Power communication network key node equipment state sensing method and system

Abstract

The invention provides an electric power communication network key node equipment state sensing method and system, and the method comprises the steps: carrying out the up-sampling of target electric power data of target key node equipment through a double-up-sampling mode, and fusing the advantages of different up-sampling modes, thereby enabling the input data to contain rich data information, and enabling the input data to be more accurate. The target state sensing model extracts features of different levels of input data through a dual-channel feature extraction layer, so that the extracted features retain rich information of the input data, and through hierarchical attention calculation, shallow features and deep features are respectively processed and then fused, the features of different levels can be fully learned, and the learning efficiency is improved. And the accuracy of the output result of the target state sensing model is improved. And due to application of dual-channel feature extraction and hierarchical attention calculation, compared with traditional single-channel feature extraction, the network depth is shallower, the model is lighter, and the deployment efficiency and the calculation efficiency of the model at edge calculation nodes are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing technology for power communication networks, and in particular to a method and system for sensing the status of key node equipment in power communication networks. Background Technology

[0002] The operating status of key node equipment in the power communication network is directly related to the safety and stability of the power grid. Traditional status sensing methods mainly rely on low-frequency sensor data and rule judgment, which have problems such as low sensing accuracy, large response delay, and difficulty in adapting to complex environments. Especially in edge node deployment, they are constrained by computing resources and it is difficult to achieve efficient real-time monitoring. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method and system for status perception of key node equipment in power communication networks, so as to improve the accuracy of status perception of key node equipment in power communication networks.

[0004] According to one aspect of the present invention, a method for status awareness of key node equipment in a power communication network is provided, applied to edge nodes, the method comprising: Acquire target power data of key nodes in the power communication network; The target power data is double-upsampled using a fusion upsampling method to obtain the input data corresponding to the target power data; The input data is fed into a pre-trained target state perception model, which includes a dual-channel feature extraction layer and a hierarchical channel attention fusion mechanism layer. The dual-channel feature extraction layer is used to extract shallow and deep features from the input data. The hierarchical channel attention fusion mechanism layer is used to perform attention calculations on the shallow and deep features respectively to obtain corresponding shallow attention features and deep attention features. The shallow and deep attention features are then fused based on preset channel weights to obtain target fusion features. The preset channel weights are used to identify the importance of different channels in the shallow and deep attention features. Obtain the target state perception result output by the target state perception model based on the target fusion features.

[0005] In one possible embodiment, the step of performing double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data includes: The target power data is upsampled using a deconvolution upsampling method to obtain the first upsampled data. The target power data is upsampled using an interpolation upsampling method to obtain second upsampled data. The first upsampled data and the second upsampled data are fused to obtain the input data corresponding to the target power data.

[0006] In one possible embodiment, the dual-channel feature extraction layer includes a shallow feature extraction sub-layer and a deep feature extraction sub-layer; the shallow feature extraction sub-layer contains multiple convolutional layers with different kernel sizes, and the deep feature extraction sub-layer contains multiple deep feature extraction units, each of which contains multiple convolutional layers with an increasing number of channels and a channel compression layer.

[0007] In one possible embodiment, the dual-channel feature extraction layer further includes a formatted convolutional layer with a kernel size of 1×1; the method further includes: The shallow features and the deep features are respectively input into the formatted convolutional layer to obtain shallow features and deep features of the same dimension.

[0008] In one possible embodiment, the hierarchical channel attention mechanism layer includes a shallow attention mechanism layer, a deep attention mechanism layer, and a weighted fusion layer; The shallow attention mechanism layer is used to output shallow attention features for the shallow features; The deep attention mechanism layer is used to output deep attention features for the deep features; The weighted fusion layer is used to generate a channel weight vector based on the shallow attention features, deep attention features, and preset parameters; and to fuse the shallow attention features and deep attention features using the channel weight vector to obtain the target fusion feature.

[0009] According to another aspect of the present invention, a status awareness system for key node equipment in a power communication network is provided, applied to edge nodes, the system comprising: The acquisition module is used to acquire target power data of key nodes in the power communication network. The upsampling module is used to perform double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data; An input module is used to input the input data into a pre-trained target state perception model. The target state perception model includes a dual-channel feature extraction layer and a hierarchical channel attention fusion mechanism layer. The dual-channel feature extraction layer extracts shallow and deep features from the input data. The hierarchical channel attention fusion mechanism layer performs attention calculations on the shallow and deep features respectively to obtain corresponding shallow and deep attention features. It then fuses the shallow and deep attention features based on preset channel weights to obtain target fusion features. The preset channel weights are used to identify the importance of different channels in the shallow and deep attention features. The output module is used to obtain the target state perception result output by the target state perception model based on the target fusion features.

[0010] In one possible embodiment, the step of performing double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data includes: The target power data is upsampled using a deconvolution upsampling method to obtain the first upsampled data. The target power data is upsampled using an interpolation upsampling method to obtain second upsampled data. The first upsampled data and the second upsampled data are fused to obtain the input data corresponding to the target power data.

[0011] In one possible embodiment, the dual-channel feature extraction layer includes a shallow feature extraction sub-layer and a deep feature extraction sub-layer; the shallow feature extraction sub-layer contains multiple convolutional layers with different kernel sizes, and the deep feature extraction sub-layer contains multiple deep feature extraction units, each of which contains multiple convolutional layers with an increasing number of channels and a channel compression layer; The dual-channel feature extraction layer also includes a formatted convolutional layer with a kernel size of 1×1; the target state perception module is further used to input the shallow features and the deep features into the formatted convolutional layer respectively to obtain shallow features and deep features with the same dimension. The hierarchical channel attention mechanism layer includes a shallow attention mechanism layer, a deep attention mechanism layer, and a weighted fusion layer; The shallow attention mechanism layer is used to output shallow attention features for the shallow features; The deep attention mechanism layer is used to output deep attention features for the deep features; The weighted fusion layer is used to generate a channel weight vector based on the shallow attention features, deep attention features, and preset parameters; and to fuse the shallow attention features and deep attention features using the channel weight vector to obtain the target fusion feature.

[0012] According to another aspect of the present invention, an electronic device is provided, comprising: Processor; and Stored program memory, The program includes instructions that, when executed by the processor, cause the processor to perform any of the above-described power communication network key node equipment status awareness methods.

[0013] According to another aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute any of the above-described power communication network key node equipment status sensing methods.

[0014] The embodiments of this invention provide one or more technical solutions. After acquiring target power data of key node equipment in a power communication network, the target power data is double-upsampled to supplement the target power data with corresponding high-frequency data, resulting in input data. A target state perception model extracts features at different levels from the input data and performs attention calculations on the features at different levels. Based on the attention calculation results of the features at different levels, channel weights are calculated, and the attention features at different levels are weighted based on the channel weights to obtain target fusion features. Based on the target fusion features, a state perception result is output. By applying the embodiments of this invention, the target power data is sampled using a double-upsampled method, which can combine the advantages of different upsampled methods, thereby enabling the input data to contain rich data information. The target state perception model extracts features at different levels from the input data through a dual-channel feature extraction layer, so that the extracted features retain rich information from the input data. Furthermore, through hierarchical attention calculation, shallow and deep features are processed separately and then fused, which can fully learn the features at different levels, thereby improving the accuracy of the output results of the target state perception model. Furthermore, due to the application of dual-channel feature extraction and hierarchical attention calculation, the network depth is shallower and the model is lighter compared to traditional single-channel feature extraction, which improves the deployment efficiency and computational efficiency of the model on edge computing nodes. Attached Figure Description

[0015] Further details, features, and advantages of the invention are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 A flowchart illustrating a method for sensing the status of key node equipment in a power communication network according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a deep feature extraction unit in the power communication network key node equipment status perception method provided in an embodiment of the present invention. Figure 3 A schematic diagram of a status awareness system for key node equipment in a power communication network provided in an embodiment of the present invention; Figure 4 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present invention is shown. Detailed Implementation

[0016] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0017] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0018] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0019] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0020] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0021] To enhance perception capabilities, related technologies propose converting operational data into pseudo-images and introducing convolutional neural networks for state recognition of key node equipment in power communication networks. However, conventional single-path structures suffer from incomplete feature extraction and low information utilization. Furthermore, while attention mechanisms can improve model discrimination capabilities, some methods are structurally complex, hindering edge deployment.

[0022] Based on this, embodiments of the present invention provide a method and system for sensing the status of key node equipment in a power communication network. The method for sensing the status of key node equipment in a power communication network provided by the present invention can be applied to any electronic device with equipment status sensing capabilities, such as a server, computer, industrial control computer, or mobile terminal. In one possible embodiment, the method for sensing the status of key node equipment in a power communication network provided by the present invention can be applied to edge nodes. The solution of the present invention is described below with reference to the accompanying drawings: Figure 1 A flowchart illustrating a method for status awareness of key node equipment in a power communication network provided in an embodiment of the present invention may include the following steps: S101. Obtain target power data of key nodes in the power communication network; S102. The target power data is double-upsampled using a fusion upsampling method to obtain the input data corresponding to the target power data. S103. The input data is input into a pre-trained target state perception model, wherein the target state perception model includes a dual-channel feature extraction layer and a hierarchical channel attention fusion mechanism layer. The dual-channel feature extraction layer is used to extract shallow and deep features of the input data. The hierarchical channel attention fusion mechanism layer is used to perform attention calculations on the shallow and deep features respectively to obtain corresponding shallow attention features and deep attention features. The shallow attention features and deep attention features are then fused based on preset channel weights to obtain target fusion features. The preset channel weights are used to identify the importance of different channels in the shallow and deep attention features. S104. Obtain the target state perception result output by the target state perception model based on the target fusion features.

[0023] In this embodiment of the invention, after acquiring the target power data of key node equipment in the power communication network, the target power data is double-upsampled to supplement the target power data with corresponding high-frequency data, resulting in input data. The target state perception model extracts features at different levels from the input data and performs attention calculations on the features at different levels respectively. Based on the attention calculation results of the features at different levels, channel weights are calculated, and the attention features at different levels are weighted based on the channel weights to obtain target fusion features. Based on the target fusion features, the state perception result is output. By applying this embodiment of the invention, the target power data is sampled using a double-upsampled method, which can combine the advantages of different upsampled methods, thereby enabling the input data to contain rich data information. The target state perception model extracts features at different levels from the input data through a dual-channel feature extraction layer, so that the extracted features retain rich information from the input data. Furthermore, through hierarchical attention calculation, shallow features and deep features are processed separately and then fused, which can fully learn the features at different levels, thereby improving the accuracy of the output results of the target state perception model. Furthermore, due to the application of dual-channel feature extraction and hierarchical attention calculation, the network depth is shallower and the model is lighter compared to traditional single-channel feature extraction, which improves the deployment efficiency and computational efficiency of the model on edge computing nodes.

[0024] The following provides an exemplary description of S101-S104: The target key node devices in the power communication network can be selected according to the actual application scenario, and may include transmission equipment, routing equipment, access equipment, etc. For example, a list of key node devices can be preset, and target power data for each key node device in the list can be acquired at preset time intervals. Alternatively, in the event of data transmission anomalies, the node device in the corresponding data link can be identified as the target key node device, and the corresponding target power data can be acquired. This target power data may include key indicators reflecting the operating status of the target key node device, such as voltage, current, temperature, harmonics, and vibration, as well as the corresponding acquisition timestamps.

[0025] Raw target power data collected from power communication networks may be affected by factors such as sensor errors or edge noise, resulting in missing or incorrect data. Therefore, after acquiring the raw target power data, it can be preprocessed. This preprocessing may include missing value completion, outlier removal, and converting the target power data into pseudo-image input, thereby transforming the target power data into input data that can be processed by the model.

[0026] For example, the preprocessing procedure may include completing small-scale missing samples using linear interpolation and removing outliers. Outliers can be values ​​that deviate significantly from the average value of the collected data, such as voltage values ​​that differ greatly from the average voltage value. Because the equipment data exhibits obvious daily periodic variations, meaning the collected target power data is actually a time series, the continuously sampled target power data can be divided into a two-dimensional matrix by day. The constructed matrix serves as the pseudo-image input.

[0027] Due to limitations in equipment bandwidth and transmission for power data sampling, high-frequency data is often impossible to acquire in real time. However, high-frequency data also contains rich information about the power communication network; therefore, it is necessary to upsample low-frequency data to reconstruct it into high-frequency data. Specifically, the pseudo-image input obtained from the preprocessing described above can be upsampled to obtain the final model input data. Traditional single upsampling methods each have their drawbacks; therefore, this invention uses a dual upsampling method, combining the advantages of two upsampling methods to improve upsampling quality and training efficiency. In one possible embodiment, the dual upsampling of the target power data using the fused upsampling method to obtain the input data corresponding to the target power data includes: S21. Upsample the target power data using deconvolution upsampling to obtain the first upsampled data.

[0028] Deconvolutional upsampling can be implemented using a pre-trained deconvolutional neural network, which can be a CNN network containing convolutional kernels of a preset size. The deconvolution operation is not a simple inverse convolution operation; instead, it expands the size of the feature map by inserting zero values ​​between pixels in the original feature map. Specifically, deconvolution first inserts a fixed number of zero elements between rows and columns of the original data according to a set upsampling factor, thereby expanding the dimension of the feature map to the target size. If the upsampling factor is λ, then... Zeros. The specific formula is:

[0029]

[0030] in, It is the index of the output signal, indicating the nth element in the target output sequence. One location; This is the insertion result at signal n. For the upsampling results, It is the offset index inside the convolution kernel, and the convolution kernel weights. The weights are automatically learned by the neural network during training. The initial values ​​are initialized with a random distribution and then continuously adjusted through backpropagation and loss function optimization to finally obtain a set of weights that best reproduces the target's high-resolution data. This deconvolutional neural network can be included in the aforementioned target state-aware model and trained together with it, or it can be trained independently. This invention does not impose any specific limitations on this.

[0031] S22. The target power data is upsampled using an interpolation upsampling method to obtain second upsampled data.

[0032] Upsampling techniques based on interpolation insert new data points between original data points using specific interpolation algorithms, thereby expanding the data length. The core of this method lies in utilizing the distribution patterns of known data points to estimate the values ​​of unknown points using mathematical interpolation formulas. The implementation process requires no parameter learning; it directly calculates the values ​​of the newly added points according to preset interpolation rules.

[0033] As one possible implementation, bicubic interpolation can be used to upsample the target power data. Bicubic interpolation uses weighted interpolation of the values ​​from 16 points surrounding the point to be filled to obtain the value of the point to be filled. Specifically, bicubic interpolation uses a cubic polynomial to approximate the theoretically optimal interpolation function, the mathematical expression of which is:

[0034] in, It is an adjustment factor that controls the smoothness and sharpness of interpolation. Indicates the relative position from the center integer point. .

[0035] S23. The first upsampled data and the second upsampled data are fused to obtain the input data corresponding to the target power data.

[0036] The fusion operation aims to combine the advantages of both upsampling methods. Bicubic interpolation is advantageous because it is a non-learning method, requiring no training parameters. It uses the weighted values ​​of its 16 nearest neighbors for calculation, resulting in smoother edges than bilinear interpolation. Furthermore, it can quickly reconstruct the structural contours and overall trend of an image. Furthermore, the algorithm is simple, computationally inexpensive, and suitable for rapid preprocessing on edge devices. It boasts strong structural restoration capabilities, stable reconstruction, and low computational cost. The advantage of deconvolutional upsampling lies in its ability to adaptively recover image details. It can be embedded into multi-layer convolutional neural networks to learn complex mapping relationships and adapt to different devices or scenarios. It exhibits strong detail compensation capabilities, high expressiveness, and can automatically optimize the upsampling process. Bicubic interpolation is responsible for restoring the overall structure, while deconvolution supplements detail changes; the fusion of these two methods achieves a balance between stability and expressiveness.

[0037] As one possible implementation, deconvolution and interpolation upsampling can be performed on the same target power data separately to obtain two upsampling results of the same size but different features. Then, by concatenating the two data into a dual-channel feature map, the subsequent convolutional network can learn the global information provided by interpolation and the detailed features recovered by deconvolution in a unified input, thereby improving the overall modeling accuracy and robustness.

[0038] Then, the target state perception model can be used to perform state perception of the corresponding key node equipment based on the input data. This target state perception model includes a dual-channel feature extraction layer and a hierarchical attention mechanism layer to extract features at different levels from the input data. This target state perception model can be trained based on historical power data and its state labels in the power communication network. For example, historical power data can be input into the initial state perception model, and the state prediction results output by the state perception model can be obtained. The difference between the historical power data state labels and the state prediction results can be calculated using loss functions such as cross-entropy. Based on this difference, the parameters of the state perception model can be adjusted through backpropagation until the difference converges, thus obtaining the target state perception model.

[0039] In one possible embodiment, the dual-channel feature extraction layer includes a shallow feature extraction sub-layer and a deep feature extraction sub-layer; the shallow feature extraction sub-layer contains multiple convolutional layers with different kernel sizes, and the deep feature extraction sub-layer contains multiple deep feature extraction units, each of which contains multiple convolutional layers with an increasing number of channels and a channel compression layer.

[0040] To effectively extract key feature information from input data within different receptive fields, multi-scale convolutional structures can be introduced into shallow feature extraction sub-layers. By setting multiple convolutional kernel sizes and extracting features in parallel, joint modeling of local details and global structure can be achieved. The kernel size in each convolutional layer follows the principle of "odd-number increment + complementary receptive fields + computational balance," meaning that the difference between the kernel sizes of adjacent convolutional layers is 2 or an odd number. This ensures that each size serves different levels of feature perception targets, guaranteeing structural rationality and clear feature division.

[0041] As one possible implementation, the convolutional layers in the shallow feature extraction sublayer can be configured with the following dimensions: 3×3 convolutional kernels are suitable for extracting fine-grained local features, with a small receptive field and fewer parameters, which can reduce computation and storage overhead while ensuring representational ability; 5×5 convolutional kernels have a medium receptive field and can cover a wide context region, making them suitable for recognizing complex edge structures or local patterns; 7×7 convolutional kernels can model global structural information at a larger scale, and are particularly advantageous when dealing with macroscopic state features such as long periods and power fluctuation trends.

[0042] The above convolutional layers are connected in parallel to extract shallow features of different sizes from the input data. That is, the above method can also include using multiple convolutional layers with different kernel sizes to extract shallow features of different sizes from the input and output respectively.

[0043] In one possible implementation, the shallow feature extraction sublayer may also contain a 1×1 convolutional layer, i.e., a convolutional layer with a kernel size of 1×1. The 1×1 convolutional kernel is mainly used for feature compression or expansion in the channel dimension, which can realize feature integration and dimensional adjustment, and enhance the interaction capability between channels.

[0044] In one possible implementation, features of different sizes obtained after all convolution operations are connected to the ReLU activation function to enhance the nonlinear expressive power of shallow features and prevent the degradation of deep structures.

[0045] Deep feature extraction consists of m deep feature extraction units, which are connected in series to extract deep features from the input data. The number m of these deep feature extraction units can be selected according to the actual application scenario. For example, the size of m can be determined by the task complexity, network channel transformation strategy, and computational resource constraints. The setting of m should ensure that the model has sufficient perceptual depth and expressive power, and facilitate the gradual guidance of the feature map towards the target output form. It is usually an even number, generally chosen between 4 and 10.

[0046] When input features enter the deep feature extraction unit, the channel dimension is first dynamically adjusted through two-dimensional convolution operations. In the front end of the network, basic units employ a channel expansion strategy, typically increasing the number of channels by a factor of 2 to enhance feature representation. In the back end of the network, the number of channels is gradually compressed. This refines high-order features, helps control model complexity, and gradually guides the feature map back to the shape of a single power image, facilitating subsequent super-resolution reconstruction. This guidance includes structural guidance: controlling the output feature dimension to gradually approach the target image through channel compression and upsampling operations; task guidance: imposing dual constraints on shape and semantics during training using reconstruction or classification loss functions; and parameter guidance: setting structural parameters such as stride and kernel size to simultaneously revert the feature map size and semantic content to the target image form.

[0047] After each convolutional layer, a ReLU activation function is introduced as a non-linear mapping to prevent the model from degenerating into a linear structure under multiple stacked layers. Since convolution is essentially a linear transformation, the introduction of the activation function enables the network to handle non-linear feature relationships, thereby improving the model's ability to perceive and express complex state information in power images.

[0048] like Figure 2 As shown, Figure 2 This is a schematic diagram of a deep feature extraction unit provided in an embodiment of the present invention. The input of the deep feature extraction unit is a feature map of size C / 2, W, H (number of channels C, width W, height H). First, the feature map is input to a two-dimensional convolutional layer. The two-dimensional convolutional layer performs a regular 2D convolution on C / 2, W, H, and then activates the convolutional features obtained by the convolution through the ReLU activation function, that is, ReLU is performed on the convolutional output feature map element by element. Then, the activated feature map is input to the attention layer for calculation to redistribute the channel or spatial weights to obtain a feature map of size C, W, H. The feature map is processed by two efficient convolutional layers to obtain the output features of the deep feature extraction unit.

[0049] The input to the efficient convolutional layer is a feature map of size C, W, and H. It is first split into two branches by branch splitting (S operation), and the number of channels in each branch becomes C / 2, thus obtaining two feature maps of size C / 2, W, and H.

[0050] For one branch, a 1×1 2D convolutional layer is used to expand the input channels, resulting in an increased-dimensional feature. This increased-dimensional feature is then activated using a ReLU activation function. The activated feature is then input into a 3×3 DW convolutional layer, which performs channel-wise convolution on this feature, meaning each channel is convolved with a separate 3×3 kernel. The convolutional feature is then input into a ReLU activation function for element-wise ReLU, and finally into a 1×1 2D convolutional layer to reduce the dimensionality of the feature, adjusting its size to C / 2, W, H. This feature is then concatenated with the other branch obtained from the initial split (size C / 2, W, H) to obtain a concatenated feature of size C, W, H. This concatenated feature is then rearranged by rearranging the channel dimensions of the concatenated C, W, H (e.g., shuffling the channel order or grouping and reorganizing), with the aim of better integrating the features learned from different branches.

[0051] After extracting features from the shallow and deep paths, shallow feature maps are obtained respectively. and deep feature maps To ensure structural consistency and complementarity between the two types of features in the subsequent fusion stage, the number of channels can be unified for both. In one possible embodiment, the dual-channel feature extraction layer further includes a formatted convolutional layer with a kernel size of 1×1; the method further includes: The shallow features and the deep features are respectively input into the formatted convolutional layer to obtain shallow features and deep features of the same dimension.

[0052] Using 1×1 convolution to adjust shallow and deep features to the same dimension can be achieved using the following formula:

[0053] The two feature maps after alignment and The input pairs are passed to the subsequent hierarchical channel attention mechanism layer. During the fusion process, unified channel attention weights are generated through interactive modeling and jointly weighted to further enhance the response intensity of key state features.

[0054] In one possible embodiment, the hierarchical channel attention mechanism layer includes a shallow attention mechanism layer, a deep attention mechanism layer, and a weighted fusion layer; The shallow attention mechanism layer is used to output shallow attention features for the shallow features; The deep attention mechanism layer is used to output deep attention features for the deep features; The weighted fusion layer is used to generate a channel weight vector based on the shallow attention features, deep attention features, and preset parameters; and to fuse the shallow attention features and deep attention features using the channel weight vector to obtain the target fusion feature.

[0055] Both the shallow and deep attention mechanisms described above are built upon the ECA (Efficient Channel Attention) mechanism. For example, the shallow and deep attention mechanisms can respectively apply the following formulas to the shallow features... and deep features Attention calculations are performed to obtain shallow attention features. and deep attention features :

[0056] The obtained shallow and deep attention features can be input into the weighted fusion layer. The weighted fusion layer can calculate the channel weight vector based on the above features. For example, the channel weight vector can be calculated using the following formula. :

[0057] in, , , For adjustable weight parameters, This indicates element-wise channel multiplication. This is a non-linear activation function. The channel weight vector is actually the overall weight scaling factor for each channel; its physical meaning is an assessment of the importance of the feature patterns extracted by that channel in the state-aware task.

[0058] The above weight vector Simultaneously affects shallow feature maps and deep feature maps The target fusion feature can be obtained by concatenating the channel dimensions. For example, the target fusion feature can be calculated through the following steps:

[0059] Subsequent steps can be based on target fusion features The system outputs the status perception of key node devices, completing the task of sensing the operating status of power equipment. For example, the target status perception model can also include a lightweight convolutional layer, which further extracts local discriminative features from the target fusion features and incorporates a ReLU activation function to enhance the model's nonlinear expressive power. Subsequently, global average pooling can be used to compress the 3D feature map into a channel-dimensional vector. This is used to capture the global state response. The vector is fed into a fully connected layer for state mapping, and depending on the application scenario, either a Softmax or a Sigmoid activation function is selected to output the device's target state perception result. This state perception result can be a state category or a health score. For example, the target state perception result can be shown in the table below: Table 1. Examples of State Awareness Results

[0060] In this embodiment of the invention, the operational data collected from key nodes of the power communication network is first normalized and imputed, and then converted into a two-dimensional pseudo-image format as input to the perception network. Parallel interpolation upsampling and deconvolution learning are then performed to reconstruct the structural contours and enhance the details of the original input, fusing the two upsampling results to form a higher-quality feature map. Subsequently, parallel shallow and deep feature extraction channels are constructed to simultaneously capture local texture and global semantic information. A hierarchical channel attention fusion mechanism is introduced, calculating channel attention weights for shallow and deep features respectively, improving the responsiveness of state-sensitive features. By modeling the operational data of key node equipment using pseudo-images, the state perception task is completed.

[0061] Based on the same inventive concept, this invention also provides a status awareness system for key node equipment in a power communication network, applied to edge nodes, such as... Figure 3 As shown, the system 300 may include: The acquisition module 301 is used to acquire target power data of key nodes in the power communication network. Upsampling module 302 is used to perform double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data; The input module 303 is used to input the input data into a pre-trained target state perception model. The target state perception model includes a dual-channel feature extraction layer and a hierarchical channel attention fusion mechanism layer. The dual-channel feature extraction layer extracts shallow and deep features from the input data. The hierarchical channel attention fusion mechanism layer performs attention calculations on the shallow and deep features respectively to obtain corresponding shallow and deep attention features. It then fuses the shallow and deep attention features based on preset channel weights to obtain target fusion features. The preset channel weights are used to identify the importance of different channels in the shallow and deep attention features. The output module 304 is used to obtain the target state perception result output by the target state perception model based on the target fusion features.

[0062] In one possible embodiment, the step of performing double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data includes: The target power data is upsampled using a deconvolution upsampling method to obtain the first upsampled data. The target power data is upsampled using an interpolation upsampling method to obtain second upsampled data. The first upsampled data and the second upsampled data are fused to obtain the input data corresponding to the target power data.

[0063] In one possible embodiment, the dual-channel feature extraction layer includes a shallow feature extraction sub-layer and a deep feature extraction sub-layer; the shallow feature extraction sub-layer contains multiple convolutional layers with different kernel sizes, and the deep feature extraction sub-layer contains multiple deep feature extraction units, each of which contains multiple convolutional layers with an increasing number of channels and a channel compression layer; The dual-channel feature extraction layer also includes a formatted convolutional layer with a kernel size of 1×1; the target state perception module is further used to input the shallow features and the deep features into the formatted convolutional layer respectively to obtain shallow features and deep features with the same dimension. The hierarchical channel attention mechanism layer includes a shallow attention mechanism layer, a deep attention mechanism layer, and a weighted fusion layer; The shallow attention mechanism layer is used to output shallow attention features for the shallow features; The deep attention mechanism layer is used to output deep attention features for the deep features; The weighted fusion layer is used to generate a channel weight vector based on the shallow attention features, deep attention features, and preset parameters; and to fuse the shallow attention features and deep attention features using the channel weight vector to obtain the target fusion feature.

[0064] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in this invention comply with relevant laws and regulations and do not violate public order and good morals.

[0065] An exemplary embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the electronic device to perform a method according to an embodiment of the present invention.

[0066] An exemplary embodiment of the present invention also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of the present invention.

[0067] An exemplary embodiment of the present invention also provides a computer program product, including a computer program, wherein, when executed by a computer's processor, the computer program is used to cause the computer to perform a method according to an embodiment of the present invention.

[0068] refer to Figure 4The present invention will now be described in the form of a structural block diagram of an electronic device 400 that can serve as a server or client of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0069] like Figure 4 As shown, the electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. The RAM 403 may also store various programs and data required for the operation of the electronic device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0070] Multiple components in electronic device 400 are connected to I / O interface 405, including: input unit 406, output unit 407, storage unit 408, and communication unit 409. Input unit 406 can be any type of device capable of inputting information to electronic device 400. Input unit 406 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 407 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 408 may include, but is not limited to, disks and optical discs. Communication unit 409 allows electronic device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth™ devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0071] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above. For example, in some embodiments, the power communication network critical node equipment status awareness method described above can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 400 via ROM 402 and / or communication unit 409. In some embodiments, the computing unit 401 can be configured by any other suitable means (e.g., by means of firmware) to perform the power communication network critical node equipment status awareness method described above.

[0072] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0073] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0074] As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0075] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0076] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0077] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

Claims

1. A method for status perception of key node equipment in a power communication network, characterized in that, Applied to edge nodes, the method includes: Acquire target power data of key nodes in the power communication network; The target power data is double-upsampled using a fusion upsampling method to obtain the input data corresponding to the target power data; The input data is fed into a pre-trained target state perception model, which includes a dual-channel feature extraction layer and a hierarchical channel attention fusion mechanism layer. The dual-channel feature extraction layer is used to extract shallow and deep features from the input data. The hierarchical channel attention fusion mechanism layer is used to perform attention calculations on the shallow and deep features respectively to obtain corresponding shallow attention features and deep attention features. The shallow and deep attention features are then fused based on preset channel weights to obtain target fusion features. The preset channel weights are used to identify the importance of different channels in the shallow and deep attention features. Obtain the target state perception result output by the target state perception model based on the target fusion features.

2. The method according to claim 1, characterized in that, The step of performing double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data includes: The target power data is upsampled using a deconvolution upsampling method to obtain the first upsampled data. The target power data is upsampled using an interpolation upsampling method to obtain second upsampled data. The first upsampled data and the second upsampled data are fused to obtain the input data corresponding to the target power data.

3. The method according to claim 1, characterized in that, The dual-channel feature extraction layer includes a shallow feature extraction sub-layer and a deep feature extraction sub-layer; the shallow feature extraction sub-layer contains multiple convolutional layers with different kernel sizes, and the deep feature extraction sub-layer contains multiple deep feature extraction units, each of which contains multiple convolutional layers with an increasing number of channels and a channel compression layer.

4. The method according to claim 3, characterized in that, The dual-channel feature extraction layer further includes a formatted convolutional layer with a kernel size of 1×1; the method further includes: The shallow features and the deep features are respectively input into the formatted convolutional layer to obtain shallow features and deep features of the same dimension.

5. The method according to claim 1, characterized in that, The hierarchical channel attention mechanism layer includes a shallow attention mechanism layer, a deep attention mechanism layer, and a weighted fusion layer; The shallow attention mechanism layer is used to output shallow attention features for the shallow features; The deep attention mechanism layer is used to output deep attention features for the deep features; The weighted fusion layer is used to generate a channel weight vector based on the shallow attention features, deep attention features, and preset parameters; and to fuse the shallow attention features and deep attention features using the channel weight vector to obtain the target fusion feature.

6. A status sensing system for key node equipment in a power communication network, characterized in that, The system, applied to edge nodes, includes: The acquisition module is used to acquire target power data of key nodes in the power communication network. The upsampling module is used to perform double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data; An input module is used to input the input data into a pre-trained target state perception model. The target state perception model includes a dual-channel feature extraction layer and a hierarchical channel attention fusion mechanism layer. The dual-channel feature extraction layer extracts shallow and deep features from the input data. The hierarchical channel attention fusion mechanism layer performs attention calculations on the shallow and deep features respectively to obtain corresponding shallow and deep attention features. It then fuses the shallow and deep attention features based on preset channel weights to obtain target fusion features. The preset channel weights are used to identify the importance of different channels in the shallow and deep attention features. The output module is used to obtain the target state perception result output by the target state perception model based on the target fusion features.

7. The system according to claim 6, characterized in that, The step of performing double upsampling on the target power data using a fusion upsampling method to obtain the input data corresponding to the target power data includes: The target power data is upsampled using a deconvolution upsampling method to obtain the first upsampled data. The target power data is upsampled using an interpolation upsampling method to obtain second upsampled data. The first upsampled data and the second upsampled data are fused to obtain the input data corresponding to the target power data.

8. The system according to claim 6, characterized in that, The dual-channel feature extraction layer includes a shallow feature extraction sub-layer and a deep feature extraction sub-layer; the shallow feature extraction sub-layer contains multiple convolutional layers with different kernel sizes, and the deep feature extraction sub-layer contains multiple deep feature extraction units, each of which contains multiple convolutional layers with an increasing number of channels and a channel compression layer; The dual-channel feature extraction layer also includes a formatted convolutional layer with a kernel size of 1×1; the target state perception module is further used to input the shallow features and the deep features into the formatted convolutional layer respectively to obtain shallow features and deep features with the same dimension. The hierarchical channel attention mechanism layer includes a shallow attention mechanism layer, a deep attention mechanism layer, and a weighted fusion layer; The shallow attention mechanism layer is used to output shallow attention features for the shallow features; The deep attention mechanism layer is used to output deep attention features for the deep features; The weighted fusion layer is used to generate a channel weight vector based on the shallow attention features, deep attention features, and preset parameters; and to fuse the shallow attention features and deep attention features using the channel weight vector to obtain the target fusion feature.

9. An electronic device, comprising: processor; as well as Stored program memory, The program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1-5.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

Images