ViT and GRU fusion-based one-key sequence control disconnecting link state identification method and system

Abstract

A ViT and GRU fusion-based one-key sequence control disconnecting link state identification method and system comprises the following steps: step 1, acquiring a disconnecting link monitoring video, and extracting a continuous frame image sequence from the disconnecting link monitoring video; 2, performing feature extraction on the multi-frame image sequence through a ViT model to obtain a frame-level feature vector, and performing action clustering and stage division on the frame-level feature vector to obtain an enhanced frame feature sequence; step 3, inputting the enhanced feature sequence into the gating circulation unit network to obtain a final time sequence feature; and step 4, inputting the final time sequence characteristics into a full-connection classification layer, processing the final time sequence characteristics through a normalized exponential function, and outputting the state category of the disconnecting link. According to the method, adaptive weighted fusion is carried out on the features at all moments through an attention weighting mechanism, rule constraint terms are introduced, abnormal weight suppression is carried out, and the time sequence sensing ability and robustness of the model can be remarkably improved while the detection precision is ensured.

Description

Technical Field

[0001] This invention relates to the technical field of disconnector status recognition, and more specifically, to a one-click sequential control disconnector status recognition method and system based on the integration of ViT and GRU. Background Technology

[0002] With the continuous improvement of the power grid's intelligence level, the automation and unmanned operation of substations are gradually becoming a trend. In substation operation, the open / closed status of disconnecting switches (also known as knife switches) is one of the important factors affecting the safe operation of equipment. To improve operational efficiency and safety, power grid dispatching systems have widely adopted "one-click sequential control" (i.e., one-click sequential control) technology to realize automatic sequential control and status monitoring of equipment such as circuit breakers, disconnecting switches, and grounding switches.

[0003] During the "one-click sequential control" operation, the system will automatically execute a series of opening and closing actions according to the preset operation ticket. To ensure the safety of the operation, the actual status of the disconnector (knife switch) needs to be double-confirmed: 1. Initial confirmation is achieved by the control system determining the open / closed state based on equipment signals (such as switch position contact quantity); 2. Secondary confirmation is usually achieved by using video image recognition or visual analysis to identify and verify the physical position and status of the disconnect switch on site, in order to prevent misjudgment caused by signal mistransmission or equipment malfunction.

[0004] Existing one-click sequential control switch identification methods typically employ the following process: After the dispatch platform issues a disconnector operation command, the video master station controls the front-end camera to capture images based on the disconnector's location information. The video host or algorithm server processes the captured disconnector images, identifies the actual open or closed state of the disconnector, and returns the identification result to the dispatch system. The dispatch platform compares the returned result with a primary signal. If they match, the disconnector status is considered successfully confirmed; otherwise, an alarm is triggered or the sequential control operation is terminated.

[0005] For identifying the open and closed states of disconnectors, the most commonly used disconnector identification methods in the industry currently include: State recognition based on traditional image processing (such as edge detection and template matching); Visual recognition based on deep learning detection networks (such as YOLO and CenterNet); Based on multi-data fusion.

[0006] The shortcomings of existing technology include at least the following: Traditional single-frame image recognition models (such as Convolutional Neural Networks (CNNs) and Visual Transformers (ViTs) can extract spatial features of devices from static images. However, in practical applications, the opening and closing processes of disconnectors exhibit significant temporal continuity, making it difficult for a single frame image to accurately reflect the switch's movement. When camera angles, lighting conditions, or occlusion factors change, single-frame recognition models are prone to misjudgments or unstable identification. Furthermore, they lack effective modeling of inter-frame temporal relationships during the dynamic switching process of disconnectors from "open" to "closed" or vice versa, resulting in low accuracy in identifying "false opening" and "false closing" states. For example, using a single frame for state recognition leads to low accuracy in cases where the opening or closing is incomplete.

[0007] While some studies have attempted to improve the accuracy of disconnector status judgment by constructing sequence models to predict action trends, these methods still have limitations in terms of feature representation accuracy, computational complexity, or transferability, making them unsuitable for scenarios with low computing power, such as edge devices. For example, existing technologies use information such as angle, speed, and time to model and predict disconnector trends, but this method is relatively complex, and due to the different aging levels of different devices, the device modeling needs to be modified, resulting in significant differences in the accuracy of the method when used on different devices. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a one-click sequential control disconnector status recognition method based on the fusion of a visual Transformer and a gated loop unit. This method solves the problem that traditional single-frame recognition methods cannot effectively characterize the continuity of the disconnector's motion process. It can balance spatial representation capabilities and temporal dependency modeling capabilities, and can perform disconnector status recognition under low computing power, thereby improving the stability and accuracy of recognition in complex environments.

[0009] The present invention adopts the following technical solution.

[0010] The beneficial effect of this invention is that, compared with the prior art, this invention proposes a multi-frame temporal recognition method based on joint modeling of ViT and GRU. This method uses continuous multi-frame images in a video segment as input, extracts the spatial feature information of each frame through a visual Transformer (ViT), and then introduces a gated recurrent unit (GRU) to model the temporal dimension, thereby capturing the dynamic evolution characteristics of the disconnector in the process of state changes such as opening, closing, false opening, and false closing. To enhance the perception of disconnector movement, this invention addresses the distinct stages of disconnector movement. After the ViT-encoded embedding output, feature clustering is used to divide the movement into initial, middle, or final stages. An additional movement stage embedding vector is added, giving the model stronger discriminative ability for different temporal stages. Simultaneously, GRU units are used to model the correlation between features across consecutive frames, solving the problem of insufficient sensitivity to time series in traditional Transformers. This continuity feature is not simply based on temporal continuity but rather on the stages of device movement, significantly improving the model's accuracy in recognizing disconnector movement trends.

[0011] To address the issues of significant differences in key frames and redundancy in non-key frame information during the operation of the disconnector, this invention further introduces an attention-weighted mechanism to adaptively weight and fuse features at each time step. Since the disconnector's movement is controlled by commands, in real-world scenarios, the next command is executed sequentially only after the previous one is completed. Therefore, the disconnector will not perform multiple repetitive actions in a short period. The attention mechanism introduces rule constraints to suppress abnormal weights and improve the stability of the results. Through this temporal attention mechanism, the model can automatically identify important frames during the disconnector's operation, thereby enhancing the response to key temporal features and reducing the interference of static frames on the final classification results.

[0012] This method employs a lightweight model structure with ample feature representation, enabling multi-frame temporal dependency modeling without relying on external time-series labels. In practical applications, it can be directly deployed in substation video monitoring systems for real-time identification and alarming of the open / closed status and abnormal states (false open, false close) of disconnectors. Unlike existing technologies that only distinguish between open and closed states, this invention, for the first time, introduces "false open" and "false close" states into the time-series model to address abnormal situations encountered in actual operation and maintenance. It improves the ability to identify abnormal states by jointly judging action integrity and temporal consistency.

[0013] Through the above structural design, the present invention significantly improves the model's temporal awareness and robustness while ensuring detection accuracy, making it suitable for high-security power equipment condition monitoring tasks. Attached Figure Description

[0014] Figure 1 This is a flowchart of the one-click sequential control switch status recognition method in this invention; Figure 2 This is a structural diagram of the one-button sequential control switch status recognition system of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0016] like Figure 1 As shown, this invention proposes a one-click sequential control switch status recognition method based on the fusion of ViT (Vision Transformer) and GRU (Gated Recurrent Unit). The method includes the following steps: Step 1: Acquire the monitoring video of the disconnect switch and extract a continuous sequence of frame images from the monitoring video of the disconnect switch; Acquire the monitoring video of the disconnect switch, extract multiple consecutive frames from the video, and normalize the size of each frame to obtain the processed frame images X. t =[B,T,C,H,W], where B is the batch size of the image, C is the number of channels per frame, and H and W are the height and width of the frame image, respectively; Based on the processed images, a continuous sequence of image frames X is obtained, where X = [X1, X2, ..., X...]. T ], where T is the total number of frames. Let represent the i-th frame image, t∈[1,T].

[0017] Step 2: Extract features from the multi-frame image sequence using the ViT model to obtain frame-level feature vectors, and perform action clustering and stage division on the frame-level feature vectors to obtain an enhanced frame feature sequence; The pre-trained Visual Transformer (ViT) model is used to perform image patch embedding and feature extraction on each frame of the image to obtain frame-level feature vectors. :

[0018] in, Let represent the i-th frame of the image, and T be the total number of frames.

[0019] The Patch Embedding layer of the Visual Transformer is implemented using convolutional projection, which divides the input image into fixed-size image patches and obtains feature vectors through linear mapping.

[0020] Meanwhile, the action of the disconnect switch can be viewed as a type of action with phased motion characteristics (initial state - intermediate state - stable state). To enhance the perception of the disconnect switch's action, an action phase vector is introduced into the feature encoding, discretizing the disconnect switch's motion process into three action phases, and assigning a phase vector to each action phase. The corresponding relationships can be distinguished through feature difference clustering. Unlike traditional division methods based on time ratio or other thresholds, this invention introduces an unsupervised clustering mechanism based on feature similarity.

[0021] Using frame-level feature vectors The method calculates the average feature vectors of the first n frames and the last n frames (n can be adjusted according to the actual characteristics of the switch) as prior representations of the initial and final states. Then, it performs cluster analysis on the feature vector set of the entire action sequence, adaptively dividing it into three state spaces corresponding to the initial stage, intermediate transition stage, and final stage. This method can automatically determine the stage boundaries based on actual motion characteristics, thus effectively improving the robustness and generalization ability of stage division. This method does not require manual annotation of stage boundaries and has good adaptability and transferability.

[0022] Constructing endpoint prior features: Calculate the average eigenvector of the first n frames of the sequence in the frame-level eigenvector. The average eigenvector of the sequence after n frames :

[0023]

[0024] Among them, T is the total number of frames. This represents the frame-level feature vector corresponding to the i-th frame, and the size of n is set by those skilled in the art according to the actual application scenario.

[0025] Based on the average feature vector of the first n frames and the average eigenvector of the next n frames Perform action phase clustering, including the start phase, intermediate phase, and termination phase, to obtain the clustering results. :

[0026] in, , , These represent the initial stage class, the intermediate stage class, and the final stage class, respectively. Among them, semantics are automatically mapped based on the distance relationship between cluster centers and anchor points: with the average feature vector of the previous n frames. The frame with the smallest distance is classified as the initial stage, and its average feature vector is compared with that of the next n frames. The frame image with the smallest distance is classified as the termination stage, and the remaining frame images are classified as the intermediate stage; To enhance the perception of changes in motion, the difference vectors of n frames are calculated. (These n terms are consistent with the average of the features at the start and end of the calculation, spanning n frames), and are concatenated with the current frame vector to obtain the concatenated vector. :

[0027]

[0028] in, This represents the frame-level feature vectors corresponding to frame t and frame (t+n). Stage embedding vector construction: Assign an action phase embedding vector to each type of action phase. And it is fused with frame features to obtain enhanced frame features. :

[0029] in, The concatenation of representation vectors yields enhanced frame features that integrate spatial information, motion changes, and semantic information about the action phase. This serves as the input for subsequent GRU time series modeling; Among them, the action phase embedding vector The construction includes: For the t-th frame image, for the label of the action phase Take the first step from the stage parameter matrix. Line, to obtain the embedding vector :

[0030] Among them, matrix K represents the number of categories in the action phase. Embedded dimensions for stages.

[0031] By using ViT feature-based clustering adaptive partitioning, this invention eliminates the need for manual annotation of action stage boundaries and can automatically identify the start, transition, and termination stages based on the actual action characteristics of the disconnector. Combined with action stage embedding vectors, the model can explicitly perceive the action evolution process, improving the accuracy and robustness of identifying abnormal states such as opening, closing, false opening, and false closing, making it particularly suitable for fixed monitoring scenarios and high-security power equipment condition monitoring tasks.

[0032] In a preferred embodiment, the present invention processes the enhanced frame features using a temporal attention mechanism through an optimization module, weights key frames, and obtains the processed enhanced frame features as input to the GRU network; specifically as follows: For certain complex scenarios, abnormal frames can be suppressed and optional modules can be optimized. A temporal attention mechanism can be added to process the enhanced frame features before inputting them into the GRU, and key frames can be weighted to improve the accuracy of knife switch recognition.

[0033] Introducing prior knowledge—minimum action interval and control command constraints—we adjust the attention weights of frames as follows: Calculate the attention weights for frame t. as follows:

[0034]

[0035] Where T is the total number of frames. Here, is the attention weight parameter, used to learn the importance score for each frame; exp(·) is the exponential function, and R(t) is the rule constraint function. For intermediate parameters; In the calculation formula This represents the summation of weights over all frames, i.e., the normalization factor; By combining attention weights and enhanced frame features, the enhanced frame features of the optimized switch in frame t are obtained. :

[0036] The optimized enhanced features are used as input to the gated recurrent unit network.

[0037] The specific rules of the rule constraint function are: 1) Control instruction minimum interval constraint to suppress short-term repeated action frames; 2) Embed information in the action stage to reduce the weight of non-critical stage frames.

[0038] After adding the attention module, the features input to the GRU are:

[0039] This reduces interference from abnormal frames and improves the stability of motion recognition.

[0040] Step 3, GRU temporal feature modeling steps; The enhanced feature sequence output from step 2 Or optimized enhanced frame features Input the gated recurrent unit (GRU) network, extract the inter-frame temporal dependencies, and obtain the GRU output. :

[0041] The gated cyclic unit network includes update gates and reset gates, and its state update formula is:

[0042]

[0043]

[0044] Where t is the sequence number of the current video frame, for example, the t-th frame image; This represents the enhanced frame feature of the switch in frame t. This represents the system's memory of the dynamic state of the disconnector up to frame t-1, for example, whether it is transitioning from closed to open. , , The linear transformation parameters corresponding to the update gate, reset gate, and candidate state are used to learn the temporal relationship; σ(·) compresses the input to (0,1), representing the degree of opening of the "gate"; tanh(·) is the hyperbolic tangent activation function, used to enable nonlinear fusion of information; This indicates the ratio of current frame information to historical memory. The larger the value, the more likely it is to update to the new state; if A smaller value indicates a greater tendency to preserve historical states; Indicates whether past memories have been forgotten. If the value is less than the preset threshold, it means that the state of the previous frame has been forgotten and the judgment is made again based only on the current frame. This indicates that at frame t, the system integrates the visual features of the current frame with the historical changes in the switch's motion state.

[0045] Step 4, state classification; The final temporal features are input into a fully connected classification layer, and the softmax function outputs the state category of the switch:

[0046] The obtained status categories include: open, closed, false open (i.e., not fully open) or false closed (i.e., not fully closed).

[0047] like Figure 2 As shown, the present invention also proposes a one-click sequential control switch status recognition system based on ViT and GRU fusion, which is used to realize the above-mentioned one-click sequential control switch status recognition method based on ViT and GRU fusion. The system includes a video acquisition module, a frame extraction and preprocessing module, a ViT+GRU timing recognition module and a status output module. The video acquisition module is used to acquire monitoring video of the disconnect switch; The frame extraction and preprocessing module is used to extract a continuous sequence of frame images from the switch monitoring video; The temporal recognition module is used to extract features from the frame image sequence using the ViT model to obtain frame-level feature vectors, perform action clustering and stage division on the frame-level feature vectors to obtain an enhanced frame feature sequence, and input the enhanced feature sequence into a gated recurrent unit network to obtain the final temporal features. The state classification and output module is used to input the final time-series features into the fully connected classification layer, and output the state category of the disconnector after processing by the normalized exponential function.

[0048] The beneficial effects of this invention are that, compared with existing technologies, it balances spatial feature extraction and temporal dependency modeling. By fusing ViT spatial features, frame difference features, and action stage embedded vector input GRU for temporal modeling, it significantly enhances the model's ability to perceive the evolution trend of disconnector actions. It incorporates the sequential relationship of disconnector actions into the features, improving the accuracy of disconnector state recognition; it can effectively identify "false separation" and "false closing" during disconnector actions, enhancing robustness; and it introduces selectable optimization modules to adapt to complex backgrounds, lighting, and occlusion environments in substation scenarios. The model can automatically focus on the action start and end keyframes, making it suitable for various substation monitoring scenarios; it can also be extended to other power equipment action recognition tasks.

[0049] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0050] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0051] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0052] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A one-click sequential control switch status recognition method based on ViT and GRU fusion, characterized in that, Includes the following steps: Step 1: Acquire the monitoring video of the disconnect switch and extract a continuous sequence of frame images from the monitoring video of the disconnect switch; Step 2: Extract features from the frame image sequence using the ViT model to obtain frame-level feature vectors, and perform action clustering and stage division on the frame-level feature vectors to obtain enhanced frame feature sequences; Step 3: Input the enhanced feature sequence into the gated recurrent unit network to obtain the final temporal features; Step 4: Input the final time-series features into the fully connected classification layer, and process them through the normalized exponential function to output the state category of the disconnector.

2. The one-click sequential control switch status recognition method based on ViT and GRU fusion according to claim 1, characterized in that, Step 1 specifically includes: Extract consecutive multi-frame images from the switch monitoring video, and normalize the size of each frame to obtain the processed frame images X. t =[B,T,C,H,W], where B is the batch size of the image, C is the number of channels per frame, and H and W are the height and width of the frame image, respectively; Based on the processed images, a continuous sequence of image frames X is obtained, where X = [X1, X2, ..., X...]. T ], where T is the total number of frames, and t∈[1,T].

3. The one-click sequential control switch status identification method based on ViT and GRU fusion according to claim 1, characterized in that, Step 2 specifically includes: The ViT model is used to extract features and embed image patches from each frame in the frame image sequence to obtain the t-th frame image. Corresponding frame-level feature vector : in, (·) indicates the ViT model processing.

4. The one-click sequential control switch status identification method based on ViT and GRU fusion according to claim 1, characterized in that, The action clustering and stage division of frame-level feature vectors specifically includes: The representational priors of the initial and final states are obtained based on the frame-level feature vectors; Based on the representation prior, cluster analysis is performed on the feature vector set of the entire action sequence, and it is adaptively divided into three state spaces, corresponding to the initial stage, intermediate transition stage and termination stage, respectively.

5. The one-click sequential control switch status identification method based on ViT and GRU fusion according to claim 4, characterized in that, The process of obtaining the representational priors of the initial and final states based on frame-level feature vectors specifically includes: Calculate the average eigenvector of the first n frames in the frame-level eigenvector. and the average eigenvector of the next n frames : Among them, T is the total number of frames. This represents the frame-level feature vector corresponding to the i-th frame.

6. The one-click sequential control switch status identification method based on ViT and GRU fusion according to claim 5, characterized in that, The clustering analysis of the feature vector set of the entire action sequence based on representational priors specifically includes: Based on the average feature vector of the first n frames and the average eigenvector of the next n frames Perform action phase clustering, including the start phase, intermediate phase, and termination phase, to obtain the clustering results. : in, , , These represent the initial stage class, the intermediate stage class, and the final stage class, respectively. Among them, semantics are automatically mapped based on the distance relationship between cluster centers and anchor points: with the average feature vector of the previous n frames. The frame with the smallest distance is classified as the initial stage, and its average feature vector is compared with that of the next n frames. The frame image with the smallest distance is classified as the termination stage, and the remaining frame images are classified as the intermediate stage; Calculate the difference vector of n frames It is then concatenated with the current frame vector to obtain the concatenated vector. : in, This represents the frame-level feature vectors corresponding to frame t and frame (t+n). Assign an action phase embedding vector to each type of action phase. And it is fused with frame features to obtain enhanced frame features. : in, The concatenation of representation vectors yields enhanced frame features that integrate spatial information, motion changes, and semantic information about the action phase. This serves as the input for subsequent GRU time series modeling; Among them, the action phase embedding vector The construction includes: For the t-th frame image, for the label of the action phase Take the first step from the stage parameter matrix. Line, to obtain the embedding vector : Among them, matrix K represents the number of categories in the action phase. Embedded dimensions for stages.

7. The one-click sequential control switch status identification method based on ViT and GRU fusion according to claim 6, characterized in that, Also includes: Calculate the attention weights for frame t. as follows: Where T is the total number of frames. Let be the attention weight parameter, exp(·) be the exponential function, and R(t) be the rule constraint function. For intermediate parameters; By combining attention weights and enhanced frame features, the optimized enhanced frame features are obtained. : The optimized enhanced features are used as input to the gated recurrent unit network.

8. A one-click sequential control switch status recognition system based on ViT and GRU fusion, used to implement the one-click sequential control switch status recognition method based on ViT and GRU fusion as described in any one of claims 1-7, characterized in that, include: The video acquisition module is used to acquire monitoring video of the disconnect switch; The frame extraction and preprocessing module is used to extract a continuous sequence of frame images from the switch monitoring video. The temporal recognition module is used to extract features from the frame image sequence using the ViT model to obtain frame-level feature vectors, and to perform action clustering and stage division on the frame-level feature vectors to obtain an enhanced frame feature sequence. The enhanced feature sequence is then input into a gated recurrent unit network to obtain the final temporal features; The state classification and output module is used to input the final time-series features into the fully connected classification layer, and output the state category of the disconnector after processing by the normalized exponential function.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the one-click sequential control switch status identification method based on ViT and GRU fusion as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the one-click sequential control switch status recognition method based on the fusion of ViT and GRU as described in any one of claims 1-7.

Images