Method and system for storing digital twin data of power transmission and transformation equipment

Abstract

The invention discloses a method and a system for storing digital twin data of power transmission and transformation equipment. The method comprises the following steps of: dividing multi-mode state monitoring data into sampling blocks which are not overlapped with each other; generating a basic recovery result and a recovery residual error from the divided sampling blocks through a variable splitting recovery network; and determining the importance of each layer of sample base by adopting a greedy algorithm so as to realize resampling and recovery of multi-modal state monitoring data under different sampling rates. According to the method, the optimization theory and the convolutional neural network are combined by means of the variable splitting technology, the trained model can quickly solve the problem of loss of different multi-modal data, and the accuracy and stability of the digital twin data of the power transmission and transformation equipment are effectively improved; an important sample base is selected through a greedy algorithm, the storage space of multi-modal state monitoring data of the power transmission and transformation equipment is effectively reduced, the effectiveness and economy of a digital twin model under a long-time dimension are enhanced, and more accurate data support is provided for equipment monitoring.

Description

Technical Field

[0001] The present invention relates to the field of digital twin technology, and in particular to a method and system for storing digital twin data of power transmission and transformation equipment. Background Art

[0002] Digital twin technology for power transmission and transformation equipment has become an effective basis for equipment status monitoring and operation and maintenance decision-making. Digital twin technology uses multimodal data such as real-time collected visual images, IMU data, laser point clouds, etc. to build a virtual model of the equipment, realize real-time monitoring of equipment status and predictive maintenance, and provide strong support for more accurate operation and maintenance decisions. However, a major problem faced by existing digital twin systems in practical engineering applications is data storage. Compared with one-dimensional signals, multimodal data collected by digital twin technology adopts two-dimensional images, three-dimensional point clouds, etc., which require larger storage space. Generally speaking, the cycle of long-term equipment status monitoring data is usually as long as several years or even decades, which is quite challenging for the storage of ordinary computer services. In addition, since visual images are highly dependent on lighting conditions in indoor or outdoor environments, data loss, such as noise and blur, is inevitable. Noise is caused by environmental factors such as heavy rain and snowfall, and also by system factors such as camera or radar damage. Due to low shutter speed or insufficient light, overexposure of the image will produce blur. Therefore, when further estimating sensitive features of equipment status, the relevant solution may be pathological and sensitive to errors caused by data loss.

[0003] Compressed sensing is a breakthrough in signal processing that can recover signals at less than half the sampling rate, which is mainly based on the Nyquist sampling theorem. The basic idea of ​​compressed sensing is to first represent the signal in a sparse transform domain and then recover the signal through an optimization process. Compressed sensing can achieve the resampling and recovery of one-dimensional state detection signals quite accurately. At the same time, the feasibility and efficiency of using compressed sensing to recover lost state monitoring signal data are also guaranteed to a certain extent. CS methods are divided into traditional methods and deep learning-based methods. Traditional CS methods recover the complete 2D image from the resampled 2D image by solving the sparse regularization optimization problem, which can be time-consuming for complex priors.

[0004] Therefore, this patent proposes a compressed sensing method based on deep learning, which aims to resample and restore multimodal state monitoring data by using compressed sensing technology. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a digital twin data storage method and system for power transmission and transformation equipment, which aims to solve the problems of model merging training calculations with different sampling rates and data loss of different types and levels by resampling and recovering multimodal state monitoring data using compressed sensing technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In the first aspect, the present invention provides a method for storing digital twin data of power transmission and transformation equipment, including: dividing multimodal state monitoring data into non-overlapping sampling blocks; generating basic recovery results and recovery residuals for the divided sampling blocks through a variable splitting recovery network; and using a greedy algorithm to determine the importance of each layer of sample basis to achieve resampling and recovery of the multimodal state monitoring data at different sampling rates.

[0009] As a preferred solution of the digital twin data storage method for power transmission and transformation equipment described in the present invention, wherein: the multimodal state monitoring data is divided into non-overlapping sampling blocks including:

[0010] In the variable split scalable convolutional neural network, a single-block sampling convolutional network is used to divide the multimodal condition monitoring data into B blocks of size 2 ×l non-overlapping sampling blocks, where l represents the number of image channels and B represents the length of the vector;

[0011] The process of rewriting sample-by-sample blocks is embedded into the deep learning based compressed sensing method by using convolutional layers with specific filter size and stride.

[0012] As a preferred solution of the digital twin data storage method for power transmission and transformation equipment described in the present invention, the variable splitting and recovery network includes:

[0013] Corrupted observation data from the multimodal condition monitoring data Recover the clean image block I R (i,t),different types of data loss problems are generated by specifying different damage matrices;

[0014] When using optimization theory, the objective function for the clean image patch is:

[0015]

[0016] in, is the fidelity term, r(·) is the regularization term, and λ is the parameter used to control the relative weight between the fidelity term and the regularization term;

[0017] In deep learning based methods, the relevant objective function is:

[0018]

[0019] Among them, Lo is the loss function and Θ is the prior parameter.

[0020] As a preferred solution of the digital twin data storage method for power transmission and transformation equipment described in the present invention, wherein: the variable splitting and recovery network also includes:

[0021] In order to decouple the regularization term and the fidelity term, an auxiliary variable w is introduced in HQS to adaptively solve different types and levels of data loss problems by integrating with HQS. The objective function of the clean image patch can be reformulated as:

[0022]

[0023] HQS attempts to minimize the following loss function:

[0024]

[0025] Among them, μ is a penalty parameter used to control The relative weight of

[0026] The loss function is solved by the following iterative solution:

[0027]

[0028] The regularization term and the fidelity term are decoupled into two separate sub-problems. The fidelity term is a quadratic regularized least squares problem whose solution is:

[0029]

[0030] Where E is the identity matrix;

[0031] The regularization term can be restated as:

[0032]

[0033] As a preferred solution of the digital twin data storage method for power transmission and transformation equipment described in the present invention, wherein: the architecture of the variable splitting and recovery network includes seven operations, the first operation consists of an expanded convolutional layer and a subsequent rectified linear unit activation layer; each of the second to sixth operations includes an expanded convolutional layer, a batch normalization layer and a ReLU activation layer; the last operation only includes an expanded convolutional layer.

[0034] As a preferred solution of the digital twin data storage method for power transmission and transformation equipment described in the present invention, the method of using a greedy algorithm to determine the importance of each layer of sample bases includes:

[0035] Given a sampling rate S;

[0036] When the sampling rate S is less than the sampling rate S of the base layer or the hth enhancement layer h , but greater than the sampling rate S of the base layer or the h-1th enhancement layer h-1 , it is necessary to determine and select the most important sample base.

[0037] As a preferred solution of the digital twin data storage method for power transmission and transformation equipment described in the present invention, it is assumed that the h-th enhancement layer uses a sample basis set G, denoted as G = {1, 2, ..., Num G}, when some sample bases and their connections to the recovery network are removed, the remaining samples should generate as accurate recovery results as possible, that is, solve the following optimization problem:

[0038]

[0039] Among them, V j is a validation set, and I (h-1) are the samples and restoration results of the h-1th enhancement layer, respectively. and are the samples and the restored results of the h-th enhancement layer after removing the samples in the subset Q of the set G;

[0040] A greedy algorithm is used to determine and select a set of important sample bases so that the solution to the optimization problem can provide the highest average peak signal-to-noise ratio value.

[0041] In a second aspect, the present invention provides a digital twin data storage system for power transmission and transformation equipment, comprising:

[0042] A sampling block division module is used to divide the multimodal condition monitoring data into non-overlapping sampling blocks;

[0043] A data recovery module, used for generating a basic recovery result and a recovery residual by using a variable splitting recovery network for the divided sampling blocks;

[0044] The resampling and recovery module is used to determine the importance of each layer of sample basis using a greedy algorithm to achieve resampling and recovery of the multimodal state monitoring data at different sampling rates.

[0045] In a third aspect, the present invention provides an electronic device, comprising:

[0046] Memory and processor;

[0047] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for storing digital twin data of power transmission and transformation equipment are implemented.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for storing digital twin data of power transmission and transformation equipment.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a method and system for storing digital twin data of power transmission and transformation equipment, which combines optimization theory with convolutional neural networks with the help of variable splitting technology. The trained model can quickly solve different multimodal data loss problems, and effectively improve the accuracy and stability of digital twin data of power transmission and transformation equipment; by selecting important sample bases through a greedy algorithm, a model can resample and restore multimodal status monitoring data at different sampling rates, effectively reducing the storage space of multimodal status monitoring data of power transmission and transformation equipment, enhancing the effectiveness and economy of the digital twin model in the long term dimension, and also providing more accurate data support for equipment monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0051] Figure 1 A schematic diagram of the overall process logic of a method for storing digital twin data of power transmission and transformation equipment according to an embodiment of the present invention;

[0052] Figure 2 A variable splitting scalable convolutional neural network structure diagram of a method for storing digital twin data of power transmission and transformation equipment according to an embodiment of the present invention;

[0053] Figure 3 This is a variable splitting and recovery network architecture diagram of a method for storing digital twin data of power transmission and transformation equipment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0055] Example 1

[0056] Reference Figure 1-Figure 3 As an embodiment of the present invention, a method for storing digital twin data of power transmission and transformation equipment is provided, such as Figure 1 The specific steps shown include:

[0057] S100: Divide the multimodal condition monitoring data into non-overlapping sampling blocks;

[0058] S200: The divided sampling blocks are restored through a variable splitting network to generate a basic restoration result and a restoration residual;

[0059] S300: A greedy algorithm is used to determine the importance of each layer of sample basis to achieve resampling and recovery of multimodal condition monitoring data at different sampling rates.

[0060] It should be noted that the present invention provides a method and system for storing digital twin data of power transmission and transformation equipment. The optimization theory is combined with the convolutional neural network by means of variable splitting technology. The trained model can quickly solve different multimodal data loss problems, and effectively improve the accuracy and stability of the digital twin data of power transmission and transformation equipment. The important sample base is selected by the greedy algorithm, so that a model can resample and restore the multimodal state monitoring data at different sampling rates, effectively reducing the storage space of the multimodal state monitoring data of the power transmission and transformation equipment, enhancing the effectiveness and economy of the digital twin model in the long-term dimension, and providing more accurate data support for equipment monitoring.

[0061] Before introducing the specific implementation plan, the compressed sensing method is introduced:

[0062] In the Nyquist sampling theorem, to recover the complete signal, the sampling rate of the sampled signal should be greater than half, and the sampled signal should contain at least half of the information of the complete signal, in order to avoid modal aliasing. Therefore, the CS method is proposed, which has the maximum sampling rate while still accurately recovering the complete signal. The mathematical theory is introduced as follows:

[0063] Assuming that the length of the complete signal x is N, the length of the sampled signal y is M, and M<<N, the relationship between the signals x and y is:

[0064] y=φx

[0065] Where Φ is a sampling matrix of size M×N, S=M / N is the sampling rate;

[0066] The CS method assumes that x can be represented by a basis Ψ of size N×N, that is, if If x has only K nonzero coefficients and K < < N, then x is K sparse in the basis Ψ. If the matrix Φ satisfies the restricted isometric property, then it can be obtained from the length Recover the K sparse signal x from the sampled signal y, that is:

[0067]

[0068] When p is 0, it is used to characterize the sparsity of the signal; however, if it is equal to 0, solving the optimization problem is complicated and is a non-deterministic polynomial problem. Therefore, it is usually converted into an easier convex problem by setting p≤1. If x is sparse enough, accurate recovery may be achieved even when S is less than 0.5.

[0069] The resampling and recovery of multimodal data signals is different from that of one-dimensional signals, which are based on a block-by-block basis. Therefore, in the acquired multimodal data, the frame I(t) at time t to be resampled and recovered should be divided into non-overlapping blocks of size B×B. The i-th block at time t is denoted by I(I,t), where i=1,...,n B , n B is the number of blocks. In the calculation, I(I,t) is squeezed into a length of B 2 The vector of I R (i, t) is squeezed into a vector M of length B Therefore, the sampling matrix is ​​a diagonal matrix:

[0070]

[0071] Among them, Φ B is a block I(I,t) of size M B ×B 2 The sampling matrix, the relevant sampling rate is S = M B / B 2 . Therefore, the relationship between signals x and y can be rewritten in a block-by-block manner:

[0072] I R (i,t)=Φ B I(i,t)

[0073] Therefore, for the resampling and restoration of multimodal information, the relevant optimization problem is:

[0074]

[0075] There are two methods to solve the above optimization problem, traditional mathematical methods and methods based on deep learning. Traditional mathematical methods recover multimodal information by solving sparse regularization optimization problems. Although traditional methods can achieve resampling and recovery at different sampling rates, their iterative nature leads to high computational burden. Compared with traditional methods, deep learning methods based on convolutional neural networks (CNNs) can achieve acceptable accuracy with lower computational burden.

[0076] In an embodiment of the present application, a deep learning CS method of variable split scalable convolutional neural network is proposed to address the problem that models with different sampling rates need to merge training calculations and different types and levels of data loss for CS methods based on deep learning. This method takes advantage of optimization theory and is integrated with the CS method based on deep learning as a modular part. The variable split scalable convolutional neural network is designed as a single-block sampling convolutional network plus a hierarchical recovery network. By using a single-block sampling convolutional network, multimodal condition monitoring data is divided into non-overlapping sampling blocks. The hierarchical recovery network consists of a base layer (BL) and multiple enhancement layers (ELs). Figure 2 As shown in the figure, the sampling blocks generated by the block sampling convolutional network are regarded as multi-band (MB) feature maps and are divided into multiple groups marked with different colors. The base layer (BL) uses a set of sampling blocks to generate basic recovery results. Each enhancement layer (EL) uses a set of sampling blocks to generate recovery residuals and refers to the lower layer results to improve the quality of basic recovery. The variable split recovery network is specially designed to solve different types and degrees of data loss problems. All three recovery networks are supervised by the mean square error (MSE) loss function. Finally, a greedy algorithm is used to determine the importance of the sample basis of each layer, and a model is used to select important sample bases for resampling and recovery at different sampling rates.

[0077] In the embodiment of the present application, the above step S100 divides the multimodal state monitoring data into non-overlapping sampling blocks including:

[0078] In the variable split scalable convolutional neural network, a single-block sampling convolutional network is used to divide the multimodal condition monitoring data into B blocks of size 2 ×l non-overlapping sampling blocks, where l represents the number of image channels and l is set to 1, and B represents the length of the vector;

[0079] The process of rewriting each sample block by sample block is embedded into the deep learning based compressed sensing method by using a convolutional layer with a specific filter size and stride, which is expressed as:

[0080] I R (i,t)=F s *I(i,t)

[0081] Among them, F s It is M B filter, which will be of size B 2 Each block I(i,t) of ×l is resampled to a length of M B Block I R (i, t), this block sampling convolutional network is represented as Conv(B 2 ,l,M B). In the following description, the convolutional layer is labeled Conv(z,in,out), where z is the spatial size of the filter, in is the number of input channels, and out is the number of output channels. In order to achieve non-overlapping resampling at a fixed sampling rate, the stride of this convolutional layer is B×B. The convolutional layer can optimize the sampling convolutional network and the subsequent recovery network to learn the sampling process.

[0082] It should be noted that the above step S100 ensures that the data in each sampling block has a high degree of correlation and consistency, thereby reducing redundancy and interference between data and improving the efficiency and accuracy of data processing. By organizing the data segmentation, it is not only convenient for parallel processing and accelerated calculation, but also provides structured input for subsequent steps such as the generation of variable splitting recovery network, so that the system can more accurately capture the intrinsic connection and change pattern between different modal data. In addition, non-overlapping sampling blocks help to maintain the time series characteristics of the data, ensuring accurate state monitoring and fault detection at different sampling rates and time scales, and enhancing the stability and reliability of the entire system.

[0083] In the embodiment of the present application, the variable splitting and restoring network in the above step S200 includes:

[0084] Corrupted Observations from Multimodal Condition Monitoring Data Recover the clean image block I R (i,t):

[0085]

[0086] Where H is a damage matrix, v is additive Gaussian noise with standard deviation σ; different types of data loss problems are generated by specifying different damage matrices;

[0087] When using optimization theory, the objective function for the clean image patch is:

[0088]

[0089] in, is the fidelity term, r(·) is the regularization term, and λ is a parameter used to control the relative weight between the fidelity term and the regularization term; the fidelity term ensures that the solution is consistent with the relevant corruption, while the regularization term limits the properties required for recovery. When applying optimization theory, the solution of the objective function of the clean image patch is usually based on time-consuming iterative estimation.

[0090] In deep learning based methods, the relevant objective function is:

[0091]

[0092] Among them, Lo is the loss function and Θ is the prior parameter;

[0093] The above-related objective function can be restated as:

[0094]

[0095] The solution to the above formula is achieved by optimizing the loss function, using multimodal data containing both corrupted and clean data for training; however, the optimization method has complex priors and is therefore not easily trained by CNN. With the help of variable splitting techniques such as the alternating direction method of multiple clamps (ADMM) and the semi-quadratic segmentation method (HQS), the regularization and fidelity terms in the optimization method can be decoupled, which enables the integration of optimization theory with CNN.

[0096] In order to decouple the regularization term and the fidelity term, an auxiliary variable w is introduced in HQS to adaptively solve different types and levels of data loss problems by integrating with HQS. The objective function of the clean image patch can be reformulated as:

[0097]

[0098] HQS attempts to minimize the following loss function:

[0099]

[0100] Among them, μ is a penalty parameter used to control The relative weight of

[0101] The loss function is solved by the following iterative solution:

[0102]

[0103] Where k is the iteration index;

[0104] The regularization term and the fidelity term are decoupled into two separate sub-problems. The fidelity term is a quadratic regularized least squares problem whose solution is:

[0105]

[0106] Where E is the identity matrix;

[0107] The regularization term can be restated as:

[0108]

[0109] In order for CNN to learn the above regularization term, it can be restated as:

[0110]

[0111] Therefore, HQS provides a feasible solution to integrate optimization theory with CNN, that is, integrating the learned regularization term formula model with the fidelity term formula.

[0112] Further, such as Figure 3 The architecture of the variable splitting and restoration network shown in the figure consists of seven operations. The first operation consists of a dilated convolutional layer followed by a rectified linear unit activation layer. Each of the second to sixth operations contains a dilated convolutional layer, a batch normalization layer, and a ReLU activation layer. The last operation only contains a dilated convolutional layer. The dilation factors of the 3×3 dilated convolutions from the first operation to the last operation are set to 1, 2, 3, 4, 3, 2, and 1, respectively. The equivalent receptive fields corresponding to each operation are 3, 5, 7, 9, 7, 5, and 3, respectively, and the receptive field of the entire network is 33×33.

[0113] It should be noted that the above step S200 can not only effectively separate the main features and subtle differences in the data, ensuring high-fidelity recovery of the original signal, but also provide valuable basis and supplementary information for subsequent analysis. Through the variable splitting recovery network, the system can more accurately capture complex and changeable data patterns and enhance the ability to identify potential faults or abnormal conditions. In addition, the generated basic recovery results and recovery residuals provide rich materials for data analysis and model training at different levels, which helps to further optimize the data processing process, improve the overall system response speed and decision-making accuracy, and thus play a key role in the real-time monitoring of the power system.

[0114] In the embodiment of the present application, the above step S300 uses a greedy algorithm to determine the importance of each layer of sample basis to achieve resampling and recovery of multimodal state monitoring data at different sampling rates, including:

[0115] Given a sampling rate S;

[0116] When the sampling rate S is less than the sampling rate S of the base layer or the hth enhancement layer h , but greater than the sampling rate S of the base layer or the h-1th enhancement layer h-1 , the most important sample base needs to be determined and selected. Since the base layer (BL) can be regarded as a specific enhancement layer (EL), the base layer (BL) and the enhancement layer (EL) are uniformly labeled as enhancement layer (EL);

[0117] Assume that the hth enhancement layer uses a sample basis set G, denoted as G = {1, 2, ..., Num G}, when some sample bases and their connections to the recovery network are removed, the remaining samples should generate as accurate recovery results as possible, that is, solve the following optimization problem:

[0118]

[0119] Among them, V j is a validation set, and I (h-1) are the samples and restoration results of the h-1th enhancement layer, respectively. and are the samples and the restored results of the h-th enhancement layer after removing the samples in the subset Q of the set G;

[0120] A greedy algorithm is used to determine and select a set of important sample bases so that the solution to the optimization problem provides the highest average peak signal-to-noise ratio value.

[0121] It should be noted that the steps of the greedy algorithm are shown in Table 1.

[0122] Table 1: Greedy algorithm for sample base selection.

[0123]

[0124]

[0125] It should be noted that the above step S300 not only ensures the efficiency and accuracy of data recovery, but also optimizes resource utilization, so that the system can flexibly respond to various sampling rate requirements while maintaining data integrity and details. The adaptive ability of the system is enhanced by dynamically selecting the most important sample base, and stable performance can be maintained in a complex and changing monitoring environment. Step S300 effectively improves the overall efficiency and reliability of multimodal condition monitoring data processing, provides strong support for real-time monitoring and fault warning of power systems, and ensures the stability and safety of system operation.

[0126] The above is a schematic scheme of a method for storing digital twin data of power transmission and transformation equipment in this embodiment. It should be noted that the technical scheme of the system for storing digital twin data of power transmission and transformation equipment and the technical scheme of the method for storing digital twin data of power transmission and transformation equipment described above belong to the same concept. For details not described in detail in the technical scheme of the system for storing digital twin data of power transmission and transformation equipment in this embodiment, please refer to the description of the technical scheme of the method for storing digital twin data of power transmission and transformation equipment described above.

[0127] This embodiment also provides a digital twin data storage system for power transmission and transformation equipment, including:

[0128] A sampling block division module is used to divide the multimodal condition monitoring data into non-overlapping sampling blocks;

[0129] A data recovery module is used to generate basic recovery results and recovery residuals from the divided sampling blocks through a variable splitting recovery network;

[0130] The resampling and recovery module is used to determine the importance of each layer of sample basis using a greedy algorithm to achieve resampling and recovery of multimodal condition monitoring data at different sampling rates.

[0131] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0132] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. 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 and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless method can be implemented through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a digital twin data storage method for power transmission and transformation equipment is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0133] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0134] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0135] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0137] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0141] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0142] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for storing digital twin data of power transmission and transformation equipment, characterized in that: include: Divide the multimodal condition monitoring data into non-overlapping sampling blocks; The divided sampling blocks are restored through a variable splitting network to generate a basic restoration result and a restoration residual; A greedy algorithm is used to determine the importance of each layer of sample basis to achieve resampling and recovery of the multimodal state monitoring data at different sampling rates.

2. The method for storing digital twin data of power transmission and transformation equipment according to claim 1, characterized in that: The step of dividing the multimodal condition monitoring data into non-overlapping sampling blocks includes: In the variable split scalable convolutional neural network, a single-block sampling convolutional network is used to divide the multimodal condition monitoring data into B blocks of size 2 ×l non-overlapping sampling blocks, where l represents the number of image channels and B represents the length of the vector; The process of rewriting sample-by-sample blocks is embedded into the deep learning based compressed sensing method by using convolutional layers with specific filter size and stride.

3. The method for storing digital twin data of power transmission and transformation equipment according to claim 2, characterized in that: The variable splitting and recovery network comprises: Corrupted observation data from the multimodal condition monitoring data Recover the clean image block I R (i,t),different types of data loss problems are generated by specifying different damage matrices; When using optimization theory, the objective function for the clean image patch is: in, is the fidelity term, r(·) is the regularization term, and λ is the parameter used to control the relative weight between the fidelity term and the regularization term; In deep learning based methods, the relevant objective function is: Among them, Lo is the loss function and Θ is the prior parameter.

4. The method for storing digital twin data of power transmission and transformation equipment according to claim 3, characterized in that: The variable splitting and recovery network also includes: In order to decouple the regularization term and the fidelity term, an auxiliary variable w is introduced in HQS to adaptively solve different types and levels of data loss problems by integrating with HQS. The objective function of the clean image patch can be reformulated as: HQS attempts to minimize the following loss function: Among them, μ is a penalty parameter used to control The relative weight of The loss function is solved by the following iterative solution: The regularization term and the fidelity term are decoupled into two separate sub-problems. The fidelity term is a quadratic regularized least squares problem whose solution is: Where E is the identity matrix; The regularization term can be restated as:

5. The method for storing digital twin data of power transmission and transformation equipment according to claim 4, characterized in that: The architecture of the variable splitting and recovery network includes seven operations, the first operation consists of a dilated convolutional layer and a subsequent rectified linear unit activation layer; each of the second to sixth operations includes a dilated convolutional layer, a batch normalization layer, and a ReLU activation layer; and the last operation only includes a dilated convolutional layer.

6. The method for storing digital twin data of power transmission and transformation equipment according to claim 5, characterized in that: The greedy algorithm used to determine the importance of sample bases at each layer includes: Given a sampling rate S; When the sampling rate S is less than the sampling rate S of the base layer or the hth enhancement layer h , but greater than the sampling rate S of the base layer or the h-1th enhancement layer h-1 , it is necessary to determine and select the most important sample base.

7. The method for storing digital twin data of power transmission and transformation equipment according to claim 6, characterized in that: Assume that the hth enhancement layer uses a sample basis set G, denoted as G = {1, 2, ..., Num G }, when some sample bases and their connections to the recovery network are removed, the remaining samples should generate as accurate recovery results as possible, that is, solve the following optimization problem: Among them, V j is a validation set, and I (h-1) are the samples and restoration results of the h-1th enhancement layer, respectively. and are the samples and the restored results of the h-th enhancement layer after removing the samples in the subset Q of the set G; A greedy algorithm is used to determine and select a set of important sample bases so that the solution to the optimization problem can provide the highest average peak signal-to-noise ratio value.

8. A system using the digital twin data storage method for power transmission and transformation equipment according to any one of claims 1 to 7, characterized in that: include: A sampling block division module is used to divide the multimodal condition monitoring data into non-overlapping sampling blocks; A data recovery module, used for generating a basic recovery result and a recovery residual by using a variable splitting recovery network for the divided sampling blocks; The resampling and recovery module is used to determine the importance of each layer of sample basis using a greedy algorithm to achieve resampling and recovery of the multimodal state monitoring data at different sampling rates.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for storing digital twin data of power transmission and transformation equipment according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for storing digital twin data of power transmission and transformation equipment as described in any one of claims 1 to 7.

Images