Acquisition method of power grid multi-source cross-domain data fusion model, power grid multi-source cross-domain data fusion method and device, computer equipment and readable storage medium

Abstract

The invention relates to the technical field of data fusion, and provides a power grid multi-source cross-domain data fusion model acquisition method, a power grid multi-source cross-domain data fusion method, a power grid multi-source cross-domain data fusion device, computer equipment and a readable storage medium. The method comprises the following steps: acquiring a dense connection convolutional network model; modifying a loss function of the dense connection convolutional network model into a regularization loss function based on a total variation norm to obtain a to-be-trained model; and according to the power grid multi-source cross-domain data sample, performing model training on the to-be-trained model to obtain a power grid multi-source cross-domain data fusion model. By adopting the method, information and modes in the multi-source cross-domain data of the power grid can be more deeply mined and utilized, so that the multi-source cross-domain data fusion model of the power grid can better perform data fusion on data of different sources.

Description

Technical Field

[0001] The present application relates to the technical field of data fusion, and in particular, to a method for obtaining a power grid multi-source cross-domain data fusion model, a power grid multi-source cross-domain data fusion method, a device, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In a complex urban power grid scenario, traditional data fusion methods usually integrate power grid data from different sources according to a unified data model. For example, pattern mapping and duplicate detection are used to ensure the consistency of data in various parts of the power grid in the database.

[0003] However, traditional data fusion methods often cannot fully utilize the complex relationships and non-linear characteristics existing in power grid multi-source cross-domain data, and it is difficult to capture the implicit information and complex patterns in power grid multi-source cross-domain data. This results in insufficient and inaccurate feature representations in the data fusion results corresponding to power grid multi-source cross-domain data, further making it difficult to fully mine and analyze power grid multi-source cross-domain data, and restricting the application of the data fusion results corresponding to power grid multi-source cross-domain data in accurate and efficient prediction. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method for obtaining a power grid multi-source cross-domain data fusion model, a power grid multi-source cross-domain data fusion method, a device, a computer device, a computer-readable storage medium, and a computer program product.

[0005] In a first aspect, the present application provides a method for obtaining a power grid multi-source cross-domain data fusion model, including:

[0006] Obtaining a densely connected convolutional network model;

[0007] Modifying the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm to obtain a model to be trained;

[0008] Training the model to be trained according to power grid multi-source cross-domain data samples to obtain a power grid multi-source cross-domain data fusion model.

[0009] In one embodiment, the step of modifying the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm to obtain a model to be trained includes:

[0010] Modifying the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm to obtain an intermediate model;

[0011] Adding an attention mechanism to the intermediate model to obtain a model to be trained.

[0012] In one embodiment, before modifying the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm to obtain an intermediate model, the method further includes:

[0013] Discretize the total variation norm to obtain an optimized total variation norm;

[0014] Obtain a total variation norm regularization term according to the regularization parameter and the optimized total variation norm;

[0015] Obtain a regularization loss function based on the total variation norm according to the data loss function and the total variation norm regularization term.

[0016] In one embodiment, before training the model to be trained according to the power grid multi-source cross-domain data samples to obtain a power grid multi-source cross-domain data fusion model, the method further includes:

[0017] Collect the location information, operating status information, image information, daily maintenance records, repair records, and upgrade records of several power grid devices;

[0018] Collect the records, types, occurrence times, and fault handling results of several power grid device failures;

[0019] Obtain power grid multi-source cross-domain data samples according to the location information, operating status information, image information, daily maintenance records, repair records, and upgrade records of several power grid devices, and the records, types, occurrence times, and fault handling results of several power grid device failures.

[0020] In one embodiment, the training the model to be trained according to the power grid multi-source cross-domain data samples to obtain a power grid multi-source cross-domain data fusion model includes:

[0021] Input the power grid multi-source cross-domain data samples into the model to be trained to obtain a data fusion result corresponding to the power grid multi-source cross-domain data samples;

[0022] Obtain a structural similarity index between the data fusion result and the data fusion result sample according to the data fusion result corresponding to the power grid multi-source cross-domain data samples and the data fusion result sample;

[0023] When the structural similarity index is greater than a preset threshold, stop the model training and use the model to be trained as the power grid multi-source cross-domain data fusion model.

[0024] In a second aspect, the present application further provides a power grid multi-source cross-domain data fusion method, and the method includes:

[0025] Obtain power grid multi-source cross-domain data;

[0026] Input the multi-source cross-domain data of the power grid into the power grid multi-source cross-domain data fusion model to obtain the data fusion result corresponding to the multi-source cross-domain data of the power grid; the power grid multi-source cross-domain data fusion model is a model obtained according to the embodiments of the method for obtaining the power grid multi-source cross-domain data fusion model.

[0027] In a third aspect, the present application also provides an apparatus for obtaining a power grid multi-source cross-domain data fusion model, including:

[0028] A network model acquisition module, configured to acquire a densely connected convolutional network model;

[0029] A to-be-trained model acquisition module, configured to modify the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm to obtain a to-be-trained model;

[0030] A data fusion model acquisition module, configured to perform model training on the to-be-trained model according to power grid multi-source cross-domain data samples to obtain a power grid multi-source cross-domain data fusion model.

[0031] In a fourth aspect, the present application also provides a power grid multi-source cross-domain data fusion apparatus, including:

[0032] A multi-source cross-domain data acquisition module, configured to acquire power grid multi-source cross-domain data;

[0033] A data fusion result acquisition module, configured to input the power grid multi-source cross-domain data into the power grid multi-source cross-domain data fusion model to obtain the data fusion result corresponding to the power grid multi-source cross-domain data; the power grid multi-source cross-domain data fusion model is a model obtained according to the embodiments of the method for obtaining the power grid multi-source cross-domain data fusion model.

[0034] In a fifth aspect, the present application also provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor executes the above method.

[0035] In a sixth aspect, the present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to perform the above method.

[0036] In a seventh aspect, the present application also provides a computer program product. The computer program product includes a computer program, and the computer program is executed by a processor to perform the above method.

[0037] This application obtains a densely connected convolutional network model; modifies the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm to obtain a model to be trained; and trains the model to be trained according to power grid multi-source cross-domain data samples to obtain a power grid multi-source cross-domain data fusion model. This application modifies the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm to obtain a model to be trained, so as to obtain a power grid multi-source cross-domain data fusion model, enabling the power grid multi-source cross-domain data fusion model to more deeply mine and utilize the information and patterns in the power grid multi-source cross-domain data, thereby significantly improving the data fusion effect of the power grid multi-source cross-domain data fusion model and enabling the power grid multi-source cross-domain data fusion model to better perform data fusion on data from different sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0039] Figure 1 It is an application environment diagram of the method for obtaining a power grid multi-source cross-domain data fusion model and the power grid multi-source cross-domain data fusion method in one embodiment;

[0040] Figure 2 It is a schematic flowchart of the method for obtaining a power grid multi-source cross-domain data fusion model in one embodiment;

[0041] Figure 3 It is a schematic structural diagram of a model to be trained in one embodiment;

[0042] Figure 4 It is a schematic network structure diagram of an attention mechanism in one embodiment;

[0043] Figure 5 It is a schematic flowchart of the power grid multi-source cross-domain data fusion method in one embodiment;

[0044] Figure 6 It is a schematic flowchart of the method for obtaining a power grid multi-source cross-domain data fusion model in another embodiment;

[0045] Figure 7 It is a block diagram of the structure of an apparatus for obtaining a power grid multi-source cross-domain data fusion model in one embodiment;

[0046] Figure 8 It is a block diagram of the structure of a power grid multi-source cross-domain data fusion apparatus in one embodiment;

[0047] Figure 9 It is the internal structure diagram of a computer device in an embodiment. Specific implementation manners

[0048] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] The method for obtaining a power grid multi-source cross-domain data fusion model provided by an embodiment of the present application can be applied to, for example Figure 1 the application environment shown. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed in the cloud or other network servers. The terminal 102 can obtain a densely connected convolutional network model to obtain a power grid multi-source cross-domain data fusion model. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0050] In an exemplary embodiment, as Figure 2 shown, a method for obtaining a power grid multi-source cross-domain data fusion model is provided. Taking the method applied to Figure 1 the terminal 102 in it as an example for illustration, it includes the following steps S201 to step S203. Among them:

[0051] Step S201, obtain a densely connected convolutional network model.

[0052] The densely connected convolutional network (Densely Connected Convolutional Networks, DenseNet) model is a densely connected convolutional neural network architecture. The core idea is to enhance feature transmission and reuse through dense connection (Dense Connection), thereby improving the performance of the model and reducing the number of parameters.

[0053] Step S202, modify the loss function of the densely connected convolutional network model to a regularization loss function based on the total variation norm to obtain a model to be trained.

[0054] The regularization loss function based on the total variation norm introduces a smoothness constraint during the optimization process and retains important structural features, and is applicable to various image processing tasks.

[0055] The loss function of the densely connected convolutional network model can be modified into a regularization loss function based on the total variation norm to obtain the model to be trained, which can effectively promote the smoothness of the solution learned by the model to be trained and suppress the influence of noise. The structure of the model to be trained is as Figure 3 shown, where DenseLayer is the densely connected layer, which is the sub-network structure of the densely connected convolutional network. Conv represents the convolution operation, and CBAM is the convolutional block attention module.

[0056] Step S203: According to the power grid multi-source cross-domain data samples, train the model to be trained to obtain a power grid multi-source cross-domain data fusion model.

[0057] Power grid device data can be obtained from multiple data sources to obtain power grid multi-source cross-domain data samples.

[0058] According to the power grid multi-source cross-domain data samples, the model to be trained can be trained, and the structural similarity index (Structural Similarity Index Measure, SSIM) can be used to evaluate the model to be trained to obtain a power grid multi-source cross-domain data fusion model.

[0059] In the above method for obtaining the power grid multi-source cross-domain data fusion model, the loss function of the densely connected convolutional network model is modified into a regularization loss function based on the total variation norm to obtain the model to be trained, so as to obtain the power grid multi-source cross-domain data fusion model, enabling the power grid multi-source cross-domain data fusion model to more deeply mine and utilize the information and patterns in the power grid multi-source cross-domain data, thereby significantly improving the data fusion effect of the power grid multi-source cross-domain data fusion model and enabling the power grid multi-source cross-domain data fusion model to better perform data fusion on data from different sources.

[0060] In one embodiment, the loss function of the densely connected convolutional network model is modified into a regularization loss function based on the total variation norm to obtain the model to be trained. The specific steps are as follows: The loss function of the densely connected convolutional network model is modified into a regularization loss function based on the total variation norm to obtain an intermediate model; an attention mechanism is added to the intermediate model to obtain the model to be trained.

[0061] The attention mechanism is a technology widely used in deep learning, which can enhance the model's understanding of data context information. The network structure of the attention mechanism is as Figure 4 shown, including a channel attention model and a spatial attention model, where MLP is a multi-layer perceptron (Multi-Layer Perceptron), and sigmoid is the S-shaped growth curve activation function.

[0062] The Channel Attention Model is generated from the color channel relationships of the input features. Since the input feature dimension is large and the calculation is complex, it is necessary to compress the spatial dimension of the input features through pooling operations. Through average pooling operations, the target distribution range is understood; through max pooling operations, the target features are collected, and the average pooling operation and the max pooling operation are used to aggregate the spatial information and context representation in the feature map.

[0063] Average pooling helps the model understand the overall distribution of features, while max pooling helps capture the most prominent features in the feature map. These two operations can be used in parallel so that the model can learn both global context information and local prominent features simultaneously.

[0064] In the channel attention mechanism, the following steps are usually used to aggregate the spatial information and context representation in the feature map:

[0065] (1) Apply average pooling and max pooling to the input feature map to obtain two compressed feature maps respectively.

[0066] (2) Concatenate these two feature maps along the channel dimension.

[0067] (3) Process the concatenated feature map through a shared convolutional layer (or fully connected layer), and the purpose of this convolutional layer is to learn the dependencies between channels.

[0068] (4) Generate the final channel weights through the sigmoid activation function, and these weights represent the importance of different channels.

[0069] The calculation formula is shown in Equation (1):

[0070] (1)

[0071] where M c (F) represents the one-dimensional channel attention map calculated by the convolutional block attention module, σ represents the sigmoid activation function, represents the average pooling feature, represents the max pooling feature, AvgPool represents the average pooling operation, and maxPool represents the max pooling operation. and respectively represent the weights of the multi-layer perceptron, and the activation function used is the Linear rectification function (ReLU).

[0072] The Spatial Attention Model focuses on the spatial relationships in the feature map, that is, how features at different positions interact with each other. Different from the channel attention model, the spatial attention model focuses on understanding the spatial location information in the image, which helps the model better locate the objects or regions of interest.

[0073] When calculating the spatial attention map, the following steps are usually adopted:

[0074] (1) Pooling operation: Perform average pooling and max pooling on the feature map along the channel axis (usually the depth dimension of the feature map). The purpose of doing this is to capture the global context information in the spatial dimension while retaining important local features.

[0075] (2) Feature fusion: Concatenate the results after average pooling and max pooling in the spatial dimension to obtain a fused feature map with a richer representation.

[0076] (3) Convolution operation: Process the fused feature map through a standard convolutional layer (usually using a 3x3 convolutional kernel) to learn the spatial dependencies.

[0077] (4) Activation function: Use an activation function (such as sigmoid) to generate spatial attention weights, which represent the importance of different spatial positions.

[0078] The calculation method is shown in Equation (2).

[0079] (2)

[0080] Among them, M s (F) represents the two-dimensional spatial attention map calculated by the convolutional block attention module, σ represents the sigmoid activation function, f7*7 represents the convolutional operation with a 7*7 convolutional kernel, AvgPool represents the average pooling operation, maxPool represents the max pooling operation, represents the two-dimensional feature of the average pooling operation, represents the two-dimensional feature of the max pooling operation.

[0081] The formula of the entire convolutional attention mechanism CBAM is shown in Equation (3).

[0082] (3)

[0083] Among them, and are the one-dimensional channel attention map and the two-dimensional spatial attention map calculated by the Convolutional Block Attention Module (CBAM) respectively, represents the output of the channel attention model, represents the output of the final convolutional block attention model CBAM, and F represents the backbone feature map.

[0084] The loss function of the densely connected convolutional network model can be modified to a regularization loss function based on the total variation norm to obtain an intermediate model; an attention mechanism can be added to the intermediate model to obtain a model to be trained.

[0085] In this embodiment, the loss function of the densely connected convolutional network model is modified to a regularization loss function based on the total variation norm, and an attention mechanism is added to the densely connected convolutional network model to obtain a model to be trained, so that the power grid multi-source cross-domain data fusion model trained according to the model to be trained can more deeply mine and utilize the information and patterns in the power grid multi-source cross-domain data, thereby significantly improving the data fusion effect of the power grid multi-source cross-domain data fusion model.

[0086] In one of the embodiments, before modifying the loss function of the densely connected convolutional network model to a regularization loss function based on the total variation norm to obtain an intermediate model, the method provided in this application further includes: discretizing the total variation norm to obtain an optimized total variation norm; obtaining a total variation norm regularization term according to a regularization parameter and the optimized total variation norm; and obtaining a regularization loss function based on the total variation norm according to a data loss function and the total variation norm regularization term.

[0087] For a two-dimensional image, the total variation norm is usually defined as the sum of the total norms of the image gradients, and the total variation norm can be expressed as Equation (4).

[0088] (4)

[0089] where f(x; θ) represents the model prediction value, and L TV f(x; θ) represents the total variation norm of the model prediction value, represents the difference operator along the x-axis, represents the difference operator along the y-axis.

[0090] The total variation norm can be discretized by the gradient descent method, and the model parameter θ is iteratively updated to minimize the loss function, so as to obtain an optimized total variation norm and make the total variation norm applicable to numerical optimization. The discretized form of the total variation norm regularization term is shown in Equation (5).

[0091] (5)

[0092] where f(x; θ) represents the model prediction value, and L TV f(x; θ) represents the total variation norm of the model prediction value.

[0093] According to the regularization parameter and the optimized total variation norm, the total variation regularization term can be obtained.

[0094] The data loss function measures the difference between the model prediction value and the true value. Different loss functions can be selected as the data loss function according to the specific application scenario, such as mean-square error (MSE), cross-entropy loss, etc. In this embodiment, the mean-square error loss as shown in Equation (6) can be adopted as the data loss function.

[0095] (6)

[0096] where f(x;θ) represents the model prediction value, and L data f(x;θ) represents the data loss function of the model prediction value, y represents the true value, N represents the number of samples, and θ represents the model parameters.

[0097] The data loss function and the total variation regularization term can be combined to obtain the regularization loss function based on the total variation norm, as shown in Equation (7).

[0098] (7)

[0099] where represents the regularization parameter, and L data f(x;θ) represents the data loss function of the model prediction value, and L TV f(x;θ) represents the total variation norm of the model prediction value.

[0100] In this embodiment, the total variation norm is discretized to obtain the optimized total variation norm; according to the regularization parameter and the optimized total variation norm, the total variation regularization term is obtained; according to the data loss function and the total variation regularization term, the regularization loss function based on the total variation norm is obtained, preparing the data for obtaining the training model.

[0101] In one embodiment, before training a model to be trained based on multi-source cross-domain data samples of a power grid to obtain a multi-source cross-domain data fusion model of the power grid, the method provided by the present application further includes: collecting the location information, operating status information, image information, daily maintenance records, overhaul records, and upgrade records of a number of power grid devices; collecting the records, types, occurrence times, and fault handling results of a number of power grid device failures; and obtaining multi-source cross-domain data samples of the power grid based on the location information, operating status information, image information, daily maintenance records, overhaul records, and upgrade records of a number of power grid devices, and the records, types, occurrence times, and fault handling results of a number of power grid device failures.

[0102] The location information, operating status information, image information, daily maintenance records, overhaul records, and upgrade records of a number of power grid devices can be collected; the records, types, occurrence times, and fault handling results of a number of power grid device failures can be collected.

[0103] The collected location information, operating status information, image information, daily maintenance records, overhaul records, and upgrade records of a number of power grid devices, and the records, types, occurrence times, and fault handling results of a number of power grid device failures can be subjected to data cleaning and formatting processing to obtain multi-source cross-domain data samples of the power grid.

[0104] In this embodiment, obtaining multi-source cross-domain data samples of the power grid based on the location information, operating status information, image information, daily maintenance records, overhaul records, and upgrade records of a number of power grid devices, and the records, types, occurrence times, and fault handling results of a number of power grid device failures prepares the data for the model training of the multi-source cross-domain data fusion model of the power grid.

[0105] In one embodiment, training a model to be trained based on multi-source cross-domain data samples of a power grid to obtain a multi-source cross-domain data fusion model of the power grid specifically includes the following steps: inputting the multi-source cross-domain data samples into the model to be trained to obtain a data fusion result corresponding to the multi-source cross-domain data samples; obtaining a structural similarity index between the data fusion result and the data fusion result sample based on the data fusion result corresponding to the multi-source cross-domain data samples and the data fusion result sample; and when the structural similarity index is greater than a preset threshold, stopping the model training and using the model to be trained as the multi-source cross-domain data fusion model.

[0106] The structural similarity index is an index used to measure the similarity between two images, aiming to evaluate the image quality, especially to evaluate the degree of distortion in lossy compressed images. Compared with traditional indexes such as mean square error and peak signal-to-noise ratio (PSNR), the structural similarity index is closer to the perceptual characteristics of the human visual system.

[0107] The calculation of the structural similarity index is based on the following three comparison dimensions:

[0108] (1) Luminance Comparison: Compare the average luminance of two images.

[0109] (2) Contrast Comparison: Compare the contrast of two images, that is, the standard deviation.

[0110] (3) Structure Comparison: Compare the structures of two images, that is, the correlation between the two images.

[0111] Given two images x and y, the calculation formula of the structural similarity index is shown in Equation (8).

[0112] (8)

[0113] Where SSIM(x, y) is the structural similarity index between images x and y, μ x is the mean of image x, μ y are the means of image y respectively, σ x 2 is the variance of image x, σ y 2 are the variances of image y respectively, σ xy is the covariance of images x and y. c1 and c2 are constants introduced to avoid a zero denominator, usually taking c1 = (k1L) and c2 = (k2L), where L is the dynamic range of pixel values, and k1 and k2 are constants less than 1.

[0114] The multi-source cross-domain data samples of the power grid can be input into the model to be trained to obtain the corresponding data fusion results of the multi-source cross-domain data samples of the power grid; the structural similarity index between the data fusion results and the data fusion result samples corresponding to the multi-source cross-domain data samples of the power grid can be obtained according to the calculation formula of the structural similarity index; when the structural similarity index is greater than the preset threshold, stop the model training and use the model to be trained as the multi-source cross-domain data fusion model of the power grid.

[0115] In this embodiment, the structural similarity index between the data fusion results and the data fusion result samples corresponding to the multi-source cross-domain data samples of the power grid is obtained; when the structural similarity index is greater than the preset threshold, stop the model training and use the model to be trained as the multi-source cross-domain data fusion model of the power grid, and a better multi-source cross-domain data fusion model of the power grid can be obtained.

[0116] The multi-source cross-domain data fusion method for power grids provided by the embodiments of this application can be applied to, for example, Figure 1 the application environment shown. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed in the cloud or on other network servers. The terminal 102 can obtain multi-source cross-domain power grid data and input it into the multi-source cross-domain power grid data fusion model to obtain the data fusion result corresponding to the multi-source cross-domain power grid data. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0117] In an exemplary embodiment, as Figure 5 shown, a multi-source cross-domain data fusion method for power grids is provided. Taking the method applied to the Figure 1 terminal 102 in it as an example, it includes the following steps S501 and S502. Among them:

[0118] Step S501, obtain multi-source cross-domain power grid data.

[0119] It is possible to perform data cleaning and formatting on the location information, operating status information, image information, daily maintenance records, repair records, and upgrade records of several power grid devices collected, as well as the records, types, occurrence times, and fault handling results of several power grid device failures, to obtain multi-source cross-domain power grid data.

[0120] Step S502, input the multi-source cross-domain power grid data into the multi-source cross-domain power grid data fusion model to obtain the data fusion result corresponding to the multi-source cross-domain power grid data; the multi-source cross-domain power grid data fusion model is a model obtained according to the embodiment of the method for obtaining the multi-source cross-domain power grid data fusion model.

[0121] The multi-source cross-domain power grid data fusion model can be obtained according to the embodiment of the method for obtaining the multi-source cross-domain power grid data fusion model.

[0122] The multi-source cross-domain power grid data can be input into the multi-source cross-domain power grid data fusion model to obtain the data fusion result corresponding to the multi-source cross-domain power grid data.

[0123] In the above-mentioned power grid multi-source cross-domain data fusion method, an attention mechanism and a regularization loss function based on the total variation norm are added to the power grid multi-source cross-domain data fusion model, enabling the power grid multi-source cross-domain data fusion model to more deeply mine and utilize the information and patterns in the power grid multi-source cross-domain data, thereby significantly improving the data fusion effect of the power grid multi-source cross-domain data fusion model and making the corresponding data fusion results of the power grid multi-source cross-domain data more accurate and stable.

[0124] To better understand the above method, the following elaborates in detail a method for obtaining the power grid multi-source cross-domain data fusion model of this application and an application embodiment of the power grid multi-source cross-domain data fusion method.

[0125] In complex urban power grid scenarios, traditional data fusion methods usually integrate power grid data from different sources according to a unified data model. For example, through schema mapping and duplicate detection to ensure the consistency of detailed information of each part of the power grid in the database. These power grid data may include power load, equipment status, power grid topology and other data. Although these power grid data come from different systems, they describe the same characteristics of the power grid. However, in the big data era, there are deep correlations between multiple data sets generated in complex urban power grids. Although the operation data, meteorological data and environmental monitoring data of the urban power grid come from different data domains, these power grid data together reflect the comprehensive operation status of the urban power grid. When fusing these power grid data from different data domains, simple schema mapping and duplicate detection are no longer applicable. Therefore, adopting a more complex power grid multi-source cross-domain data fusion method has become an important way to solve these problems, which can better utilize the information and patterns in the power grid multi-source cross-domain data, improve the accuracy and stability of the model, and bring new possibilities and opportunities for the construction of complex urban power grids.

[0126] In the field of traditional data fusion, the common practice is to integrate power grid data from different sources according to a unified data model, which usually involves techniques such as schema mapping and duplicate detection. However, traditional data fusion methods often cannot fully utilize the complex relationships and non-linear features existing in the data. Especially when dealing with power grid data (power data), due to the characteristics of large scale and high dimension of power grid data, traditional data fusion methods often face limitations in terms of computing and storage resources.

[0127] In addition, traditional data fusion methods usually adopt simple network structures to extract features, failing to capture the implicit information and complex patterns in the multi-source cross-domain data of the power grid. This results in insufficient and inaccurate feature representations in the data fusion results corresponding to the multi-source cross-domain data of the power grid, further making it difficult to fully mine and analyze the multi-source cross-domain data of the power grid. Therefore, in order to more effectively process the multi-source cross-domain data of the power grid, it is necessary to develop new data fusion methods to better capture the complexity and non-linear characteristics of the multi-source cross-domain data of the power grid and improve the accuracy and sufficiency of feature representations in the data fusion results corresponding to the multi-source cross-domain data of the power grid.

[0128] When dealing with the complex multi-source cross-domain data of urban power grids and actual urban power grid application scenarios, traditional data fusion methods face various technical challenges in the power industry, which limit their applications in accurate and efficient prediction. To overcome these problems, in this embodiment, an attention mechanism and a regularization loss function based on the total variation norm are added to the multi-source cross-domain data fusion model of the power grid, enabling the multi-source cross-domain data fusion model of the power grid to more deeply mine and utilize the information and patterns in the multi-source cross-domain data of the power grid, thereby significantly enhancing the effect of the multi-source cross-domain data fusion model on the multi-source cross-domain data fusion of the power grid. The main problems solved in this embodiment are as follows:

[0129] (1) Traditional data fusion: In this embodiment, data fusion is performed on multi-source cross-domain data based on the DenseNet model. Feature enhancement is carried out by adding an attention mechanism to the DenseNet model. A regularization loss function based on the total variation norm is used, and the structural similarity index SSIM is adopted for index evaluation to obtain the multi-source cross-domain data fusion model of the power grid, solving the problem that traditional data fusion methods cannot fully utilize the complex relationships and non-linear characteristics existing in the data.

[0130] (2) Limited data processing capacity: In this embodiment, the neural scaling law is adopted to elaborate on the dynamic relationship between model performance and the amount of pre-trained data, which helps to reveal the interaction between the performance of the multi-source cross-domain data fusion model of the power grid and various parameters, and can effectively solve the problems of computational and storage resource limitations encountered by traditional technologies when dealing with large-scale and high-dimensional data, thus making it possible to deeply mine and analyze the multi-source cross-domain data of the power grid.

[0131] (3)Difficulty in feature acquisition: This embodiment utilizes the advantages of the densely connected convolutional network model and extends it by introducing an attention module to effectively capture the extensive dependencies in the multi-source cross-domain data of the power grid. This method solves the problem that traditional simple network structures may miss implicit information and complex patterns in the data during feature extraction, thereby improving the sufficiency and accuracy of feature representation. In this way, it can be ensured that the multi-source cross-domain data fusion model of the power grid can better understand and express the internal structure of the data.

[0132] This embodiment is based on the densely connected convolutional network model, adds an attention mechanism to the model, performs convolutional operations on different data sources, uses the method of feature splicing, uses a regularization loss function based on the total variation norm, and uses the structural similarity index for index evaluation to obtain the multi-source cross-domain data fusion model of the power grid, enabling the multi-source cross-domain data fusion model of the power grid to better perform data fusion on the multi-source cross-domain data from different sources, thereby making the fusion effect of the multi-source cross-domain data fusion model of the power grid more accurate and stable. The training process of the multi-source cross-domain data fusion model of the power grid includes Figure 6 the six steps shown, namely: data collection, data preprocessing, adding an attention mechanism and modifying the regularization loss function of the total variation norm, model training, and fusion result evaluation.

[0133] Step description:

[0134] Step 1 Data collection: Collect the location information, operation status information, image information, daily maintenance records, overhaul records, and upgrade records of different power grid devices, as well as the records, types, occurrence times, and fault handling results of power grid device failures.

[0135] Step 2 Data preprocessing: Clean and format the collected data to obtain multi-source cross-domain data samples of the power grid.

[0136] Step 3 Adding an attention mechanism: By adding an attention module, enhance the understanding of the data context information by the multi-source cross-domain data fusion model of the power grid. By modifying the loss function to a regularization loss function based on the total variation norm, it can effectively promote the smoothness of the solution learned by the model and suppress the influence of noise, obtaining an improved densely connected convolutional network model, that is, the training model.

[0137] Step 4 Model training: Use the model to be trained to train the multi-source cross-domain data samples of the power grid.

[0138] Step 5 Fusion result evaluation: Evaluate the fusion effect of the multi-source cross-domain data samples of the trained model to be trained.

[0139] The key technology of this application lies in improving the densely connected convolutional network, and separately inputting power grid multi-source cross-domain data samples into multiple improved densely connected convolutional networks for data fusion. The following is the detailed improved network structure and detailed description:

[0140] 1. Improved network structure

[0141] The improved network structure is as Figure 3 shown, where DenseLayer is the densely connected layer, which is a sub-network structure of the densely connected convolutional network, Conv represents the convolution operation, and CBAM is the convolutional block attention module.

[0142] 2. Adding attention mechanism

[0143] By adding an attention module, the power grid multi-source cross-domain data fusion model's understanding of data context information is enhanced. The network structure of the attention mechanism is as Figure 4 shown, including the channel attention model and the spatial attention model, where MLP is the Multi-Layer Perceptron, and sigmoid is the S-shaped growth curve activation function.

[0144] The Channel Attention Model is an important concept in deep learning, especially in Convolutional Neural Networks (CNN) for enhancing the model's understanding of the importance of feature channels. When dealing with high-dimensional input features, in order to reduce computational complexity, pooling operations are usually used to reduce the spatial dimension. Average pooling and max pooling are two commonly used pooling methods, which play key roles in the channel attention mechanism.

[0145] The channel attention model is generated from the color channel relationship of the input features. Since the input feature dimension is large and the calculation is complex, it is necessary to compress the spatial dimension of the input features through pooling operations. Through average pooling operations, the target distribution range is understood; through max pooling operations, the target features are collected, and the average pooling operation and max pooling operation are used to aggregate the spatial information and context representation in the feature map.

[0146] Average pooling helps the model understand the overall distribution of features, while max pooling helps capture the most significant features in the feature map. These two operations can be used in parallel so that the model can learn both global context information and local significant features simultaneously.

[0147] In the channel attention mechanism, the following steps are usually used to aggregate the spatial information and context representation in the feature map:

[0148] (1) Apply average pooling and max pooling to the input feature map to obtain two compressed feature maps respectively.

[0149] (2) Concatenate these two feature maps along the channel dimension.

[0150] (3) Process the concatenated feature map through a shared convolutional layer (or fully connected layer), and the purpose of this convolutional layer is to learn the dependencies between channels.

[0151] (4) Generate the final channel weights through the sigmoid activation function, and these weights represent the importance of different channels.

[0152] The calculation formula is shown in Equation (1):

[0153] (1)

[0154] Where, M c (F) represents the one-dimensional channel attention map calculated by the convolutional block attention module, σ represents the sigmoid activation function, represents the average pooling feature, represents the max pooling feature, AvgPool represents the average pooling operation, and maxPool represents the max pooling operation. and respectively represent the weights of the multi-layer perceptron, and the activation function used is the rectified linear unit (ReLU).

[0155] The Spatial Attention Model focuses on the spatial relationships in the feature map, that is, how features at different positions interact with each other. Different from the channel attention model, the spatial attention model focuses on understanding the spatial location information in the image, which helps the model better locate the objects or regions of interest.

[0156] When calculating the spatial attention map, the following steps are usually adopted:

[0157] (1) Pooling operation: Perform average pooling and max pooling on the feature map along the channel axis (usually the depth dimension of the feature map). The purpose of doing this is to capture the global context information in the spatial dimension while retaining important local features.

[0158] (2) Feature fusion: Concatenate the results of average pooling and max pooling in the spatial dimension to obtain a fused feature map with a richer representation.

[0159] (3) Convolution operation: Process the fused feature map through a standard convolutional layer (usually using a 3x3 convolutional kernel) to learn spatial dependencies.

[0160] (4) Activation function: Use an activation function (such as sigmoid) to generate spatial attention weights, which represent the importance of different spatial positions.

[0161] The calculation method is shown in Equation (2).

[0162] (2)

[0163] where M s (F) represents the two-dimensional spatial attention map calculated by the convolutional block attention module, σ represents the sigmoid activation function, f7*7 represents the convolutional operation with a 7*7 convolutional kernel, AvgPool represents the average pooling operation, maxPool represents the max pooling operation, represents the two-dimensional feature of the average pooling operation, represents the two-dimensional feature of the max pooling operation.

[0164] The formula for the entire convolutional attention mechanism CBAM is shown in Equation (3).

[0165] (3)

[0166] where, and are the one-dimensional channel attention map and the two-dimensional spatial attention map calculated by the Convolutional Block Attention Module (CBAM) respectively, represents the output of the channel attention model, represents the output of the final convolutional block attention model CBAM, and F represents the backbone feature map.

[0167] 3. Regularization Loss Function Based on Total Variation Norm

[0168] Modify the loss function to a regularization loss function based on the total variation norm, which can effectively promote the smoothness of the solution learned by the power grid multi-source cross-domain data fusion model and suppress the influence of noise.

[0169] 3.1 Composition of the Regularization Loss Function

[0170] The total variation norm regularization loss function usually consists of two parts: the data loss function and the total variation regularization term.

[0171] 3.2 Data Loss

[0172] The data loss function measures the difference between the model's predicted values and the true values. Depending on the specific application scenario, different loss functions can be selected as the data loss function, such as mean-square error (MSE), cross-entropy loss, etc. In this embodiment, the mean-square error loss shown in Equation (6) can be adopted as the data loss function.

[0173] (6)

[0174] Among them, f(x; θ) represents the model's predicted value, and L data f(x; θ) represents the data loss function of the model's predicted value, y represents the true value, N represents the number of samples, and θ represents the model parameters.

[0175] 3.3 Total Variation Norm Regularization Term

[0176] The total variation norm regularization term is used to measure the smoothness of an image or a function. For a two-dimensional image, the total variation norm is usually defined as the sum of the total modulus of the image gradients. The total variation norm can be expressed as Equation (4).

[0177] (4)

[0178] Among them, f(x; θ) represents the model's predicted value, and L TV f(x; θ) represents the total variation norm of the model's predicted value,[[]] represents the difference operator along the x-axis,[[]] represents the difference operator along the y-axis.

[0179] 3.4 Total Variation Norm Regularization Loss Function

[0180] The data loss function and the total variation norm regularization term can be combined to obtain a regularization loss function based on the total variation norm, as shown in Equation (7).

[0181] (7)

[0182] Among them, represents the regularization parameter, and L data f(x; θ) represents the data loss function of the model's predicted value, and L TV f(x; θ) represents the total variation norm of the model's predicted value.

[0183] 3.5 Discretization and Optimization

[0184] In practical applications, it is usually necessary to discretize the total variation norm to make it applicable to numerical optimization. The discretized form of the total variation norm regularization term is shown in Equation (5).

[0185] (5)

[0186] Among them, f(x; θ) represents the model prediction value, and L TV f(x; θ) represents the total variation norm of the model prediction value.

[0187] Optimizing this loss function usually involves the gradient descent method or other optimization algorithms, and the model parameters θ are iteratively updated to minimize the loss function.

[0188] 4. Fusion result evaluation

[0189] The Structural Similarity Index (SSIM) is a metric used to measure the similarity between two images. The SSIM index was proposed by Wang et al. in 2004 with the aim of evaluating image quality, especially the degree of distortion in lossy compressed images. Compared with traditional metrics such as the Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR), SSIM is closer to the perceptual characteristics of the human visual system.

[0190] 4.1. Calculation principle of SSIM

[0191] The calculation of the SSIM index is based on the following three comparison dimensions:

[0192] (1) Luminance Comparison: Compare the average luminance of the two images.

[0193] (2) Contrast Comparison: Compare the contrast of the two images, that is, the standard deviation.

[0194] (3) Structure Comparison: Compare the structures of the two images, that is, the correlation between the two images.

[0195] 4.2. Calculation formula of SSIM

[0196] Given two images x and y, the calculation formula of the structural similarity index is shown in Equation (8).

[0197] (8)

[0198] Among them, SSIM(x, y) is the structural similarity index between images x and y, and μ x is the mean of image x, and μ yThey are the mean of image y, σ x 2 is the variance of image x, σ y 2 They are the variances of image y, σ xy is the covariance of images x and y. c1 and c2 are constants introduced to avoid a zero denominator, usually taking c1 = (k1L) and c2 = (k2L), where L is the dynamic range of pixel values and k1 and k2 are constants less than 1.

[0199] After training the power grid multi-source cross-domain data fusion model, the power grid multi-source cross-domain data to be fused can be obtained. Inputting this power grid multi-source cross-domain data into the power grid multi-source cross-domain data fusion model, the corresponding data fusion result of this power grid multi-source cross-domain data can be obtained.

[0200] In this embodiment, the power grid multi-source cross-domain data fusion model is added with an attention mechanism and a regularization loss function based on the total variation norm, enabling the power grid multi-source cross-domain data fusion model to more deeply mine and utilize the information and patterns in the power grid multi-source cross-domain data, thereby significantly improving the data fusion effect of the power grid multi-source cross-domain data fusion model and enabling the power grid multi-source cross-domain data fusion model to better perform data fusion on power grid data from different sources.

[0201] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential either, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0202] Based on the same inventive concept, the embodiment of the present application also provides an acquisition device for the power grid multi-source cross-domain data fusion model for implementing the acquisition method of the power grid multi-source cross-domain data fusion model involved above. The implementation solution for solving problems provided by this device is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more embodiments of the acquisition device for the power grid multi-source cross-domain data fusion model provided below can refer to the limitations on the acquisition method of the power grid multi-source cross-domain data fusion model in the above text, and will not be repeated here.

[0203] In an exemplary embodiment, as Figure 7As shown, a device for obtaining a multi-source cross-domain data fusion model of a power grid is provided, where:

[0204] A network model acquisition module 701, configured to acquire a densely connected convolutional network model;

[0205] A model to be trained acquisition module 702, configured to modify the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm, to obtain a model to be trained;

[0206] A data fusion model acquisition module 703, configured to perform model training on the model to be trained according to power grid multi-source cross-domain data samples, to obtain a power grid multi-source cross-domain data fusion model.

[0207] In one embodiment, the model to be trained acquisition module 702 is further configured to: modify the loss function of the densely connected convolutional network model into a regularization loss function based on the total variation norm, to obtain an intermediate model; add an attention mechanism to the intermediate model, to obtain a model to be trained.

[0208] In one embodiment, the device further includes a regularization loss function acquisition module, configured to: perform discretization processing on the total variation norm, to obtain an optimized total variation norm; obtain a total variation norm regularization term according to a regularization parameter and the optimized total variation norm; obtain a regularization loss function based on the total variation norm according to a data loss function and the total variation norm regularization term.

[0209] In one embodiment, the device further includes a data sample acquisition module, configured to: collect location information, operating status information, image information, daily maintenance records, repair records, and upgrade records of a plurality of power grid devices; collect records, types, occurrence times, and fault handling results of faults of a plurality of power grid devices; obtain power grid multi-source cross-domain data samples according to the location information, operating status information, image information, daily maintenance records, repair records, and upgrade records of a plurality of power grid devices, and the records, types, occurrence times, and fault handling results of faults of a plurality of power grid devices.

[0210] In one embodiment, the data fusion model acquisition module 703 is further configured to: input the power grid multi-source cross-domain data samples into the model to be trained, to obtain a data fusion result corresponding to the power grid multi-source cross-domain data samples; obtain a structural similarity index between the data fusion result and a data fusion result sample according to the data fusion result corresponding to the power grid multi-source cross-domain data samples and the data fusion result sample; when the structural similarity index is greater than a preset threshold, stop model training, and use the model to be trained as the power grid multi-source cross-domain data fusion model.

[0211] Each module in the above-mentioned acquisition device of the power grid multi-source cross-domain data fusion model can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above modules can be embedded in the processor of the computer device in the form of hardware or be independent of it, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0212] Based on the same inventive concept, an embodiment of the present application also provides a power grid multi-source cross-domain data fusion device for implementing the above-mentioned power grid multi-source cross-domain data fusion method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the power grid multi-source cross-domain data fusion device provided below can refer to the limitations on the power grid multi-source cross-domain data fusion method in the above text, and will not be repeated here.

[0213] In an exemplary embodiment, as Figure 8 shown, a power grid multi-source cross-domain data fusion device is provided, where:

[0214] A multi-source cross-domain data acquisition module 801, configured to acquire power grid multi-source cross-domain data;

[0215] A data fusion result acquisition module 802, configured to input the power grid multi-source cross-domain data into a power grid multi-source cross-domain data fusion model to obtain a data fusion result corresponding to the power grid multi-source cross-domain data; the power grid multi-source cross-domain data fusion model is a model obtained according to the embodiment of the acquisition method of the power grid multi-source cross-domain data fusion model.

[0216] Each module in the above-mentioned power grid multi-source cross-domain data fusion device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above modules can be embedded in the processor of the computer device in the form of hardware or be independent of it, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0217] In an exemplary embodiment, a computer device is provided. This computer device can be a server, and its internal structure diagram can be as Figure 9As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the data of the embodiments of the method for obtaining the power grid multi-source cross-domain data fusion model and the embodiments of the power grid multi-source cross-domain data fusion method. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for obtaining a power grid multi-source cross-domain data fusion model and a power grid multi-source cross-domain data fusion method.

[0218] Those skilled in the art can understand that Figure 9 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0219] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0220] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0221] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0222] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0223] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0224] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present application.

[0225] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for acquiring a multi-source cross-domain data fusion model of a power grid, characterized in that: The method comprises: Get a densely connected convolutional network model; Modifying the loss function of the densely connected convolutional network model into a regularized loss function based on the total variation norm to obtain a model to be trained; According to the power grid multi-source cross-domain data samples, the model to be trained is trained to obtain a power grid multi-source cross-domain data fusion model.

2. The method according to claim 1, characterized in that The loss function of the densely connected convolutional network model is modified to a regularized loss function based on the total variation norm to obtain a model to be trained, including: Modifying the loss function of the densely connected convolutional network model into a regularized loss function based on the total variation norm to obtain an intermediate model; Add an attention mechanism to the intermediate model to obtain a model to be trained.

3. The method according to claim 2, characterized in that Before modifying the loss function of the densely connected convolutional network model to a regularized loss function based on a total variation norm to obtain an intermediate model, the method further includes: Discretize the total variation norm to obtain the optimized total variation norm; According to the regularization parameter and the optimized total variation norm, a total variation norm regularization term is obtained; According to the data loss function and the total variation norm regularization term, a regularization loss function based on the total variation norm is obtained.

4. The method according to claim 1, characterized in that: Before performing model training on the model to be trained according to the power grid multi-source cross-domain data samples to obtain the power grid multi-source cross-domain data fusion model, the method further includes: Collect location information, operating status information, image information, daily maintenance records, repair records and upgrade records of several power grid equipment; Collect records, types, occurrence times and fault handling results of several power grid equipment faults; Based on the location information, operation status information, image information, daily maintenance records, inspection records and upgrade records of several power grid equipment, as well as the records, types, occurrence times and fault handling results of several power grid equipment failures, a multi-source cross-domain data sample of the power grid is obtained.

5. The method according to claim 1, characterized in that: The method of training the model to be trained according to the power grid multi-source cross-domain data samples to obtain a power grid multi-source cross-domain data fusion model includes: Inputting a multi-source cross-domain data sample of a power grid into a to-be-trained model to obtain a data fusion result corresponding to the multi-source cross-domain data sample of the power grid; According to the data fusion results and the data fusion result samples corresponding to the multi-source cross-domain data samples of the power grid, a structural similarity index between the data fusion results and the data fusion result samples is obtained; When the structural similarity index is greater than a preset threshold, the model training is stopped, and the model to be trained is used as a power grid multi-source cross-domain data fusion model.

6. A method for fusion of multi-source cross-domain data of power grid, characterized in that: The method comprises: Acquire multi-source cross-domain data of power grid; The power grid multi-source cross-domain data is input into a power grid multi-source cross-domain data fusion model to obtain a data fusion result corresponding to the power grid multi-source cross-domain data; the power grid multi-source cross-domain data fusion model is a model obtained according to the method according to any one of claims 1 to 5.

7. A device for acquiring a multi-source cross-domain data fusion model of a power grid, characterized in that: The device comprises: Network model acquisition module, used to obtain densely connected convolutional network models; A model acquisition module to be trained, used to modify the loss function of the densely connected convolutional network model into a regularized loss function based on the total variation norm to obtain the model to be trained; The data fusion model acquisition module is used to perform model training on the model to be trained according to the multi-source cross-domain data samples of the power grid to obtain the multi-source cross-domain data fusion model of the power grid.

8. A power grid multi-source cross-domain data fusion device, characterized in that: The device comprises: A multi-source cross-domain data acquisition module is used to acquire multi-source cross-domain data of the power grid; A data fusion result acquisition module is used to input the multi-source cross-domain data of the power grid into a multi-source cross-domain data fusion model of the power grid to obtain a data fusion result corresponding to the multi-source cross-domain data of the power grid; the multi-source cross-domain data fusion model of the power grid is a model obtained according to the method described in any one of claims 1 to 5.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images