Zero-crossing opening control method and system based on arc energy transfer prediction and application in ring main unit

Abstract

The application discloses a zero-crossing breaking control method and system based on arc energy transfer prediction and application in ring network cabinet, and relates to the technical field of zero-crossing breaking. Current data and burning images caused by arc energy due to the decay of current data over time are collected. Through Fourier transform, the current data is converted into a current spectrum diagram, which can more accurately reflect the information of the current. Then, a convolutional neural network is used to detect the state of current change. Using an associative neural network, the burning condition of the surface of the contact is combined with the current data, and association and fusion are performed to obtain an arc correlation feature vector. The change of the arc correlation feature vector at multiple time points can more accurately predict the zero-crossing point.

Description

Technical Field

[0001] This invention relates to the field of zero-crossing interruption technology, and more specifically, to a zero-crossing interruption control method, system, and application based on arc energy transfer prediction. Background Technology

[0002] Alternating current will naturally pass through zero point once every certain period of time. An ideal switching device should extinguish the arc the instant the current crosses zero, because at this moment the arc energy is minimal, extinguishing it is easiest, and the resulting overvoltage is also minimal.

[0003] To achieve precise zero-crossing interruption, the switch cannot passively wait for the current to cross zero. It needs to actively manipulate the arc, creating optimal conditions in the arc-extinguishing medium (such as SF6 gas) in advance, predicting and guiding the arc to extinguish precisely when the next current zero point arrives. This process involves efficiently transferring, dispersing, and cooling the arc energy from the contact gap.

[0004] Circuit breakers, as crucial equipment in the entire power supply system, are essential for the safe operation of the power system. Therefore, improving the intelligence level of circuit breaker breaking operations is of great significance to the safe operation of the power grid. During the interruption of short-circuit current, circuit breakers undergo transient changes, and the short-circuit current contains a non-periodic component that decays over time. This non-periodic decay component is random and uncertain, which increases the difficulty of accurately predicting the zero-crossing point of the short-circuit current. How to predict the zero-crossing point of the short-circuit current based on its characteristic parameters is the primary problem that must be solved for the controllable interruption of short-circuit current by circuit breakers. Summary of the Invention

[0005] The purpose of this invention is to provide a zero-crossing interruption control method and system based on arc energy transfer prediction, so as to solve the above-mentioned problems existing in the prior art.

[0006] This invention is used to address situations where an electric arc is generated because the interruption time does not cross zero due to time decay.

[0007] In a first aspect, embodiments of the present invention provide a zero-crossing interruption control method based on arc energy transfer prediction, comprising: Collect current data and scorch images at m+n time points; the current data represents the current waveform passing through the contact; the scorch image represents an image of the contact surface; A trained correlation neural network is obtained; the correlation neural network is capable of detecting the damage to the contacts caused by the arc when different currents are disconnected. The current data is converted into a current spectrum using Fourier transform. By using a trained neural network, based on m+n scorching images and current spectrum diagrams, the relationship between current and arc energy shown in the scorching images is detected, resulting in m+n-1 arc-related feature vectors. According to the time points from morning to night, m+n-1 arc-related feature vectors are input into a temporal convolutional network for prediction to obtain the zero-crossing points.

[0008] Optionally, the associated neural network includes a first convolutional network, a second convolutional network, and a fusion neural network; The first convolutional network contains 2*2 convolutional kernels; The second convolutional network contains 2*2*2 three-dimensional convolutional kernels.

[0009] Optionally, the step involves using a trained correlation neural network to detect the relationship between the current and the arc energy shown in the burning images, based on m+n ablation images and current spectrum diagrams, to obtain m+n-1 arc-related feature vectors, including: Based on m+n ablation images, the gray value changes at adjacent time points are detected to obtain m+n-1 ablation difference images and arc change vectors. With a stride of 1, the scorching difference image is convolved with a 2*2 convolution kernel in the first convolutional network to detect the scorching state of the scorching difference image and obtain a first scorching feature map; the scorching feature map represents the characteristics of the contact surface changes as the electric arc burns. m+n-1 ablation difference images correspond to m+n-1 first ablation feature maps; Based on the current spectrum diagram, the current change state is detected to obtain m+n-1 current spectrum feature diagrams; Based on the arc change vector, m+n-1 first burning feature maps and m+n-1 current spectrum feature maps, an association relationship is constructed to obtain the arc association feature vector; the arc association feature vector represents the association feature between the area and color change of the contact burning and the current change.

[0010] Optionally, the step of detecting the current change state based on the current spectrum to obtain m+n-1 current spectrum feature maps includes: The current spectrum is divided according to time points to obtain m+n time-current change spectrum diagrams; The time-current spectrum diagrams corresponding to two adjacent time points are superimposed to obtain a superimposed three-dimensional spectrum diagram. With a stride of 1, the superimposed three-dimensional spectrum map is convolved with a 2*2*2 three-dimensional convolution kernel in the second convolutional network to obtain a current spectrum feature map. The m+n time-current change spectrum diagrams correspond to the m+n-1 current spectrum feature diagram.

[0011] Optionally, the step of constructing an association relationship based on the arc change vector, m+n-1 first scorching feature maps, and m+n-1 current spectrum feature maps to obtain an arc association feature vector includes: The first scorching feature map, the corresponding current spectrum feature map, and the scorching depth and scorching area values ​​in the corresponding arc change vector are input into the fusion neural network to obtain the arc-related feature vector. m+n-1 current feature maps correspond to m+n-1 arc-related feature vectors; The fusion neural network comprises an input layer, a hidden layer, and an output layer; the input layer comprises input neurons; the hidden layer comprises hidden neurons; and the output layer comprises output neurons. Use the elements of the electric arc change vector as the input to the input layer; The sum of the number of elements in the first ablation feature map and the number of elements in the current spectrum feature map plus 2 is used as the correlation number; The number of associated input neurons corresponds to one hidden neuron, and one input neuron corresponds to only one hidden neuron; the output of the output layer is the arc change feature vector.

[0012] Optionally, the step of detecting grayscale value changes at adjacent time points based on m+n ablation images to obtain m+n-1 ablation difference images and arc change vectors includes: Obtain the initial time point; the initial time point is a time point earlier than other time points. The ablation image at the initial time point is used as the background image; Based on the time points from morning to night, multiple ablation images are compared with the background image to obtain ablation similarity value; If the scorching similarity value is less than or equal to the similarity threshold, set all m-1 scorching difference images corresponding to m scorching images to 0; If the scorching similarity value is greater than the similarity threshold, edge detection is performed on the n scorching images to obtain n scorching positions; the gray value of the scorching position is subtracted from the gray value of the corresponding position in the background image to obtain the scorching gray value; Based on n scorching locations, n scorching gray values, and the corresponding n scorching images, n scorching difference images are obtained; the scorching difference images represent the changes in the contact scorching state caused by the electric arc at two adjacent time points. Based on m+n scorching positions and m+n scorching gray values, an arc change vector is obtained; the arc change vector contains (m+n-1)*2 elements.

[0013] Optionally, the step of obtaining n scorching difference images based on n scorching locations, n scorching grayscale values, and the corresponding n scorching images includes: The ablation grayscale values ​​are marked at the corresponding ablation positions to obtain the ablation image to be judged; the values ​​of the ablation image to be judged are all 0 except for the ablation positions; The earliest time point among the n ablation images to be judged is taken as one ablation difference image; Subtract the corresponding gray value in the ablation image at the previous time point from the gray value of the ablation image at the next time point to obtain n-1 ablation difference images.

[0014] Optionally, the step of obtaining the arc change vector based on m+n scorching positions and m+n scorching gray values ​​includes: If the scorching similarity value is less than or equal to the similarity threshold, set the n-1 scorching depth values ​​and n-1 scorching area values ​​corresponding to the n scorching images to 0; If the scorching similarity value is greater than the similarity threshold, the number of scorching locations at the next time point is subtracted from the number of scorching locations at the previous time point to obtain m scorching area values. The scorching ash value at the same scorching position at the next time point as the previous time point is taken as the scorching depth value. Based on the time points from morning to night, construct the arc change vector by taking m+n-1 scorching depth values ​​and m+n-1 scorching area values.

[0015] Optionally, the training method for the associated neural network includes: Acquire training current data, corresponding training ablation images, and marked zero-crossing points at historical time points; the training current data and training ablation images represent the data used for training; the marked zero-crossing points represent the marked zero-crossing points at the next time point after the historical time point used for training. Based on the ablation images at historical time points, the predicted zero-crossing points are obtained through correlation neural networks and temporal convolutional networks; The loss is calculated by comparing the predicted zero-crossing points with the labeled zero-crossing points, and then used to train the correlation neural network and the temporal convolutional network.

[0016] Secondly, embodiments of the present invention provide a zero-crossing interruption control system based on arc energy transfer prediction, comprising: The acquisition module is used to collect current data and scorch images at m+n time points; the current data represents the current waveform passing through the contact; the scorch image represents the image of the contact surface; and a trained correlation neural network is acquired; the correlation neural network can detect the damage to the contact caused by the arc when different currents are disconnected. The Fourier transform module is used to convert current data into a current spectrum graph through Fourier transform. The correlation module is used to detect the relationship between the current and the arc energy shown in the burning images based on m+n burning images and current spectrum diagrams through a trained correlation neural network, and obtain m+n-1 arc correlation feature vectors. The prediction module is used to input m+n-1 arc-related feature vectors into the temporal convolutional network according to the time points from morning to night, and to make predictions to obtain the zero-crossing points.

[0017] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects: This invention also provides a zero-crossing interruption control method and system based on arc energy transfer prediction.

[0018] In this invention, current data and images of arcing caused by the decay of current over time are collected. Fourier transform is used to convert the current data into a current spectrum, which more accurately reflects the current information. A convolutional neural network is then used to detect changes in the current. A correlation neural network is used to combine the current data with the corresponding arcing conditions (area and grayscale changes) on the contact surface, correlated and fused to obtain an arc-related feature vector. By using the changes in the arc-related feature vector at multiple time points, a more accurate prediction of zero-crossing points is achieved. Attached Figure Description

[0019] Figure 1 This is a flowchart of a zero-crossing interruption control method based on arc energy transfer prediction provided by an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings. Example

[0021] like Figure 1 As shown, this embodiment of the invention provides a zero-crossing interruption control method based on arc energy transfer prediction, the method comprising: S101: Collect current data and scorch images at m+n time points; the current data represents the current waveform passing through the contact; the scorch image represents an image of the contact surface.

[0022] Where m and n are positive integers.

[0023] The ablation image is a grayscale image.

[0024] The contact mentioned above is the contact of a circuit breaker.

[0025] The current waveform in the current data can simulate waveforms under various power grid conditions.

[0026] The above method is used because there is a non-periodic component that decays over time during the short circuit process, which prevents the current zero point from changing periodically. Therefore, deep learning is needed to predict the zero point.

[0027] S102: Obtain the trained correlation neural network; the correlation neural network is capable of detecting the damage to the contacts caused by the arc when different currents are disconnected.

[0028] S103: Convert the current data into a current spectrum graph through Fourier transform.

[0029] In the current spectrum graph, the horizontal axis represents time, the vertical axis represents frequency, and the grayscale value represents energy intensity.

[0030] S104: Using a trained correlation neural network, based on m+n scorch images and current spectrum diagrams, detect the relationship between current and arc energy shown in the scorch images, and obtain m+n-1 arc correlation feature vectors.

[0031] Using the above method, the scorching image can reflect the energy of the electric arc, and the electric arc extinguishes at the zero crossing point, so the contact surface will not be scorched.

[0032] S105: According to the time points from morning to night, input m+n-1 arc-related feature vectors into the temporal convolutional network (TCN) for prediction to obtain the zero-crossing points.

[0033] The zero-crossing point refers to the time point at which the current crosses zero in the predicted future time period.

[0034] The zero-crossing point refers to the time point at which no electric arc will occur. The circuit breaker closes at the predicted zero-crossing point and performs an arc-extinguishing operation.

[0035] Optionally, the associated neural network includes a first convolutional network, a second convolutional network, and a fusion neural network; The first convolutional network contains 2*2 convolutional kernels.

[0036] The first convolutional network is a convolutional neural network (CNN).

[0037] The second convolutional network contains 2*2*2 three-dimensional convolutional kernels.

[0038] The second convolutional network is a three-dimensional convolutional neural network (3D CNN).

[0039] Optionally, the step involves using a trained correlation neural network to detect the relationship between the current and the arc energy shown in the burning images, based on m+n ablation images and current spectrum diagrams, to obtain m+n-1 arc-related feature vectors, including: Based on m+n ablation images, the gray value changes at adjacent time points are detected to obtain m+n-1 ablation difference images and arc change vectors.

[0040] With a stride of 1, the scorching difference image is convolved with a 2*2 convolution kernel in the first convolutional network to detect the scorching state of the scorching difference image and obtain a first scorching feature map; the scorching feature map represents the characteristics of the contact surface changes as the electric arc burns.

[0041] m+n-1 ablation difference images correspond to m+n-1 first ablation feature maps; Based on the current spectrum diagram, the current change state is detected to obtain m+n-1 current spectrum feature diagrams; Based on the arc change vector, m+n-1 first burning feature maps and m+n-1 current spectrum feature maps, an association relationship is constructed to obtain the arc association feature vector; the arc association feature vector represents the association feature between the area and color change of the contact burning and the current change.

[0042] Optionally, the step of detecting the current change state based on the current spectrum to obtain m+n-1 current spectrum feature maps includes: The current spectrum is divided into m+n time current spectrums centered on the time point.

[0043] The current spectrum represents the change in current at consecutive time points. The current spectrum is divided into time segments centered on each time point, with a window representing the time length. The window is smaller than the length between two time points.

[0044] The time-current spectrum diagrams corresponding to two adjacent time points are superimposed to obtain a superimposed three-dimensional spectrum diagram.

[0045] The two time-current spectrum diagrams are superimposed by superimposing them with the corresponding time point as the center, and zeros are padded at both ends to make them the same size.

[0046] With a stride of 1, the superimposed three-dimensional spectrum map is convolved with a 2*2*2 three-dimensional convolution kernel in the second convolutional network to obtain a current spectrum feature map.

[0047] The m+n time-current change spectrum diagrams correspond to the m+n-1 current spectrum feature diagram.

[0048] Optionally, the step of constructing an association relationship based on the arc change vector, m+n-1 first scorching feature maps, and m+n-1 current spectrum feature maps to obtain an arc association feature vector includes: The first scorching feature map, the corresponding current spectrum feature map, and the scorching depth and scorching area values ​​in the corresponding arc change vector are input into the fusion neural network to obtain the arc-related feature vector.

[0049] The fusion neural network is a deep neural network (DNN).

[0050] m+n-1 current feature maps correspond to m+n-1 arc-related feature vectors; The fusion neural network comprises an input layer, a hidden layer, and an output layer; the input layer comprises input neurons; the hidden layer comprises hidden neurons; and the output layer comprises output neurons.

[0051] The elements in the electric arc change vector are used as the input to the input layer.

[0052] The sum of the number of elements in the first ablation feature map and the number of elements in the current spectrum feature map plus 2 is used as the correlation number.

[0053] The addition of 2 is because the number of scorching depth and scorching area values ​​in the arc change vector is 2.

[0054] The number of associated input neurons corresponds to one hidden neuron, and one input neuron corresponds to only one hidden neuron; the output of the output layer is the arc change feature vector.

[0055] Optionally, the step of detecting grayscale value changes at adjacent time points based on m+n ablation images to obtain m+n-1 ablation difference images and arc change vectors includes: Obtain the initial time point; the initial time point is a time point earlier than other time points.

[0056] The burn image at the initial time point is used as the background image.

[0057] Based on the time points from morning to night, multiple ablation images are compared with the background image to calculate the ablation similarity value.

[0058] There are m+n scorching similarity values.

[0059] The Euclidean distance algorithm is used to calculate the similarity of gray values ​​at corresponding positions in the scorched image and the background image.

[0060] If the scorching similarity value is less than or equal to the similarity threshold, set all m-1 scorching difference images corresponding to m scorching images to 0.

[0061] In this embodiment, the similarity threshold is 0.95.

[0062] If the scorching similarity value is greater than the similarity threshold, edge detection is performed on the n scorching images to obtain n scorching positions; the gray value of the scorching position is subtracted from the gray value of the corresponding position in the background image to obtain the scorching gray value.

[0063] In this embodiment, the Canny operator is used for edge detection.

[0064] For example, if the scorching position is [7,10], then the gray value corresponding to [7,10] in the scorched image is subtracted from the gray value corresponding to [7,10] in the background image to obtain the scorching gray value.

[0065] Based on n scorching locations, n scorching gray values, and the corresponding n scorching images, n scorching difference images are obtained; the scorching difference images represent the changes in the contact scorching state caused by the electric arc at two adjacent time points.

[0066] Based on m+n scorching positions and m+n scorching gray values, an arc change vector is obtained; the arc change vector contains (m+n-1)*2 elements.

[0067] Where m and n are positive integers. Since the degree of burning will be superimposed if the contacts are not replaced, a burning similarity value is used to divide multiple burning images into two parts, m and n.

[0068] Optionally, the step of obtaining n scorching difference images based on n scorching locations, n scorching grayscale values, and the corresponding n scorching images includes: The ablation grayscale values ​​are marked at the corresponding ablation positions to obtain the ablation image to be judged; the values ​​of the ablation image to be judged are all 0 except for the ablation positions.

[0069] Wherein, the size of the ablation image to be judged is equal to the size of the ablation image.

[0070] The above method removes the influence of the background image.

[0071] The earliest time point among the n ablation images to be judged is obtained as one ablation difference image.

[0072] Since the first scorched image among the n scorched images needs to be compared with the last scorched image among the m scorched images, and the weighted image among the m scorched images is the image similar to the background image, that is, the image that has not been scorched due to the electric arc, the scorched image to be judged that is affected by the background image is taken as the scorched difference image between the two.

[0073] Subtract the corresponding gray value in the ablation image at the previous time point from the gray value of the ablation image at the next time point to obtain n-1 ablation difference images.

[0074] Subtraction is performed according to the subscript.

[0075] Optionally, obtaining the arc change vector based on m+n scorching locations and m+n scorching gray values ​​includes: If the scorching similarity value is less than or equal to the similarity threshold, set the m-1 scorching depth values ​​and m-1 scorching area values ​​corresponding to the m scorching images to 0.

[0076] If the scorching similarity value is greater than the similarity threshold, the number of scorching locations at the next time point is subtracted from the number of scorching locations at the previous time point to obtain n scorching area values.

[0077] The number of scorched locations at the first time point in the n scorched images is subtracted from the number of scorched locations at the last time point in the m scorched images.

[0078] The scorching ash value at the same scorching location at the next time point as at the previous time point is taken as the scorching depth value.

[0079] Based on the time points from morning to night, construct the arc change vector by taking m+n-1 scorching depth values ​​and m+n-1 scorching area values.

[0080] Specifically, if the first scorching image does not produce an electric arc, meaning the scorching image remains unchanged (i.e., both the scorching depth and scorching area values ​​are 0), then the first and second values ​​in the arc change vector are both set to 0. If the second scorching image does not produce an electric arc, meaning the scorching image remains unchanged (i.e., both the scorching depth and scorching area values ​​are 0), then the third and fourth values ​​in the arc change vector are both set to 0. If the third scorching image produces an electric arc, meaning the scorching image changes, then the fifth and sixth values ​​in the arc change vector are set to the calculated scorching depth and scorching area values.

[0081] Optionally, the method for training the association neural network includes: Acquire training current data, corresponding training burn images, and marked zero-crossing points at historical time points; the training current data and training burn images represent the data used for training; the marked zero-crossing points represent the marked zero-crossing points at the next time point after the historical time point used for training.

[0082] Among them, the time point corresponding to the zero point is later than the time point used for training.

[0083] Based on the ablation images at historical time points, the predicted zero-crossing points are obtained through correlation neural networks and temporal convolutional networks.

[0084] The loss is calculated by comparing the predicted zero-crossing points with the labeled zero-crossing points, and then used to train the correlation neural network and the temporal convolutional network.

[0085] In this embodiment, the cross-entropy loss function is used to calculate the loss. Example

[0086] Based on the above-mentioned zero-crossing interruption control method based on arc energy transfer prediction, this embodiment of the invention also provides a zero-crossing interruption control system based on arc energy transfer prediction, the system including an acquisition module, a Fourier transform module, an association module, and a prediction module.

[0087] The acquisition module is used to collect current data and scorch images at m+n time points; the current data represents the current waveform passing through the contact; the scorch image represents the image of the contact surface; and a trained correlation neural network is acquired; the correlation neural network can detect the damage to the contact caused by the arc when different currents are disconnected.

[0088] The Fourier transform module is used to convert current data into a current spectrum graph through Fourier transform. The correlation module is used to detect the relationship between the current and the arc energy shown in the burning images based on m+n burning images and current spectrum diagrams through a trained correlation neural network, and obtain m+n-1 arc correlation feature vectors. The prediction module is used to input m+n-1 arc-related feature vectors into the temporal convolutional network according to the time points from morning to night, and to make predictions to obtain the zero-crossing points. Example

[0089] Based on the aforementioned zero-crossing interruption control method based on arc energy transfer prediction, this embodiment of the invention also provides an application of the zero-crossing interruption control method based on arc energy transfer prediction in a ring main unit. By predicting the zero-crossing point, the arc extinguishing operation is actively performed when the predicted zero-crossing point is reached in the ring main unit at the time corresponding to the zero-crossing point.

[0090] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0091] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0092] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. A zero-crossing interruption control method based on arc energy transfer prediction, characterized in that, include: Collect current data and scorch images at m+n time points; the current data represents the current waveform passing through the contact; the scorch image represents an image of the contact surface; A trained correlation neural network is obtained; the correlation neural network is capable of detecting the damage to the contacts caused by the arc when different currents are disconnected. The current data is converted into a current spectrum using Fourier transform. By using a trained neural network, based on m+n scorching images and current spectrum diagrams, the relationship between current and arc energy shown in the scorching images is detected, resulting in m+n-1 arc-related feature vectors. According to the time points from morning to night, m+n-1 arc-related feature vectors are input into a temporal convolutional network for prediction to obtain the zero-crossing points.

2. The zero-crossing interruption control method based on arc energy transfer prediction according to claim 1, characterized in that, The associated neural network includes a first convolutional network, a second convolutional network, and a fusion neural network; The first convolutional network contains 2*2 convolutional kernels; The second convolutional network contains 2*2*2 three-dimensional convolutional kernels.

3. The zero-crossing interruption control method based on arc energy transfer prediction according to claim 1, characterized in that, The process involves using a trained neural network to detect the relationship between the current and the arc energy shown in the burning images, based on m+n ablation images and current spectrum diagrams, to obtain m+n-1 arc-related feature vectors, including: Based on m+n ablation images, the gray value changes at adjacent time points are detected to obtain m+n-1 ablation difference images and arc change vectors. With a stride of 1, the scorching difference image is convolved with a 2*2 convolution kernel in the first convolutional network to detect the scorching state of the scorching difference image and obtain a first scorching feature map; the scorching feature map represents the characteristics of the contact surface changes as the electric arc burns. m+n-1 ablation difference images correspond to m+n-1 first ablation feature maps; Based on the current spectrum diagram, the current change state is detected to obtain m+n-1 current spectrum feature diagrams; Based on the arc change vector, m+n-1 first burning feature maps and m+n-1 current spectrum feature maps, an association relationship is constructed to obtain the arc association feature vector; the arc association feature vector represents the association feature between the area and color change of the contact burning and the current change.

4. The zero-crossing interruption control method based on arc energy transfer prediction according to claim 3, characterized in that, Based on the current spectrum, the current change state is detected to obtain m+n-1 current spectrum feature maps, including: The current spectrum is divided according to time points to obtain m+n time-current change spectrum diagrams; The time-current spectrum diagrams corresponding to two adjacent time points are superimposed to obtain a superimposed three-dimensional spectrum diagram. With a stride of 1, the superimposed three-dimensional spectrum map is convolved with a 2*2*2 three-dimensional convolution kernel in the second convolutional network to obtain a current spectrum feature map. The m+n time-current change spectrum diagrams correspond to the m+n-1 current spectrum feature diagram.

5. The zero-crossing interruption control method based on arc energy transfer prediction according to claim 3, characterized in that, The process involves constructing a correlation based on the arc change vector, m+n-1 first scorching feature maps, and m+n-1 current spectrum feature maps to obtain an arc correlation feature vector, including: The first scorching feature map, the corresponding current spectrum feature map, and the scorching depth and scorching area values ​​in the corresponding arc change vector are input into the fusion neural network to obtain the arc-related feature vector. m+n-1 current feature maps correspond to m+n-1 arc-related feature vectors; The fusion neural network comprises an input layer, a hidden layer, and an output layer; the input layer comprises input neurons; the hidden layer comprises hidden neurons; and the output layer comprises output neurons. Use the elements of the electric arc change vector as the input to the input layer; The sum of the number of elements in the first ablation feature map and the number of elements in the current spectrum feature map plus 2 is used as the correlation number; The number of associated input neurons corresponds to one hidden neuron, and one input neuron corresponds to only one hidden neuron; the output of the output layer is the arc change feature vector.

6. The zero-crossing interruption control method based on arc energy transfer prediction according to claim 3, characterized in that, The method involves detecting grayscale value changes at adjacent time points based on m+n ablation images to obtain m+n-1 ablation difference images and an arc change vector, including: Obtain the initial time point; the initial time point is a time point earlier than other time points. The ablation image at the initial time point is used as the background image; Based on the time points from morning to night, multiple ablation images are compared with the background image to obtain ablation similarity value; If the scorching similarity value is less than or equal to the similarity threshold, set all m-1 scorching difference images corresponding to m scorching images to 0; If the scorching similarity value is greater than the similarity threshold, edge detection is performed on the n scorching images to obtain n scorching positions; the gray value of the scorching position is subtracted from the gray value of the corresponding position in the background image to obtain the scorching gray value; Based on n scorching locations, n scorching gray values, and the corresponding n scorching images, n scorching difference images are obtained; the scorching difference images represent the changes in the contact scorching state caused by the electric arc at two adjacent time points. Based on m+n scorching positions and m+n scorching gray values, an arc change vector is obtained; the arc change vector contains (m+n-1)*2 elements.

7. The zero-crossing interruption control method based on arc energy transfer prediction according to claim 6, characterized in that, The process of obtaining n scorching difference images based on n scorching locations, n scorching grayscale values, and the corresponding n scorching images includes: The ablation grayscale values ​​are marked at the corresponding ablation positions to obtain the ablation image to be judged; the values ​​of the ablation image to be judged are all 0 except for the ablation positions; The earliest time point among the n ablation images to be judged is taken as one ablation difference image; Subtract the corresponding gray value in the ablation image at the previous time point from the gray value of the ablation image at the next time point to obtain n-1 ablation difference images.

8. The zero-crossing interruption control method based on arc energy transfer prediction according to claim 1, characterized in that, The method for obtaining the arc change vector based on m+n scorching positions and m+n scorching gray values ​​includes: If the scorching similarity value is less than or equal to the similarity threshold, set the n-1 scorching depth values ​​and n-1 scorching area values ​​corresponding to the n scorching images to 0; If the scorching similarity value is greater than the similarity threshold, the number of scorching locations at the next time point is subtracted from the number of scorching locations at the previous time point to obtain m scorching area values. The scorching ash value at the same scorching position at the next time point as the previous time point is taken as the scorching depth value. Based on the time points from morning to night, construct the arc change vector by taking m+n-1 scorching depth values ​​and m+n-1 scorching area values.

9. A zero-crossing interruption control system based on arc energy transfer prediction, characterized in that, include: The acquisition module is used to collect current data and scorch images at m+n time points; the current data represents the current waveform passing through the contact; the scorch image represents the image of the contact surface; and a trained correlation neural network is acquired; the correlation neural network can detect the damage to the contact caused by the arc when different currents are disconnected. The Fourier transform module is used to convert current data into a current spectrum graph through Fourier transform. The correlation module is used to detect the relationship between the current and the arc energy shown in the burning images based on m+n burning images and current spectrum diagrams through a trained correlation neural network, and obtain m+n-1 arc correlation feature vectors. The prediction module is used to input m+n-1 arc-related feature vectors into the temporal convolutional network according to the time points from morning to night, and to make predictions to obtain the zero-crossing points.

10. The application of the system according to claim 9 in a ring main unit, characterized in that, The ring main unit is configured with the system described in claim 9.

Images