Controller monitoring protection diagnosis control method and system
Through video acquisition equipment and generator identification model, the longitudinal and lateral vibrations in the generator set are analyzed and marked, which solves the problem of poor accessories connection in the prior art, improves production efficiency and reduces costs.
Patent Information
- Application Number
- CN202411903129.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2025-03-18
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the accessories cannot be well connected, resulting in irregular layout, cumbersome wiring and low production efficiency.
The reference video stream of longitudinal and lateral vibrations in the generator set is collected through the video acquisition equipment, and the longitudinal vibration frame and lateral vibration frame are analyzed using the pre-trained generator identification model, and the vibration conditions of each generator are identified and marked to achieve effective monitoring and diagnosis of each accessories in the generator set.
It effectively solves the problem of poor connection between various accessories, improves production efficiency and reduces production costs.
Smart Images
Figure CN119645008A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of monitoring and protection of generator sets, and in particular to a controller monitoring protection diagnosis control method and system. Background Art
[0002] The controller of the generator set can monitor the various parameters of the generator set and the power system in real time and perform operation protection. However, since the controllers of all brands of generator sets on the market cannot directly control the operation of the generator set stably and reliably, the controller must add high-power relays and related accessories through the secondary circuit to control the operation of the generator set stably and reliably.
[0003] In the prior art, for example, patent (CN 1992464A) discloses a wind turbine control system, including a central controller PLC, an intermediate relay board, a soft grid-connected control board, a display operation panel and wiring terminals. The central controller PLC mainly includes a CPU module 315-2DP, a digital input module DI, a digital output module D0, a high-speed counting module FM350-2, an analog input module AI, and a communication module CP340; the intermediate relay board is composed of a plurality of DC coil relays, and the intermediate relay board and the soft grid-connected control board are respectively connected to the input and output modules of the central controller PLC, and are connected to the control object through the wiring terminals. The extended connection control between the controller and the generator set is realized by setting the central controller PLC, the intermediate relay board, the soft grid-connected control board, the display operation panel and the wiring terminals.
[0004] Based on the above-mentioned existing technical solutions, although the controller can achieve stable control of the generator set, due to the poor connection between the various components, the layout is difficult to be regular and the wiring is cumbersome, resulting in low production efficiency. Summary of the invention
[0005] The present invention provides a controller monitoring protection diagnosis control method and system, which are used to solve the defect that various accessories cannot be well connected in the prior art.
[0006] In one aspect, the present invention provides a controller monitoring protection diagnosis control method, comprising: The video acquisition device pre-acquires reference images of the longitudinal vibration and the transverse vibration of each generator in the generator set, and displays the vibration amplitudes of the longitudinal vibration and the transverse vibration of the generator in the reference image by adding marks to the reference layer; During monitoring, the video stream of the generator set operation is collected by the video acquisition device installed above the generator set, and the collected video stream is decoded into longitudinal vibration frames. and lateral vibration frame ; The collected longitudinal vibration frames and lateral vibration frame Input the pre-trained generator identification model for processing, obtain the encoding of the longitudinal vibration and lateral vibration of each generator in the generator set, and identify the type of each generator; According to the encoding of the longitudinal vibration and lateral vibration of each generator, the corresponding generator is identified, and the correspondence between the longitudinal vibration and lateral vibration in the reference image and the longitudinal vibration and lateral vibration of the identified generator is determined; then, the marks added in the reference image are used to obtain the marks of the longitudinal vibration and lateral vibration in the monitoring frame during the actual monitoring process.
[0007] Preferably, an arbitrarily rotated and scaled frame of each generator in the generator set is processed based on the ORB algorithm, and a set of marker point invariant features related to the longitudinal vibration and lateral vibration of each generator in the generator set in the frame is obtained; target points are respectively taken from the marker point invariant feature set to form different target geometric structures, and the target geometric structures indicate the amplitudes of the longitudinal vibration and lateral vibration of each generator in the generator set.
[0008] Preferably, the generator identification model comprises: an encoder, a feature fusion module and a decoder; the encoder extracts the longitudinal vibration frame through four feature processing stages and lateral vibration frame Features at different levels; vibration frames at different levels in the four feature encoding stages of the encoder and lateral vibration frame The features are fused through the feature fusion module respectively; the decoder analyzes the fused features at each level and the decoded features of the decoder through four feature decoding stages; finally, the decoded features are input to the mapping layer to obtain the encoding of the longitudinal vibration and lateral vibration of each generator in the generator set; the decoded features are input to the fully connected layer to identify the type of each generator in the generator set.
[0009] Preferably, the first feature encoding stage of the encoder consists of a parsing layer; each of the parsing layers comprises a convolution layer and a clustering layer with a convolution kernel size of a×a and a convolution shift step of a, and the two groups of parsing layers respectively convert the dimension Longitudinal vibration frame and dimensions are Lateral vibration frame Convert to Analysis and , and respectively analyze and Perform analysis and processing.
[0010] Preferably, the analysis and Perform analytical processing, including: performing standardization, parsing before performing standardization and They are divided into two sets, one of which is fused with subsequent features, and the other is normalized. The results of normalization are divided into two sets, one of which forms dynamic weights, and the other passes through the linear layer and convolution to obtain feature maps. , feature map Through the two-dimensional selective scanning process, the two-dimensional selective scanning first performs scanning expansion, and the feature map Scan forward horizontally and vertically from top to bottom to expand the feature map into two groups of sequences and , taking the two sets of sequences as input, and passing through three linear layers to obtain : , ; but, , ,in, is a matrix whose parameters can be learned; is the identity matrix; represents a linear layer, ; The state equations of the two sets of sequences are: ; ,in, is a matrix; is the intermediate state of the corresponding input; The output sequences corresponding to the two groups of sequences are obtained through the state equation, and the output sequences are sorted and fused in the same order.
[0011] Preferably, the second to fourth feature encoding stages include: Perform the normalization process twice. Each time the input before normalization is divided into two sets, one of which is fused with subsequent features, and the other set is divided into two sets by the results after normalization. One set of the results after each normalization is passed through a linear layer to form a dynamic weight, and the other set is passed through a linear layer to obtain a feature map; set the two obtained feature maps as feature maps and ,Through 2D selective scanning processing: 2D selective scanning first performs scanning expansion, and the feature maps and Scan forward from top to bottom in the horizontal and vertical directions. Expand into two sets of sequences and ; Take two sets of flattened sequences as input and pass them through the linear layer to obtain and : , ; , ; but, , , , , in, , is a matrix whose parameters can be learned; is the identity matrix; Swapping matrices in the state matrix , ; Two sets of flat sequences The state equation is: ; ; in, is a matrix with learnable parameters, and the corresponding four sets of flat sequences are obtained through the state equation The output sequences are sorted and merged in the same order; Four flat sequences The state equation is: ; ; in, is a matrix with learnable parameters, and the corresponding four sets of flat sequences are obtained through the state equation The output sequences are sorted and merged in the same order; The two fusion results are normalized by two layers and weighted with their respective dynamic weights. The weighted results are processed by two linear layers and fused with their respective skip chains, and finally processed by linear layers respectively.
[0012] Preferably, the feature fusion module includes: a first fusion layer, a second fusion layer, a third fusion layer and a fourth fusion layer; the structures of the fusion layers at each level are consistent.
[0013] By forward scanning and expanding the spliced features, two sets of spliced scan sequences are obtained. Then each fusion layer takes the two spliced scan sequences as input and obtains them through the linear layer respectively. : , ,
[0014] but, , , in, is a matrix whose parameters can be learned; is the identity matrix; The state equation of the two spliced scan sequences is: ; ; in, is a matrix with learnable parameters. The output sequence corresponding to the two concatenated scan sequences is obtained through the state equation. The two output sequences are sorted and fused in the same order. The fusion results are split and normalized by two layers respectively, and then weighted with their respective weights. The weighted results are processed by two linear layers and weighted with their respective dynamic weights. The weighted results are concatenated in dimension and finally mapped by linear layer to reduce the channel dimension.
[0015] Preferably, constructing training data: collecting longitudinal vibration frames by a video acquisition device arranged above the generator set and lateral vibration frame ; Use the longitudinal vibration and transverse vibration codes of each generator in the generator set as labels to mark the collected longitudinal vibration frames and lateral vibration frame Obtain training data; and use the training data to train the generator identification model.
[0016] Preferably, training the generator identification model using training data comprises: The IOU loss between the label and the encoding predicted by the generator recognition model is used as the loss function; the fully connected layer parameters of the generator recognition model are frozen, and based on the loss function, the parameters of the generator recognition model except the fully connected layer are updated using the gradient descent algorithm; until the loss function is less than the set threshold.
[0017] Preferably, the first three feature decoding stages of the decoder are implemented by decoding the visual state space and upsampling, and the last feature decoding stage is implemented by decoding the visual state space, a fully connected layer and a coding mapping layer, wherein the decoded visual state space includes: a convolution layer, a visual state space, and the output of the visual state space is fused with subsequent features through skip chaining on the one hand, and is weighted with the channel attention after layer normalization on the other hand, and is fused with the output of the visual state space after weighting; the output of the visual state space is processed by a convolution layer after normalization, the output of the convolution layer is divided into two sets, one set is processed by a maximum pooling layer, and the other set is processed by an average pooling layer, and the outputs of the maximum pooling layer and the average pooling layer are added and processed by a sigmoid activation function to obtain the channel attention.
[0018] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the controller monitoring, protection, diagnosis and control methods described above.
[0019] On the other hand, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the controller monitoring, protection, diagnosis and control methods described above.
[0020] The present invention provides a controller monitoring protection diagnosis control method and system, which collects the reference video stream of the longitudinal vibration and lateral vibration of each generator in the generator set through a video acquisition device, and obtains the longitudinal vibration frame and the lateral vibration frame after decoding the reference video stream. The longitudinal vibration frame and the lateral vibration frame are analyzed to obtain the longitudinal vibration threshold and lateral vibration threshold After identifying the corresponding generator according to the operating status code of each generator in the generator set, the longitudinal vibration threshold and the lateral vibration threshold of this generator are determined, and the longitudinal vibration frame and the lateral vibration frame of this generator in the actual monitoring process are marked respectively using the longitudinal vibration threshold and the lateral vibration threshold of this generator, which solves the problem of low production efficiency due to the poor connection between the various accessories, the difficulty in making the layout regular, and the cumbersome wiring, thereby achieving the beneficial effect of effectively improving productivity while reducing production costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0022] Figure 1 It is a flow chart of a controller monitoring protection diagnosis control method provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0024] Combine the following Figure 1 The controller monitoring protection diagnosis control method of the present invention is described.
[0025] like Figure 1 As shown, the controller monitoring protection diagnosis control method provided by the embodiment of the present invention can be executed by the controller, which mainly monitors the operating status of each generator in the generator set and integrates the following methods into the controller: The reference video stream of the longitudinal vibration and transverse vibration of each generator in the generator set is collected by video acquisition equipment, and the longitudinal vibration frame and transverse vibration frame are obtained after decoding the reference video stream. The longitudinal vibration threshold is obtained by analyzing the longitudinal vibration frame and the transverse vibration frame. and lateral vibration threshold ; The purpose of constructing reference markers is to construct the points in the actual acquired image that match the reference markers in the reference image into markers in the image. The principle is to process the image of the generator with arbitrary rotation and scaling based on the ORB (Oriented FAST and Rotated BRIEF) algorithm to obtain the image with the longitudinal vibration frame. and lateral vibration frame A related marker point invariant feature point set; target points are respectively taken from the marker point invariant feature point set to form different target geometric structures, and the target geometric structures formed by the target points, such as triangle and T-shape, can indicate the longitudinal and lateral vibration amplitudes of the generator; the image in the actual monitoring process can be obtained by vibrating the reference image up, down, left and right relative to the reference image, because the feature point matching from the marker point invariant feature point set is not affected by the vibration, the reference image reference marker can be used to obtain the marker of the image in the actual monitoring process.
[0026] Although the ORB (Oriented FAST and Rotated BRIEF) algorithm can extract the invariant features of the marker points, it cannot identify the corresponding generator. Therefore, it is impossible to determine the correspondence between the target vibration form and the generator.
[0027] In view of the identification requirements of each generator in a generator set, the present application provides a generator identification model, which includes: an encoder, a feature fusion module and a decoder.
[0028] Construct training data: The video acquisition device set above the generator set collects the video stream of the generator set operation and decodes the collected video stream into longitudinal vibration frames and lateral vibration frame ; Use the encoding of the longitudinal vibration and transverse vibration of each generator in the generator group as a label to mark the longitudinal vibration frame and lateral vibration frame The training data is obtained and the generator identification model is trained using the training data.
[0029] The encoder extracts the vibration frames in the training data through four feature processing stages. and lateral vibration frame Features at different levels. The encoder consists of one feature encoding stage based on visual state space and three feature encoding stages based on cross visual state space.
[0030] The first feature encoding stage of the encoder is implemented by the parsing layer and the visual state space connected to the parsing layer. Each of the parsing layers includes a convolution layer and a clustering layer with a convolution kernel size of a×a and a convolution step size of a. The two groups of parsing layers respectively convert the dimension Longitudinal vibration frame and dimensions are Lateral vibration frame Convert to Analysis and , and respectively analyze and Perform analysis and processing.
[0031] Performing the normalization process: Parsing before layer normalization and They are divided into two sets, one of which is fused with subsequent features through skip chaining, and the other is processed by layer normalization. The results of layer normalization are divided into two sets, one of which is processed by linear layer to form dynamic weights, and the other is processed by linear layer and depth-separable convolution to obtain feature maps. , feature map Through the two-dimensional selective scanning process, the two-dimensional selective scanning first performs scanning expansion, and the feature map Scan forward horizontally and vertically from top to bottom to expand the feature map into two groups of sequences and , taking the two sets of sequences as input, and passing through three linear layers to obtain : , ; but, , ,in, is a matrix whose parameters can be learned; is the identity matrix; represents a linear layer, ; The state equations of the two sets of sequences are: ; ,in, is a matrix; is the intermediate state of the corresponding input; The output sequences corresponding to the two groups of sequences are obtained through the state equation, and the output sequences are sorted and merged in the same order; The fusion result is processed by layer normalization and then weighted with dynamic weights. The weighted result is processed by a linear layer and then fused with skip chaining, and finally processed by a linear layer again.
[0032] The first feature encoding stage encodes the longitudinal vibration frames through the visual state space and lateral vibration frame Each analysis of is encoded with contextual long-range modeling.
[0033] The second feature encoding stage is implemented by two downsampling and cross visual state spaces of the second feature encoding stage. The output of the two visual state spaces of the first feature encoding stage and The two downsamples are respectively input into the second feature encoding stage, and the two downsamples are respectively and Height and width downsampling, and The channel expansion of Features and , downsampling is achieved through convolutional layers. The two downsampling layers respectively and The corresponding input of the cross-visual state space is provided to the second feature encoding stage.
[0034] The third feature encoding stage and the fourth feature encoding stage are similar to the second feature encoding stage.
[0035] The third feature encoding stage is implemented by two downsampling and the cross visual state space of the third feature encoding stage. The two outputs of the cross visual state space of the second feature encoding stage are and The two downsamples are respectively input into the third feature encoding stage, and the two downsamples are respectively and Height and width downsampling, and The channel expansion of Features and , downsampling is achieved through convolutional layers. The two downsampling layers respectively and The corresponding input of the cross-visual state space is provided to the third feature encoding stage.
[0036] The fourth feature encoding stage is implemented by two downsampling and the cross visual state space of the fourth feature encoding stage. The two outputs of the cross visual state space of the third feature encoding stage and The two downsamples are respectively input into the third feature encoding stage, and the two downsamples are respectively and Height and width downsampling, and The channel expansion of Features and , downsampling is achieved through convolutional layers. The two downsampling layers respectively and The corresponding input provided to the cross-visual state space of the fourth feature encoding stage is obtained after being processed by the cross-visual state space of the fourth feature encoding stage. and .
[0037] The cross-vision state space includes: Two layers of standardization, the input of each layer before standardization is divided into two sets, one of which is fused with subsequent features through skip chaining, and the other is processed through layer standardization. The results of each layer standardization are divided into two sets, one of which is processed by a linear layer to form dynamic weights, and the other is processed by a linear layer and depth-separable convolution to obtain a feature map; the two feature maps obtained by the two paths are set as feature maps and . Feature map and Through two-dimensional selective scanning processing: two-dimensional selective scanning first performs scanning expansion, respectively, and Scan forward from top to bottom in the horizontal and vertical directions to and Expand into two sets of flat sequences respectively and The two sets of flattened sequences are taken as input and passed through the linear layer to obtain and : , ; , ; but, , , , , in, , is a matrix whose parameters can be learned; is the identity matrix; Swapping matrices in the state matrix , ; Two sets of flat sequences The state equation is: ; ; in, is a matrix with learnable parameters, and the corresponding four sets of flat sequences are obtained through the state equation The output sequences are sorted and fused in the same order.
[0038] Two sets of flat sequences The state equation is: ; ; in, is a matrix with learnable parameters, and the corresponding four sets of flat sequences are obtained through the state equation The output sequences are sorted and fused in the same order.
[0039] The two fusion results are normalized by two layers and weighted with their respective dynamic weights. The weighted results are processed by two linear layers and fused with their respective skip chains, and finally processed by linear layers respectively.
[0040] When modeling long-range context using the cross-visual state space, the exchange matrix , By using the state matrix generated by the other party to construct the state equation and realize the cross-fusion of features, it can effectively extract the common linear invariant relationship from RGB images and thermal imaging, and achieve better modeling.
[0041] Vertical vibration frames at different levels of the four feature encoding stages of the visual state space encoder and lateral vibration frame The features are fused respectively through a feature fusion module. Specifically, the feature fusion module includes: four levels of fusion layers implemented based on the visual state space, namely, a first fusion layer, a second fusion layer, a third fusion layer and a fourth fusion layer.
[0042] The structure of each fusion layer is consistent, including: Two linear layers that process different received features respectively, the input features before each linear layer are divided into two sets, one set is processed by a linear layer into sets to form dynamic weights, and the other set is input into the linear layer and the depthwise separable convolution layer; in the linear layer and the depthwise separable convolution layer structure, the output of the linear layer is divided into two sets, one set is input into the depthwise separable convolution layer, and the other set is processed by a convolution layer to obtain weights, and the output of the depthwise separable convolution layer is processed by feature splicing, and the feature splicing splices the two received features in the channel dimension. The result of the feature splicing is processed by the fused visual state space, and the fused visual state space expands the spliced features by forward scanning and reverse scanning to obtain two sets of spliced scan sequences. For example, the linear layer and the depthwise separable convolution layer of the first fusion layer are processed and get and , feature splicing processing and get ,right Scanned , ; The linear layer and depth-separable convolutional layer processing of the second fusion layer and get and , feature splicing processing and get ,right Scanned , ; The third fusion layer is processed by the linear layer and the depth-separable convolution layer and get and , feature splicing processing and get ,right Scanned , ; The fourth fusion layer is processed by the linear layer and the depth-separable convolution layer and get and , feature splicing processing and get ,right Scanned , Then the fused visual state space of each fusion layer takes the two spliced scan sequences as input and obtains them through the linear layer respectively. : , ,
[0043] but, , , in, is a matrix whose parameters can be learned; is the identity matrix; The state equation of the two spliced scan sequences is: ; ; in, is a matrix with learnable parameters. The output sequence corresponding to the two concatenated scan sequences is obtained through the state equation. The two output sequences are sorted in the same order and fused.
[0044] The fusion results are split and normalized by two layers respectively, and then weighted with their respective weights. The weighted results are processed by two linear layers and weighted with their respective dynamic weights. The weighted results are concatenated in dimension and finally mapped by linear layer to reduce the channel dimension.
[0045] The visual state space feature fusion module strengthens and fuses the features of each level of the visual state space encoder through the visual state space.
[0046] The decoder analyzes the fused features at each level and the decoded features of the previous decoder through four feature decoding stages. The decoder contains four feature decoding stages based on the decoded visual state space.
[0047] The output of the fourth fusion layer is provided to the decoded visual state space of the first feature decoding stage. The output of the decoded visual state space of the first feature decoding stage is spliced with the output of the third fusion layer after upsampling, and the splicing result is input to the decoded visual state space of the second feature decoding stage; the output of the decoded visual state space of the second feature decoding stage is spliced with the output of the second fusion layer after upsampling, and the splicing result is input to the decoded visual state space of the third feature decoding stage; the output of the decoded visual state space of the third feature decoding stage is spliced with the output of the first fusion layer after upsampling, and the splicing result is input to the decoded visual state space of the fourth feature decoding stage; the output of the decoded visual state space of the fourth feature decoding stage obtains the encoding of the longitudinal vibration and lateral vibration of each generator in the generator set through the encoding mapping layer formed by the linear layer, and the decoded features of the decoded visual state space output of the fourth feature decoding stage are analyzed by the fully connected layer to identify the type of each generator. Among them, upsampling is realized by dilated convolution.
[0048] Decoding the visual state space includes: convolution layer, visual state space, the output of the visual state space is fused with subsequent features through skip chaining on the one hand, and weighted with the channel attention after layer normalization on the other hand, and fused with the output of the visual state space after weighting; the output of the visual state space is processed by the convolution layer after normalization, and the output of the convolution layer is divided into two sets, one set is processed by the maximum pooling layer, and the other set is processed by the average pooling layer. The outputs of the maximum pooling layer and the average pooling layer are added and processed by the sigmoid activation function to obtain the channel attention.
[0049] During monitoring, the longitudinal vibration frames are collected by the video acquisition device installed above the generator set. and lateral vibration frame ; The generator identification model analyzes the longitudinal vibration frame and lateral vibration frame The output of the decoded visual state space in the fourth feature decoding stage obtains the encoding of the longitudinal vibration and lateral vibration of each generator in the generator set through the encoding mapping layer formed by the linear layer, and the output of the decoded visual state space in the fourth feature decoding stage is analyzed by the fully connected layer to identify the type of each generator.
[0050] In order to make the generator identification model based on the longitudinal vibration frame and lateral vibration frame The longitudinal vibration and lateral vibration codes of each generator in the generator set are obtained, and the type of each generator is identified.
[0051] The IOU loss between the label and the encoding predicted by the generator recognition model is used as the loss function; first, the fully connected layer parameters are frozen, and based on the loss function, the parameters of the generator recognition model except the fully connected layer are updated using the gradient descent algorithm; until the loss function is less than the set threshold.
[0052] After the generator is identified according to the coding of the longitudinal vibration and transverse vibration of each generator in the generator set, the corresponding relationship between the longitudinal vibration and transverse vibration of each generator in the generator set in the reference image and the identified longitudinal vibration and transverse vibration is determined; then the longitudinal vibration frame in the image during the actual monitoring process is obtained using the reference mark in the reference image and lateral vibration frame Mark.
[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A controller monitoring protection diagnosis control method, characterized in that: include: The video acquisition device pre-acquires reference images of the longitudinal vibration and the transverse vibration of each generator in the generator set, and displays the vibration amplitudes of the longitudinal vibration and the transverse vibration of the generator in the reference image by adding marks to the reference layer; During monitoring, the video stream of the generator set operation is collected by the video acquisition device installed above the generator set, and the collected video stream is decoded into longitudinal vibration frames. and lateral vibration frame ; The collected longitudinal vibration frames and lateral vibration frame Input the pre-trained generator identification model for processing, obtain the encoding of the longitudinal vibration and lateral vibration of each generator in the generator set, and identify the type of each generator; Identify the corresponding generator according to the coding of the longitudinal vibration and the transverse vibration of each generator, and determine the corresponding relationship between the longitudinal vibration and the transverse vibration in the reference image and the longitudinal vibration and the transverse vibration of the identified generator; Then, the markers added in the reference image are used to obtain the markers of the longitudinal vibration and the lateral vibration in the monitoring frame during the actual monitoring process.
2. A controller monitoring protection diagnosis control method according to claim 1, characterized in that: Processing the arbitrarily rotated and scaled frames of each generator in the generator set based on the ORB algorithm, the invariant feature set of the marker points related to the longitudinal vibration and lateral vibration of each generator in the generator set in the frame; Target points are respectively taken from the marker point invariant feature set to form different target geometric structures, where the target geometric structures indicate the amplitudes of the longitudinal vibration and the lateral vibration of each generator in the motor group.
3. A controller monitoring protection diagnosis control method according to claim 1, characterized in that: The generator identification model includes: an encoder, a feature fusion module and a decoder; the encoder extracts the longitudinal vibration frame through four feature processing stages and lateral vibration frame Features at different levels; vibration frames at different levels in the four feature encoding stages of the encoder and lateral vibration frame The features are fused through the feature fusion module respectively; the decoder analyzes the fused features at each level and the decoded features of the decoder through four feature decoding stages; finally, the decoded features are input to the mapping layer to obtain the encoding of the longitudinal vibration and lateral vibration of each generator in the generator set; the decoded features are input to the fully connected layer to identify the type of each generator in the generator set.
4. A controller monitoring protection diagnosis control method according to claim 3, characterized in that: The first feature encoding stage of the encoder consists of a parsing layer; each of the parsing layers includes a convolution layer and a clustering layer with a convolution kernel size of a×a and a convolution shift step of a. The two groups of parsing layers respectively convert the dimensions of Longitudinal vibration frame and dimensions are Lateral vibration frame Convert to Analysis and , and respectively analyze and Perform analysis and processing.
5. A controller monitoring protection diagnosis control method according to claim 4, characterized in that: The analysis will and Perform analytical processing, including: performing standardization, parsing before performing standardization and They are divided into two sets, one of which is fused with subsequent features, and the other is normalized. The results of normalization are divided into two sets, one of which forms dynamic weights, and the other passes through the linear layer and convolution to obtain feature maps. , feature map Through the two-dimensional selective scanning process, the two-dimensional selective scanning first performs scanning expansion, and the feature map Scan forward horizontally and vertically from top to bottom to expand the feature map into two groups of sequences and , taking the two sets of sequences as input, and passing through three linear layers to obtain : ， ； but, , ,in, is a matrix whose parameters can be learned; is the identity matrix; represents a linear layer, ; The state equations of the two sets of sequences are: ; ,in, is a matrix; is the intermediate state of the corresponding input; The output sequences corresponding to the two groups of sequences are obtained through the state equation, and the output sequences are sorted and fused in the same order.
6. A controller monitoring protection diagnosis control method according to claim 3, characterized in that: The second to fourth feature encoding stages include: Perform the normalization process twice. Each time the input before normalization is divided into two sets, one of which is fused with subsequent features, and the other set is divided into two sets by the results after normalization. One set of the results after each normalization is passed through a linear layer to form a dynamic weight, and the other set is passed through a linear layer to obtain a feature map; set the two obtained feature maps as feature maps and ,Through 2D selective scanning processing: 2D selective scanning first performs scanning expansion, and the feature maps and Scan forward from top to bottom in the horizontal and vertical directions. Expand into two sets of sequences and ; Take two sets of flattened sequences as input and pass them through the linear layer to obtain and : ， ； ， ； but, , , ， ， in, , is a matrix whose parameters can be learned; is the identity matrix; Swapping matrices in the state matrix , ; Two sets of flat sequences The state equation is: ； ； in, is a matrix with learnable parameters, and the corresponding four sets of flat sequences are obtained through the state equation The output sequences are sorted and merged in the same order; Four flat sequences The state equation is: ； ； in, is a matrix with learnable parameters, and the corresponding four sets of flat sequences are obtained through the state equation The output sequences are sorted and merged in the same order; The two fusion results are normalized by two layers and weighted with their respective dynamic weights. The weighted results are processed by two linear layers and fused with their respective skip chains, and finally processed by linear layers respectively.
7. A controller monitoring protection diagnosis control method according to claim 3, characterized in that: The feature fusion module includes: a first fusion layer, a second fusion layer, a third fusion layer and a fourth fusion layer; the structures of each fusion layer are consistent; By forward scanning and expanding the spliced features, two sets of spliced scan sequences are obtained. Then each fusion layer takes the two spliced scan sequences as input and obtains them through the linear layer respectively. : ， ， but, , , in, is a matrix whose parameters can be learned; is the identity matrix; The state equation of the two spliced scan sequences is: ； ； in, is a matrix with learnable parameters. The output sequence corresponding to the two concatenated scan sequences is obtained through the state equation. The two output sequences are sorted and fused in the same order. The fusion results are split and normalized by two layers respectively, and then weighted with their respective weights. The weighted results are processed by two linear layers and weighted with their respective dynamic weights. The weighted results are concatenated in dimension and finally mapped by linear layer to reduce the channel dimension.
8. The controller monitoring protection diagnosis control method according to claim 3 is characterized in that: Constructing training data: longitudinal vibration frames are collected by video acquisition equipment installed above the generator set. and lateral vibration frame ; Use the longitudinal vibration and transverse vibration codes of each generator in the generator set as labels to mark the collected longitudinal vibration frames and lateral vibration frame Obtain training data; and use the training data to train the generator identification model.
9. A controller monitoring protection diagnosis control method according to claim 8, characterized in that: Training the generator identification model using training data includes: The IOU loss between the label and the encoding predicted by the generator recognition model is used as the loss function; the fully connected layer parameters of the generator recognition model are frozen, and based on the loss function, the parameters of the generator recognition model except the fully connected layer are updated using the gradient descent algorithm; until the loss function is less than the set threshold.
10. A controller monitoring protection diagnosis control method according to claim 3, characterized in that: The first three feature decoding stages of the decoder are realized by decoding the visual state space and upsampling, and the last feature decoding stage is realized by decoding the visual state space, a fully connected layer and a coding mapping layer, wherein the decoded visual state space includes: a convolution layer, a visual state space, and the output of the visual state space is fused with subsequent features through skip chaining on the one hand, and is weighted with the channel attention after layer normalization on the other hand, and is fused with the output of the visual state space after weighting; the output of the visual state space is processed by the convolution layer after normalization, and the output of the convolution layer is divided into two sets, one set is processed by the maximum pooling layer, and the other set is processed by the average pooling layer, and the outputs of the maximum pooling layer and the average pooling layer are added and processed by the sigmoid activation function to obtain the channel attention.
Citation Information
Patent Citations
Control system of wind turbines
CN1992464A