter; the arc generation device includes a lead screw sliding table, a copper rod, a driver, and a controller, the driver controls the stepping motor to separate the two copper rods to generate arc, and the arc generation device is connected in series in different positions of each branch of the DC microgrid to simulate the actual arc fault. In this embodiment, the DC-DC converter is a 110V to 48V one-way DC / DC converter, and the most common Buck circuit in the DC microgrid is used.

[0060] In this embodiment, as shown in Figure 2 , the external resistance load is connected in parallel to the load of the DC microgrid system through a circuit breaker switch to simulate the normal working condition of changing the load; the arc generation device is controlled by the driver to generate arc, and the arc fault condition is simulated under different arc lengths and different positions, and the current at each position under each condition is measured and recorded.

[0061] In the DC microgrid system, only the output end of the DC-DC converter and the input end of the DC-DC converter (corresponding to Figure 2 Position ① / ①' and Position ② / ②', respectively) can occur arc fault, the arc faults occurring at the two positions have different properties, and the data of the source domain and the target domain are similar but not the same, so the power data of the two positions can be used as the source domain and the target domain data, respectively.

[0062] The constant power load in the DC microgrid system can be one or more. When the constant power load is one, the data of the source domain and the target domain come from different positions of the same load branch, i.e. the input end and the output end of the same DC-DC converter (such as Position ① and Position ②'); when the constant power load is multiple, the data of the source domain and the target domain can come from different positions of the same load branch (such as Position ① and Position ②'), or from different positions of different load branches (such as Position ① and Position ②). As shown in Figure 2 , the system includes two constant power loads, and in this embodiment, Position ① and Position ② are used as the arc generation device access positions when simulating arc fault conditions.

[0063] According to the above, the arc fault detection method based on domain adaptation is as follows:Figure 2 The system shown, in particular, S1 includes the following steps:

[0064] S11: Set the DC source output voltage to 110V, adjust the arc generation device controller so that the two copper rods are in complete contact, open the circuit breaker switch so that the entire experimental circuit is closed, and start the experiment;

[0065] S12: Record the current waveform and data at both ends of the arc generation device under normal operating conditions through an oscilloscope;

[0066] S13: Switch the circuit breaker switch of the external resistance load, and record the current waveform and data at the load end and the power end under the changed load condition through an oscilloscope;

[0067] S14: Place the arc generation device at position ①, adjust the controller to separate the two copper rods for arc drawing, and record the current waveform and data at both ends of the arc generation device during the entire process through an oscilloscope;

[0068] S15: Place the arc generation device at position ②, adjust the controller to separate the two copper rods for arc drawing, and record the current waveform and data at both ends of the arc generation device during the entire process through an oscilloscope;

[0069] S16: Change the DC microgrid system current and voltage, repeat steps S12 and S13, and record the current and data at different positions under different operating conditions;

[0070] S17: Change the arc length and DC microgrid system current, repeat steps S14 and S15, and record the arc fault current and data under different arc lengths and arc fault occurrence positions.

[0071] In this embodiment, the switch at the resistance load of the converter 1 branch is switched and an arc fault occurs at position ①, and the oscilloscope is used to record the current change process at position ①. The switch at the resistance load of the converter 2 branch is switched and an arc fault occurs at position ②, and the oscilloscope is used to record the current change process at position ②. Under the above two operating conditions, the current is adjusted respectively, the experiment is repeated multiple times, and the load current is recorded.

[0072] In one embodiment, S1 is implemented by the following steps: 1. Set different arc generation positions, current sizes, and arc lengths, such as Figure 2①-②, referring to the UL1699B test standard and common voltage levels of DC microgrids, the voltage was selected as 110V. The current at point ① was set to 3A, 4A, 5A, 6A, 7A, 8A, 9A and 10A respectively, and the current at point ② was set to 8A, 9A and 10A respectively. The arc lengths were 0.5mm, 0.7mm, 0.9mm, 1.1mm and 1.3mm respectively. Other experimental conditions remained unchanged. 2. The arc generating device was adjusted to control the copper rod to separate at a uniform speed to generate an arc. 3. The experimental data and current waveforms were recorded using an oscilloscope. The experimental parameters were changed and the experiment was repeated 10 times for each experimental condition.

[0073] Step S1 is the data acquisition process, and steps S2-S5 are the implementation process of the arc fault detection algorithm, such as... Figure 3 As shown, arc fault detection is achieved through time-frequency domain feature extraction of current data, joint image feature extraction, and adaptive data distribution. The initial work for time-frequency domain feature extraction of current data includes, as described in S1, inserting arc generating devices at different locations, repeating experiments with varying parameters, and recording data. The obtained current data is then analyzed in the time-frequency domain to obtain a time-frequency domain information image. Multiple image feature extraction methods are then used to obtain a joint feature vector. Finally, unsupervised transfer learning is performed on the obtained data to achieve accurate arc fault detection even with inconsistent data distribution.

[0074] S2: Analyze the time-frequency domain characteristics of the current at different locations under various operating conditions, record the time-frequency domain images, and mark the corresponding operating conditions.

[0075] Based on the current data obtained from the above two locations under two working conditions, time-frequency domain analysis was performed on the current to obtain the current images with respect to time-domain information and frequency-domain information. The functional relationship is shown in Equation (1).

[0076] I = F1(t) + F2(f) (1)

[0077] Where I represents the current data recorded by the oscilloscope, F1(t) represents the time domain information of the current, and F2(f) represents the frequency domain information of the current. The decomposed time and frequency domain information is plotted into an image and saved, and its corresponding operating condition is marked.

[0078] S3: Employ multiple image feature extraction methods to extract features from the time-frequency domain feature image, and merge the multiple feature vectors to obtain the joint feature vector of the image.

[0079] This embodiment uses two image feature extraction methods to extract features from time-frequency domain information images at different locations under various working conditions. The obtained multidimensional image features are converted into one-dimensional feature vectors, and then the one-dimensional features obtained by the two image feature extraction methods are combined to form a new feature vector, as shown in Equation (2).

[0080] vector=[vector1 vector2] (2)

[0081] Wherein, vector represents the joint feature vector obtained by two image feature extraction methods, vector1 represents the feature vector obtained by image feature extraction method 1, and vector2 represents the feature vector obtained by image feature extraction method 2.

[0082] In an embodiment, S3 is realized by the following steps: 1, saving the obtained time-frequency domain information images uniformly as a format; 2, storing the images at position ① into a training data folder according to the working conditions, and storing the images at position ② into a detection data folder; 3, using two image feature extraction methods to extract features from the images respectively, and obtaining the feature vectors of each image under the two methods; 4, combining the two feature vectors corresponding to each image together to obtain a new joint feature vector.

[0083] S4: The joint feature vectors of images at different positions and state labels are taken as source domain data and target domain data respectively, and the feature vectors of the source domain and the target domain are reduced in dimension, wherein the source domain data has working condition state labels, and the target domain data does not contain labels.

[0084] Since the arc fault current waveform at position ① is more common, and presents a downward trend in the time domain, while the current at position ② presents an upward trend, therefore, in this embodiment, the feature vector of the current at position ① and the state label are taken as source domain data, and the feature vector of the current at position ② is taken as target domain data. In another embodiment, the feature vector of the current at position ② and the state label can be taken as source domain data, and the feature vector of the current at position ① is taken as target domain data.

[0085] Specifically, S4 includes the following steps:

[0086] S41, selecting the current data at position ① and the state label to constitute the source domain data:

[0087]

[0088] Wherein, D s represents the source domain data, i.e. the data at position ①, x i represents the feature vector of the i-th group of current data, y i represents the category vector corresponding to the i-th group of current data, and n represents a total of n groups of source domain current data.

[0089] S42, selecting the feature vector of the current at position ② to constitute the target domain data:

[0090]

[0091] wherein D t represents target domain data, i.e. data at position ②, x j represents the feature vector of the jth group of current data, and m represents a total of m groups of target domain current data;

[0092] S43, using the method of principal component analysis, the feature vectors in the source domain and the target domain are reduced to k dimensions, to maximize the information and structure of the data retained.

[0093] S5: Arc fault detection based on domain adaptation: constructing a classifier and adjusting parameters, performing data distribution adaptive processing on the source domain and the target domain data, training the classifier by minimizing the edge distribution distance and the conditional distribution distance between the source domain and the target domain, and realizing arc fault detection based on the trained classifier.

[0094] S51: mapping the source domain and the target domain data to Hilbert space;

[0095] S52: using maximum mean difference to measure the distribution distance between the source domain and the target domain data:

[0096]

[0097] MMD represents the distance of maximum mean difference, X1 represents the feature space of the source domain, X2 represents the feature space of the target domain, n1 and n2 represent the number of feature vectors of the source domain and the target domain respectively, and Φ(·) represents mapping, which is used to map the original variable to the reproducing Hilbert space;

[0098] S53: minimizing the edge distribution distance between the source domain and the target domain by using the Lagrange multiplier method;

[0099] S54: substituting the target domain data into the classifier trained by the source domain data to obtain the target domain pseudo label;

[0100] S55: minimizing the conditional distribution distance between the source domain and the target domain by using the Lagrange multiplier;

[0101] S56: updating the distribution distance between the source domain and the target domain, repeating the above steps S52-S55 until convergence or reaching a preset iteration number, completing the classifier training (since the pseudo label accuracy is not high, multiple iterations are needed to make the pseudo label close to the true label);

[0102] S57: realizing arc fault detection based on the trained classifier.

[0103] In this embodiment, the relationship between the algorithm classification error and the number of iterations is as shown in Figure 4 When the data is reduced to 10 dimensions, the radial basis function is used, the accuracy is optimal, the number of iterations is set to 5 times, and the relationship between the classification accuracy and the number of iterations is as shown inFigure 4 As shown in the figure Figure 4 It can be found that the initial classification accuracy is about 0.815 after data dimension reduction, and the accuracy can reach 0.888 as the number of iterations increases, and the accuracy is improved by about 0.073.

[0104] In summary, the application comprehensively considers the time-frequency domain features of arc faults, the joint image feature extraction method and the field adaptation problem, and the algorithm has higher accuracy and smaller calculation amount, and is more suitable for actual application scenarios.

[0105] The preferred embodiments of the application are described in detail above. It should be understood that those skilled in the art can make many modifications and changes without creative labor according to the concept of the application. Therefore, any technical solutions obtained by logical analysis, reasoning or limited experiments on the basis of the prior art according to the concept of the application shall be within the protection scope determined by the claims.

Claims

1. A DC arc fault detection method based on image analysis and domain adaptation, characterized in that, Includes the following steps: S1: Build a DC microgrid arc fault test experimental platform. Based on the actual working conditions in the DC microgrid, use the control variable method to adjust the platform parameters for the experiment. Connect an external resistive load to the load of the DC microgrid system through a circuit breaker switch to simulate the normal working conditions of the load. Simulate the arc fault working conditions under different arc lengths and positions, and measure and record the current at each position under each working condition. S2: Analyze the time-frequency domain characteristics of the current under various operating conditions at different locations, record the time-frequency domain images, and mark the corresponding operating conditions; S3: Employ multiple image feature extraction methods to extract features from the time-frequency domain feature image, and merge the multiple feature vectors to obtain the joint feature vector of the image; S4: Use the joint feature vector and state label of images at different locations as source domain data and target domain data respectively, and reduce the dimensionality of the feature vectors of the source domain and target domain. The source domain data has working condition status labels, while the target domain data does not have labels. S5: Domain-Adaptive Arc Fault Detection: A classifier is constructed and its parameters are adjusted. Data distribution of source and target domain data is adaptively processed. The classifier is trained by minimizing the edge distribution distance and conditional distribution distance between the source and target domains. Arc fault detection is achieved based on the trained classifier.

2. The DC arc fault detection method based on image analysis and domain adaptation according to claim 1, characterized in that, The DC microgrid arc fault testing experimental platform includes a DC microgrid system and an arc generating device. The DC microgrid system includes a DC source and a constant power load. The DC source is used for power supply, and the constant power load is connected to the DC source through a DC-DC converter. The arc generating device includes a lead screw slide, copper rods, a driver, and a controller. The driver controls a stepper motor to separate two copper rods to generate an arc. The arc generating device is connected in series at different positions in each branch of the DC microgrid to simulate actual arc faults.

3. The DC arc fault detection method based on image analysis and domain adaptation according to claim 2, characterized in that, The simulated arc fault conditions include a first position and a second position. The first position is the output terminal of the DC-DC converter, and the second position is the input terminal of the DC-DC converter.

4. The DC arc fault detection method based on image analysis and domain adaptation according to claim 3, characterized in that, S1 includes the following steps: S11: Set the DC source output voltage, adjust the arc generator controller to make the two copper rods fully contact each other, open the circuit breaker switch to close the entire experimental circuit, and start the experiment. S12: Record the current waveforms and data at both ends of the arc generator under normal operating conditions using an oscilloscope; S13: Switch the circuit breaker switch for the external resistive load, and record the current waveforms and data at the load end and power supply end under the changed load conditions using an oscilloscope; S14: Place the arc generator in the first position, adjust the controller to separate the two copper rods to draw the arc, and record the current waveform and data at both ends of the arc generator through an oscilloscope throughout the process. S15: Place the arc generator in the second position, adjust the controller to separate the two copper rods to draw the arc, and record the current waveform and data at both ends of the arc generator through an oscilloscope throughout the process. S16: Change the current and voltage of the DC microgrid system, repeat steps S12 and S13, and record the current and data at different locations under different operating conditions; S17: Change the arc length and DC microgrid system current, repeat steps S14 and S15, and record the arc fault current and data under different arc lengths and arc fault locations.

5. The DC arc fault detection method based on image analysis and domain adaptation according to claim 1, characterized in that, Specifically, S2 involves: performing time-frequency domain analysis on the current based on the measured current data to obtain images of the current with respect to time-domain and frequency-domain information, with the following functional relationships: I = F1(t) + F2(f) Where I represents the current data recorded by the oscilloscope, F1(t) represents the time domain information of the current, and F2(f) represents the frequency domain information of the current. The decomposed time and frequency domain information is plotted into an image and saved, and its corresponding operating condition is marked.

6. The DC arc fault detection method based on image analysis and domain adaptation according to claim 1, characterized in that, Specifically, S3 involves simultaneously using two image feature extraction methods to extract features from time-frequency domain information images at different locations under various working conditions, converting the obtained multi-dimensional image features into a one-dimensional feature vector, and then combining the one-dimensional features obtained from the two image feature extraction methods to form a new feature vector. vector = [vector1 vector2] Where vector represents the joint feature vector obtained by the two image feature extraction methods, and vector1 and vector2 represent the feature vectors obtained by the two different image feature extraction methods, respectively.

7. The DC arc fault detection method based on image analysis and domain adaptation according to claim 3, characterized in that, In step S4, the feature vector and state marker of the current at the first position are used as source domain data, and the feature vector of the current at the second position is used as target domain data.

8. The DC arc fault detection method based on image analysis and domain adaptation according to claim 1, characterized in that, The constant power load can be one or more. When there is only one constant power load, the data in the source domain and the target domain come from different locations of the same load branch. When there are multiple constant power loads, the data in the source domain and the target domain come from different locations of the same load branch or different locations of different load branches.

9. The DC arc fault detection method based on image analysis and domain adaptation according to claim 1, characterized in that, S4 includes the following steps: S41. Select current data and status labels at a certain location to form source domain data: Among them, D s Indicates source domain data, x i y represents the feature vector of the i-th group of current data. i This represents the category vector corresponding to the i-th group of current data, and n represents the total number of n groups of source domain current data. S42. Select the feature vectors of the current at different locations from the source domain data to construct the target domain data: Among them, D t Represents the target domain data, x j Let represent the feature vector of the j-th current data group, and m represent the total number of m groups of target domain current data. S43. Use principal component analysis to reduce the dimensionality of the eigenvectors in the source and target domains.

10. The DC arc fault detection method based on image analysis and domain adaptation according to claim 1, characterized in that, S5 includes the following steps: S51: Map source and target domain data to Hilbert space; S52: The maximum mean difference is used to measure the distributional distance between the data in the source and target domains. MMD represents the distance of the maximum mean difference, X1 represents the feature space of the source domain, X2 represents the feature space of the target domain, n1 and n2 represent the number of feature vectors in the source domain and the target domain, respectively, and Φ(·) represents the mapping used to map the original variables to the regenerated Hilbert space. S53: Minimize the edge distribution distance between the source and target domains using the Lagrange multiplier method; S54: Substitute the target domain data into the classifier trained from the source domain data to obtain the target domain pseudo-label; S55: Minimize the conditional distribution distance between the source and target domains using Lagrange multipliers; S56: Update the distribution distance between the source domain and the target domain. Repeat steps S52-S55 above. Through multiple iterations, make the pseudo-labels close to the real labels until convergence or the preset number of iterations is reached to complete the classifier training. S57: Arc fault detection is achieved based on a trained classifier.

Images