Current transformer error state evaluation method and system based on multi-modal characteristics

Abstract

The invention belongs to the technical field of power monitoring, and discloses a current transformer error state evaluation method and system based on multi-modal characteristics, and the method comprises the steps: constructing a training data set; converting the training data into image data in multiple modes, performing global and local feature extraction on the image data in the multiple modes, and splicing the global and local features to obtain spliced features; splicing the global features and the local features in various modes to obtain full-mode global features and local features, performing calculation in combination with original image data to obtain an attention feature graph, and operating the attention feature graph and the spliced features to obtain final fusion features; and performing error state evaluation by using the final fusion feature, and outputting an error state condition. According to the method, on the basis of current time sequence data, feature information of three different modes is efficiently fused and utilized, and the reliability and accuracy of an evaluation result are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power monitoring, and in particular to a method and system for evaluating the error state of a current transformer based on multi-modal characteristics. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the power system, the current transformer is a key device in the electric energy metering system. The accuracy of its metering performance is directly related to the fairness and reliability of electric energy trading. The current transformer (CT), a precision device based on the principle of electromagnetic induction, can convert a high-amplitude primary current into an easily measurable secondary current. Its main structure includes a closed iron core and a coil wound around it. However, in the actual operating environment, the metering error of the CT may gradually deviate from the standard range over time due to the limitations of the acquisition mechanism and the influence of environmental conditions. This requires us not only to have accurate and rapid fault diagnosis capabilities, but also to be able to grasp the real-time status of the CT metering error at all times, so as to guide the operation and maintenance personnel to reasonably plan maintenance strategies. If the abnormality of the transformer status is not detected in time, it will pose a potential threat to the overall operation of the power grid.

[0004] Therefore, ensuring the real-time accuracy of current transformers, reducing power metering losses, and ensuring the stable operation of measurement, control and protection equipment have become urgent technical challenges. How to achieve real-time online evaluation of CT error status is undoubtedly a key problem that the power industry needs to overcome. The existing real-time online evaluation method of CT error status only utilizes the single time series data characteristics of current transformer data, and does not combine multi-modal data feature information. The evaluation results are not accurate and reliable enough. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a current transformer error state evaluation method and system based on multi-modal characteristics. On the basis of current time series data, the characteristic information of three different modes is efficiently integrated and utilized, thereby improving the reliability and accuracy of the evaluation results.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for evaluating a current transformer error state based on multimodal features, comprising the following steps: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

[0007] As an optional implementation, the multimodal global and local fusion network includes a data modality conversion layer, a global and local feature extraction layer, a multimodal global and local feature fusion layer, and a fully connected layer.

[0008] As an optional implementation, the training data is input into a multimodal global and local fusion network and converted into image data under multiple modalities, specifically: The training data are converted into two-dimensional image data under three modes through three algorithms: GAF, MTF and RPM. GAF converts the training data into a matrix representation in the angular domain. MTF converts the training data into a probability matrix through a Markov process. RPM generates a reproduction graph by recording the distance relationship between the sequence points of the training data.

[0009] As an optional implementation, global features are extracted by global average pooling, and then channel-adjusted by point-by-point convolution; Local features are extracted through point-by-point convolution, and the residual structure is used to connect the original image data of the same size to maintain the original context information.

[0010] As an optional implementation method, the attention feature map is converted into a feature weight matrix through a sigmoid function and then Hadamard product is performed with the concatenated features corresponding to the three modalities to adjust the importance of the feature points in each concatenated feature.

[0011] As an optional implementation, the final fusion feature is used to evaluate the error state, specifically: The final fusion features are sent to the fully connected layer, integrated into the corresponding probability distribution, and classified using the SVM classifier to output the error state of the current current transformer.

[0012] In a second aspect, the present invention provides a current transformer error state evaluation system based on multi-modal features, comprising: The training set building module is configured to: collect current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; The feature extraction module is configured to: construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The feature fusion module is configured to: respectively splice the global features and local features under multiple modalities step by step to obtain full-modal global features and full-modal local features, perform attention calculation based on the full-modal global features, the full-modal local features and the original image data to obtain an attention feature map, convert the attention feature map into a weight matrix and then calculate it with the splicing feature, adjust the importance of feature points in the splicing feature, and fuse the adjusted splicing features under multiple modalities to obtain the final fusion feature; The error state evaluation module is configured to: use the final fusion feature to perform error state evaluation and output the error state of the current transformer.

[0013] In a third aspect, the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the following steps are performed: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the following steps are performed: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

[0015] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the following steps are implemented: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a current transformer error state evaluation method and system based on multi-modal features, designs a multi-modal global and local feature fusion method, converts one-dimensional current time series data into two-dimensional image features under three modes, extracts and fuses the image features under different modes respectively, obtains comprehensive features that fuse information of different modes, and uses the comprehensive features to evaluate the error of the current transformer. Based on the current time series data, the present invention efficiently fuses and utilizes the feature information of three different modes, comprehensively analyzes and utilizes the inherent structure and dynamic characteristics of the current time series data, thereby improving the reliability and accuracy of the error evaluation results.

[0017] The present disclosure proposes a method and system for evaluating the error state of a current transformer based on multimodal features, and proposes a multimodal global and local feature fusion network structure, which is mainly composed of a data mode conversion layer, a global and local feature extraction layer, a multimodal global and local feature fusion layer, and a fully connected layer. The data mode conversion layer converts the one-dimensional current time series data into two-dimensional image data under three modes; the global and local feature extraction layer extracts global features and local features from the multimodal two-dimensional image; the multimodal global and local feature fusion layer performs efficient fusion of the splicing features of features under different modes. Finally, the fused comprehensive features are sent to the fully connected layer to integrate the comprehensive features into the corresponding probability distribution, and the SVM classifier is used for classification, and the error state of the current transformer is output, so as to realize the real-time online evaluation of the error state of the current transformer.

[0018] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0020] Figure 1 A flowchart of a current transformer error state evaluation method based on multi-modal features provided in Example 1 of the present invention; Figure 2 The figure is a confusion matrix of the result of using the method of the present invention to identify the error state of the current transformer. DETAILED DESCRIPTION

[0021] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0023] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0025] Terminology explanation: Gramian Angular Field (GAF): is an effective method to convert one-dimensional time series data into two-dimensional image representation. Its core principle is to calculate the cosine values ​​of the angles between data points in the time series and map these values ​​to the pixels of the two-dimensional image, thereby generating an image that can reflect the dynamic and periodic characteristics of the time series.

[0026] Markov Transition Field (MTF): is a method for converting time series data into a two-dimensional image, mainly used in the fields of computer vision and image processing, especially in video sequence analysis. MTF is based on the concept of Markov chain. It converts time series data into a two-dimensional image by modeling the probability transition between states in the time series, thereby providing local information of the signal in time and frequency. ‌The basic principle is that in MTF, the video sequence is divided into a grid-like cell, and the transitions between the pixel intensity values ​​in each cell are modeled as a Markov chain. By calculating and storing the transition probabilities between each pixel intensity value, these probabilities are stored in a transition matrix, which can be visualized through a heat map, showing the transition probability from one intensity value to another. This heat map is called a Markov transition field and provides a compact representation of spatiotemporal patterns in video sequences. ‌

[0027] Relative Position Matrix (RPM): is a matrix representation method used to describe the relative position relationship of objects. It converts the relative position relationship of objects into data through mathematical models, providing a basis for subsequent analysis and application. The relative position matrix is ​​established based on the coordinate system of a two-dimensional plane or three-dimensional space. In a certain coordinate system, the position of each object can be expressed by coordinates, and the relative position is obtained by calculating the coordinate difference between different objects. The relative position matrix presents these differences in the form of a matrix. For example, in a two-dimensional plane, if there are n objects, the position of each object can be expressed as a two-dimensional coordinate (x, y), then the size of the relative position matrix is ​​n×n, where the element in the i-th row and j-th column represents the position difference of the i-th object relative to the j-th object.

[0028] Example 1 like Figure 1 As shown, this embodiment provides a current transformer error state evaluation method based on multi-modal features, comprising the following steps: S1, collecting the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; S2. construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; S3, respectively, splicing the global features and local features under multiple modalities step by step to obtain full-modal global features and full-modal local features, performing attention calculation based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map, converting the attention feature map into a weight matrix and then calculating it with the splicing features, adjusting the importance of feature points in the splicing features, and fusing the adjusted splicing features under multiple modalities to obtain the final fusion feature; S4. Use the final fusion features to evaluate the error state and output the error state of the current transformer.

[0029] The general idea of ​​the present invention is: first, collect the secondary current data of multiple current transformers during historical operation, including normal current transformers and abnormal current transformers, and use the historical current data samples (one-dimensional current time series data) provided by normal and abnormal current transformers to build a training data set. Then, construct a multimodal global and local fusion network, which is mainly composed of a data mode conversion layer, a global and local feature extraction layer, a multimodal global and local feature fusion layer, and a fully connected layer. In the data mode conversion layer, the one-dimensional current time series data will be converted into two-dimensional image data of three modes through three algorithms: Gramian Angular Field (GAF), Markov Transition Field (MTF), and Relative Position Matrix (RPM); in the global and local feature extraction layer, the two-dimensional image data under the three modes will be extracted globally and locally; the three sets of global and local multimodal features extracted will be globally and locally efficiently fused with different modal features in the multimodal global and local feature fusion layer. Finally, the fused features are sent to the fully connected layer to accurately evaluate the error state of the current transformer. In practical applications, the network is first trained with a large number of data samples, and then the trained network is connected to the secondary current data of the current transformer to be tested, so that the error state of the current transformer can be evaluated online in real time.

[0030] (1) Data modality conversion layer After the one-dimensional current time series data is input into the multimodal global and local fusion network, it will first be converted into a data mode through the data mode conversion layer. The data mode conversion layer mainly converts the one-dimensional data into two-dimensional image data under three modes through three algorithms: GAF, MTF and RPM. GAF mainly converts the time series data into a matrix representation in the angular domain to capture the periodicity and phase information of the data; MTF converts the sequence into a probability matrix through the Markov process, reflecting the state transition characteristics of the sequence; RPM generates a recurrence diagram by recording the distance relationship between sequence points, revealing the complexity and self-similarity of the sequence. The process can be expressed as:

[0031]

[0032]

[0033] In the formula, X is the one-dimensional current time series data of the input network, , and They are the two-dimensional image data converted by GAF, MTF and RPM algorithms respectively.

[0034] (2) Global and local feature extraction layer In the global and local feature extraction layers, three sets of two-dimensional image data after modality conversion are , and The global features and local features are extracted and spliced ​​for output. In the global feature extraction part, the global features are first extracted by global average pooling, and then the channels are adjusted by point-by-point convolution. Taking global feature extraction as an example, the calculation process can be expressed as follows:

[0035] In the formula, Indicates from The global features extracted from It is the two-dimensional image feature obtained after modal transformation by GAF algorithm. Represents a global pooling operation, Represents a point-by-point convolution operation. and After global feature extraction, the corresponding global features can also be obtained and .

[0036] In the local feature extraction part, local features are extracted mainly through point-by-point convolution, and the residual structure is used to connect the original features of the same size to maintain the original context information. Taking local feature extraction as an example, the calculation process can be expressed as follows:

[0037] In the formula, It is the two-dimensional image feature obtained after modal transformation by GAF algorithm. Representative from The local features extracted from Represents a point-by-point convolution operation. and After local feature extraction, the corresponding global features can also be obtained. and . Then, the obtained global features and the corresponding local features are spliced ​​to obtain three sets of spliced ​​multimodal global and local features (i.e., spliced ​​features). The process can be expressed as:

[0038]

[0039]

[0040] In the formula, for The concatenated features extracted after the global and local feature extraction layers, for The concatenated features extracted after the global and local feature extraction layers, for The concatenated features extracted after the global and local feature extraction layers, Concat(;) It is a feature concatenation operation.

[0041] (3) Multimodal global and local feature fusion layer In the multimodal global and local feature fusion layer, the global features of the three modalities are concatenated step by step to obtain a global feature containing all modal information. (i.e., full-modal global features). The local features of the three modalities are concatenated step by step to obtain local features containing full-modal information. (i.e., full-modal local features). Subsequently, the initial information (original image data) is combined with and The attention mechanism is calculated and the initial information (i.e. original image data) under three modes and the attention feature map generated by the corresponding global and local information are output. sigmoid The function is converted into a feature weight matrix in the range (0,1), and then the Hadamard product is performed with the global and local features (i.e., splicing features) corresponding to the three modes to adjust the importance of the feature points in each global and local feature (i.e., splicing feature). Finally, the adjusted global and local features (i.e., splicing features) under the three modes are resized and superimposed and fused to output the final fusion feature. The process can be expressed by the following formula:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047] In the above formula, , and They are the global features under three different modes, It is a global feature that contains all modal information after concatenating the global features under three modes. , and They are the local features under three different modes, It is a local feature that contains full modal information after concatenating the local features under three modes. , , They are the initial information of the three modes (i.e. original image data) , and The obtained attention feature map is sigmoid The three feature weight matrices converted by the function. 、 、 They are all dimension mapping functions, which are used to convert the dimensions of the multimodal features and original information involved in the calculation into a unified dimension. d Perform calculations. r It is a scale reshaping function, which is used to reshape the feature scales under different modes to be consistent so as to facilitate superposition and fusion. is the final output feature of the multimodal global and local fusion network. ⊙ represents the Hadamard product operation. Softmax is the normalized exponential function.

[0048] (4) Fully connected layer The fused features are then sent to the fully connected layer to integrate the comprehensive features into the corresponding probability distribution, and the SVM classifier is used for classification to output the error state of the current transformer.

[0049] The multimodal global and local fusion networks are trained offline through the collected current transformer data. After the fully trained network is obtained, the secondary side current signal of the current transformer to be tested is connected to perform real-time error state evaluation to obtain the error state of the current transformer to be tested.

[0050] The following is a specific implementation effect of the method of the present invention. The experiment was conducted through the collected historical current data samples. For 300 groups of experimental samples, including 150 normal transformer samples and 150 abnormal transformer samples, the error state recognition was performed using the method of the present invention. The confusion matrix of the recognition result is as follows: Figure 2 As shown: The confusion matrix shows that among the 150 normal samples in the test samples, the algorithm evaluates 149 normal samples correctly, and among the 150 abnormal samples in the test samples, the algorithm evaluates 150 abnormal samples correctly. Based on this result, it can be calculated that the accuracy of the algorithm reaches 99.67%, indicating that the method of the present invention can accurately identify the error state of the current transformer.

[0051] The present invention designs a multi-modal global and local feature fusion method, converts one-dimensional current time series data into two-dimensional image features under three modes, extracts and fuses the image features under different modes respectively, obtains comprehensive features that fuse information of different modes, and uses the comprehensive features to perform error evaluation of current transformers. Based on the current time series data, the present invention efficiently fuses and utilizes the feature information of three different modes, comprehensively analyzes and utilizes the inherent structure and dynamic characteristics of the current time series data, thereby improving the reliability and accuracy of the error evaluation results.

[0052] The present invention proposes a multimodal global and local feature fusion network structure, which is mainly composed of a data mode conversion layer, a global and local feature extraction layer, a multimodal global and local feature fusion layer, and a fully connected layer. The data mode conversion layer converts the one-dimensional current time series data into two-dimensional image data under three modes; the global and local feature extraction layer extracts global features and local features from the multimodal two-dimensional image; the multimodal global and local feature fusion layer performs efficient fusion of the splicing features of features under different modes. Finally, the fused comprehensive features are sent to the fully connected layer to integrate the comprehensive features into the corresponding probability distribution, and the SVM classifier is used for classification, and the error state of the current transformer is output to realize the real-time online evaluation of the error state of the current transformer.

[0053] Example 2 This embodiment provides a current transformer error state evaluation system based on multi-modal features, including: The training set building module is configured to: collect current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; The feature extraction module is configured to: construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The feature fusion module is configured to: respectively splice the global features and local features under multiple modalities step by step to obtain full-modal global features and full-modal local features, perform attention calculation based on the full-modal global features, the full-modal local features and the original image data to obtain an attention feature map, convert the attention feature map into a weight matrix and then calculate it with the splicing feature, adjust the importance of feature points in the splicing feature, and fuse the adjusted splicing features under multiple modalities to obtain the final fusion feature; The error state evaluation module is configured to: use the final fusion feature to perform error state evaluation and output the error state of the current transformer.

[0054] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0055] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

[0056] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0057] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0058] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the following method is performed: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

[0059] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0060] A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the following method is implemented: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

[0061] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0062] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0063] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, and the like. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, and the like.

[0064] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0065] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A current transformer error state evaluation method based on multi-modal characteristics, characterized in that: The following steps are involved: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

2. The current transformer error state evaluation method based on multi-modal features according to claim 1, characterized in that: The multimodal global and local fusion network includes a data modality conversion layer, a global and local feature extraction layer, a multimodal global and local feature fusion layer, and a fully connected layer.

3. The current transformer error state evaluation method based on multi-modal features according to claim 1, characterized in that: The training data is input into the multimodal global and local fusion network and converted into image data under multiple modalities, specifically: The training data are converted into two-dimensional image data under three modes through three algorithms: GAF, MTF and RPM. GAF converts the training data into a matrix representation in the angular domain. MTF converts the training data into a probability matrix through a Markov process. RPM generates a reproduction graph by recording the distance relationship between the sequence points of the training data.

4. The current transformer error state evaluation method based on multi-modal features according to claim 1, characterized in that: The global features are extracted by global average pooling, and then the channels are adjusted by point-by-point convolution. Local features are extracted through point-by-point convolution, and the residual structure is used to connect the original image data of the same size to maintain the original context information.

5. The current transformer error state evaluation method based on multi-modal features according to claim 1, characterized in that: The attention feature map is converted into a feature weight matrix through the sigmoid function and then Hadamard product is performed with the concatenated features corresponding to the three modalities to adjust the importance of the feature points in each concatenated feature.

6. The current transformer error state evaluation method based on multi-modal features according to claim 1, characterized in that: The final fusion features are used to evaluate the error state, specifically: The final fusion features are sent to the fully connected layer, integrated into the corresponding probability distribution, and classified using the SVM classifier to output the error state of the current current transformer.

7. A current transformer error state evaluation system based on multi-modal characteristics, characterized in that: include: The training set building module is configured to: collect current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; The feature extraction module is configured to: construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The feature fusion module is configured to: respectively splice the global features and local features under multiple modalities step by step to obtain full-modal global features and full-modal local features, perform attention calculation based on the full-modal global features, the full-modal local features and the original image data to obtain an attention feature map, convert the attention feature map into a weight matrix and then calculate it with the splicing feature, adjust the importance of feature points in the splicing feature, and fuse the adjusted splicing features under multiple modalities to obtain the final fusion feature; The error state evaluation module is configured to: use the final fusion feature to perform error state evaluation and output the error state of the current transformer.

8. An electronic device, characterized in that: The invention comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein the computer instructions are executed by the processor to complete the following steps: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, the following steps are implemented: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the following steps: Collect the current data of normal current transformers and abnormal current transformers during historical operation to build a training data set; Construct a multimodal global and local fusion network, input the training data into the multimodal global and local fusion network, convert it into image data under multiple modalities, extract global and local features from the image data under multiple modalities respectively, and splice the global and local features to obtain spliced ​​features; The global features and local features under multiple modalities are spliced ​​step by step to obtain full-modal global features and full-modal local features. Attention calculation is performed based on the full-modal global features, full-modal local features and original image data to obtain an attention feature map. The attention feature map is converted into a weight matrix and then calculated with the splicing features. The importance of the feature points in the splicing features is adjusted, and the splicing features under multiple modalities after adjustment are fused to obtain the final fused features. The final fusion features are used to evaluate the error state and output the error state of the current transformer.

Images