Voltage sag state estimation method based on space-time three-dimensional tensor

Abstract

The invention belongs to the technical field of voltage sag state estimation, and particularly discloses a voltage sag state estimation method based on a space-time three-dimensional tensor, which comprises the following steps of: acquiring three-phase voltage time sequence data of a power grid, converting the three-phase voltage time sequence data into three-phase GAF two-dimensional image data through Gramb angle field transformation, and acquiring three-phase GAF two-dimensional image data; through time series data similarity calculation, obtaining a coupling relationship between monitoring nodes, and stacking image data into a space-time three-dimensional tensor to reflect spatial characteristics of a system; a CNN + Transform series model is constructed, and the CNN + Transform series model is constructed; and inputting the space-time three-dimensional tensor into a CNN + Transform series model, and outputting a voltage sag state estimation result. According to the method, the problems of complex voltage sag state estimation process and poor accuracy caused by the fact that a traditional method needs to obtain power grid element parameters and considers a power grid topological structure, electrical parameters and operation conditions are solved, and the accuracy of voltage sag state estimation can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of voltage sag state estimation, and in particular relates to a voltage sag state estimation method based on time-space three-dimensional tensor. Background Art

[0002] Among power quality issues, voltage sags occur most frequently and cause significant economic losses. Resolving voltage sags requires a system that can capture voltage sag conditions at every node in the power grid. This information can be used to identify weak links, assess economic costs, verify responsible parties, and develop mitigation plans. However, in reality, the power grid is a vast system with numerous nodes and a complex topology. Cost-effectiveness makes it impossible to install voltage sag measurement devices at every node, so only a limited number of monitoring devices can be installed. Traditional voltage sag state estimation methods rely on analyzing the power grid topology and electrical parameters and constructing mathematical models. These typically require the parameters of all power grid components, which are extremely difficult to obtain. Furthermore, the measured data in the power grid contains disturbances such as harmonics and noise. The complex system operation and changes in electrical parameters and even topology during faults result in inaccurate voltage sag state estimation using these methods. Data-driven voltage sag state estimation, on the other hand, does not require obtaining grid component parameters, nor does it analyze the system's operating conditions, topology, and electrical parameters. It only considers the nonlinear coupling relationship between voltage sag data between monitored and unmonitored nodes. However, the input data selected by current voltage sag data-driven state estimation methods is generally one-dimensional voltage time series data or voltage data at a certain time node, without deeply exploring the spatiotemporal mapping patterns of voltage sag data in the power grid. Traditional deep learning models can generally only extract local features and have difficulty establishing long-distance dependencies on overall data. The models are easily affected by system noise, and large amounts of noisy data can affect the model's training and prediction results. During the model training process, the data feature information generated by the algorithm is lost. All of these factors will reduce the accuracy of the data-driven state estimation method, and the computational complexity of the model will directly affect the timeliness of the state estimation. Summary of the Invention

[0003] The purpose of the present invention is to solve the problem that the traditional method needs to obtain the parameters of the power grid components and consider the power grid topology, electrical parameters and operating conditions, resulting in a complex voltage sag state estimation process and poor accuracy. A voltage sag state estimation method based on a three-dimensional space-time tensor is proposed.

[0004] The technical solution of the present invention is: a voltage sag state estimation method based on time-space three-dimensional tensor, comprising the following steps: S1. Collect actual detection data of the power grid and obtain the three-phase voltage time series data of the power grid; S2. Perform Gram angle field transformation and similarity calculation on the three-phase voltage time series data to obtain a three-dimensional space-time tensor; S3. Build a CNN+Transformer tandem model; S4. Input the spatiotemporal three-dimensional tensor into the CNN+Transformer series model and output the voltage sag state estimation result.

[0005] The beneficial effects of the present invention are: 1. The present invention does not require component parameters to model the system, and does not require analysis of the system's operating conditions, topology, and electrical parameters.

[0006] 2. The present invention constructs a three-dimensional tensor based on the voltage time series data of multiple monitoring points in the power grid, explores the spatiotemporal mapping rules of the propagation of sag events in the power grid, and can effectively improve the accuracy of voltage sag state estimation.

[0007] Preferably, the step S2 specifically includes the following sub-steps: S21. Performing Gram angle field transform on the three-phase voltage time series data to obtain three-phase GAF two-dimensional image data; S22. Taking the outermost monitoring node in the grid topology as the base point, calculate the similarity between the three-phase voltage time series data of the base point and the three-phase voltage time series data of the remaining monitoring nodes using the DTW algorithm; S23. Based on the similarity, the three-phase GAF two-dimensional image data are stacked into a three-channel spatiotemporal three-dimensional tensor.

[0008] Preferably, the step S21 specifically includes the following steps: S211. Normalize the three-phase voltage time series data. The calculation formula for the normalization is:

[0009] in, represents the normalized three-phase voltage time series data, represents the three-phase voltage time series before normalization, Represents the three-phase voltage timing sequence The Time series data, Represents the three-phase voltage timing sequence The maximum value in Represents the three-phase voltage timing sequence The minimum value in ; S212. Convert the normalized time series data into polar coordinates, where the cosine value of the polar angle is the time series data value, and the polar diameter of the polar coordinate is the timestamp. The specific calculation formula is:

[0010] in, represents the polar angle, represents the inverse cosine function, represents the polar axis, Represents time series data The timestamp of the point, Indicates the number of all time points contained in the time series data; S213. Use the Gram matrix to perform Gram angular field transformation on the polar coordinate time series data to obtain three-phase GAF two-dimensional image data. The specific calculation formula is:

[0011] in, represents three-phase GAF two-dimensional image data, Represents the three-phase voltage time series in polar coordinate form The transpose of Represents the three-phase voltage time series in polar coordinate form The Time series data, Represents the inner product of two vectors, that is, the correlation information of features with different timestamps.

[0012] As an example, the step S22 is specifically as follows: according to the voltage timing sequence of the base point And the voltage timing sequence of the remaining monitoring nodes , build a Matrix ,in, Indicates the first voltage in the base point timing sequence. Time series data, Indicates the voltage timing sequence of the remaining monitoring nodes. Time series data, matrix Chinese elements Indicates the Voltage timing sequence With the Voltage timing sequence The Euclidean distance between , ; From the matrix Find a line from arrive The path makes the path have the minimum cumulative Euclidean distance, and the minimum cumulative Euclidean distance is the similarity between the three-phase voltage time series data of the base point and the three-phase voltage time series data of the remaining monitoring nodes.

[0013] Preferably, the CNN+Transformer series model in step S3 includes a first 3DCNN layer, a maximum pooling layer, a series module of M convolutional layers and a Transformer encoder, a data flattening layer, a fully connected layer and a Softmax layer connected in sequence.

[0014] Preferably, the series module of the convolutional layer and the Transformer encoder includes an average pooling layer, a second 3DCNN layer, a first Transformer encoder, a second Transformer encoder and a third 3DCNN layer connected in sequence.

[0015] Preferably, the first Transformer encoder and the second Transformer encoder both include a multi-head self-attention layer, a first residual connection and layer normalization layer, a feedforward fully connected layer, and a second residual connection and layer normalization layer connected in sequence.

[0016] Preferably, the multi-head attention mechanism of the multi-head self-attention layer is specifically as follows: splitting the T tokens of the input data of the multi-head self-attention layer into T / t groups, calculating t tokens in each group; using multiple stacked Transformer encoders to respectively calculate the T tokens of the input data of the multi-head self-attention layer; according to the calculation results of the multiple stacked Transformer encoders, using the average pooling layer to perform feature fusion to obtain the global features of the input data; Among them, when splitting the T tokens of the input data of the multi-head self-attention layer into T / t groups, the relative position encoding of each group of tokens is required. Specifically, a three-dimensional coordinate system is constructed according to the size of the spatiotemporal three-dimensional tensor. The three-dimensional coordinate system contains the coordinate information of the input data points of each multi-head self-attention layer; the length, width, and height coordinates of a certain point are subtracted from the length, width, and height coordinates of the remaining points to obtain the coordinate difference; the coordinate difference is added to obtain the distance relationship between the two coordinates, and finally the relative position encoding of all input data points is obtained. The specific calculation formula is:

[0017] in, represents the relative position encoding matrix, Represents the input data point Location and input data points The relative position information of , .

[0018] Preferably, the step S4 specifically includes the following formula: S41. Input the spatiotemporal three-dimensional tensor into the first 3DCNN layer for three-dimensional convolution processing to obtain the output value of the first 3DCNN layer , the specific calculation formula is:

[0019] in, Represents a three-dimensional space-time tensor through the convolution kernel The output obtained, that is, the output value after three-dimensional convolution processing, represents the space-time three-dimensional tensor, Represents the number of channels of the space-time three-dimensional tensor, , Represents the total number of channels of the space-time three-dimensional tensor, 、 and Respectively represent the length, width and height of the space-time three-dimensional tensor, 、 and Represent the length, width and height of the convolution kernel respectively, 、 and Respectively represent the step size of the convolution kernel in the length, width and height directions, Represents the convolution kernel The weight parameters in Represents the convolution kernel The bias on 、 and They respectively represent the range of movement of the three-dimensional convolution window in the length, width and height directions of the spatiotemporal three-dimensional tensor; S42. The output value of the first 3DCNN layer Input the maximum pooling layer for processing to obtain the output value of the maximum pooling layer , the specific calculation formula is:

[0020] in, Indicates the maximum value; S43. The output value of the maximum pooling layer Input is a series module of M convolutional layers and Transformer encoders, and the output is the fusion feature of the spatiotemporal three-dimensional tensor; S44. Flattening the fused features of the spatiotemporal three-dimensional tensor to obtain one-dimensional data; S45. Convert the number of one-dimensional data into the number of classifications for voltage sag state estimation through linear mapping.

[0021] Preferably, the voltage sag state estimation result in step S4 includes six voltage amplitude categories: category V-1, category V-2, category V-3, category V-4, category V-5 and category V-6, which correspond to different voltage amplitudes. Scope, specifically: Category V-1: ; Category V-2: ; Category V-3: ; Category V-4: ; Category V-5: ; Category V-6: .

[0022] The beneficial effects of the above preferred solution are: 1. The voltage sag time series data is converted into three-phase GAF two-dimensional image data through Gram angular field transform, which can reflect the correlation information of different timestamp features in the time series. The similarity of voltage sag time series data is calculated through the DTW algorithm to obtain the coupling relationship between monitoring nodes. The spatiotemporal mapping law of voltage sag data in the power grid is deeply explored, which can effectively improve the accuracy of voltage sag state estimation.

[0023] 2. The CNN+Transformer cascade model integrates the multi-scale feature extraction and noise-resistance capabilities of CNN with the global modeling capabilities of the transformer. It overcomes the shortcomings of CNN in building global contextual understanding and the poor robustness of the transformer, and can effectively capture the local and global information of voltage time series data.

[0024] 3. The multi-head attention mechanism splits the T tokens of input data into T / t groups and uses multiple stacked Transformer encoders to separately process the tokens of each group of input data. This effectively reduces the computational complexity of the traditional attention mechanism and overcomes the problem of high-dimensional information loss during the calculation process of the traditional attention mechanism. It can effectively explore the relative position relationship of the input samples in the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The figure shows a flow chart of a voltage sag state estimation method based on time-space three-dimensional tensor.

[0026] Figure 2 The figure shows the structure of the CNN+Transformer series model provided in an embodiment of the present invention.

[0027] Figure 3 The figure shows a structural diagram of a serial module of a convolutional layer and a Transformer encoder provided in an embodiment of the present invention.

[0028] Figure 4Shown is a structural diagram of a Transformer encoder provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the accompanying drawings are merely exemplary and are intended to illustrate the principles and spirit of the present invention, rather than to limit the scope of the present invention.

[0030] Before describing the specific embodiments of the present invention, in order to make the solutions of the present invention clearer and more complete, the abbreviations and key terms used in the present invention are first defined as follows: Voltage sag: refers to a power quality disturbance with a voltage amplitude in the range of 0.1~0.9pu and a duration of 0.5T~1min, where T is the fundamental frequency period.

[0031] Voltage sag state estimation: refers to estimating the voltage sag amplitude at unmonitored points based on data from monitored points.

[0032] Example: like Figure 1 As shown, a voltage sag state estimation method based on time-space three-dimensional tensor includes the following steps: S1. Collect actual detection data of the power grid and obtain the three-phase voltage time series data of the power grid; S2. Perform Gram angle field transformation and similarity calculation on the three-phase voltage time series data to obtain a three-dimensional space-time tensor; S3. Build a CNN+Transformer tandem model; S4. Input the spatiotemporal three-dimensional tensor into the CNN+Transformer series model and output the voltage sag state estimation result.

[0033] In this embodiment, step S2 specifically includes the following sub-steps: S21. Perform a Gram angular field (GAF) transform on the ABC three-phase voltage time series data to obtain ABC three-phase GAF two-dimensional image data; S22. Using the outermost monitoring node in the grid topology as a base point, calculate the similarity between the ABC three-phase voltage time series data of the base point and the ABC three-phase voltage time series data of each remaining monitoring node using a dynamic time warping (DTW) algorithm; S23. Stack the three-phase GAF two-dimensional image data of ABC from high to low according to the similarity to obtain the three-dimensional space-time data of ABC, and stack the three-dimensional space-time data of ABC according to the ABC three-phase sequence into a three-channel space-time three-dimensional tensor.

[0034] In this embodiment, step S21 specifically includes the following steps: S211. Normalize the three-phase voltage time series data. The calculation formula for the normalization is:

[0035] in, represents the normalized three-phase voltage time series data, represents the three-phase voltage time series before normalization, Represents the three-phase voltage timing sequence The Time series data, Represents the three-phase voltage timing sequence The maximum value in Represents the three-phase voltage timing sequence The minimum value in ; S212. Convert the normalized time series data into polar coordinates, where the cosine value of the polar angle is the time series data value, and the polar diameter of the polar coordinate is the timestamp. The specific calculation formula is:

[0036] in, represents the polar angle, represents the inverse cosine function, represents the polar axis, Represents time series data The timestamp of the point, Indicates the number of all time points contained in the time series data; in the polar coordinate system, each time series point data contains two pieces of information, one is the normalized three-phase voltage time series data of the data point, and the other is the time series position of the data point, that is, the polar axis The time relationship of the three-phase voltage time series data is retained, and the polar angle The numerical relationship of the three-phase voltage time series data at a certain time stamp is retained.

[0037] S213. Use the Gram matrix to perform Gram angular field transformation on the polar coordinate time series data to obtain three-phase GAF two-dimensional image data. The specific calculation formula is:

[0038] in, represents three-phase GAF two-dimensional image data, Represents the three-phase voltage time series in polar coordinate form The transpose of Represents the three-phase voltage time series in polar coordinate form The Time series data, It represents the inner product of two vectors, that is, the correlation information of different timestamp features, which can be considered as the similarity of the two vectors, that is, the correlation information of different timestamp features.

[0039] In this embodiment, the current voltage sag data-driven state estimation method does not consider the degree of impact of a single set of data on the global situation during the pre-processing of input data, that is, the spatial coupling characteristics between monitoring nodes, which will lead to a decrease in the accuracy of state estimation. Therefore, it is also necessary to consider the connection between each monitoring node. When a voltage sag occurs, the closer the spatial distance between the monitoring node and the fault location and the higher the degree of electrical coupling, the deeper the impact, resulting in more severe deformation of the voltage waveform of the node. Therefore, the DTW algorithm can be used to measure the similarity between the time series of two different monitoring nodes to reflect the topological space and electrical coupling between the monitoring nodes. The step S22 is specifically as follows: according to the voltage timing sequence of the base point And the voltage timing sequence of the remaining monitoring nodes , build a Matrix ,in, Indicates the first voltage in the base point timing sequence. Time series data, Indicates the voltage timing sequence of the remaining monitoring nodes. Time series data, matrix Chinese elements Indicates the Voltage timing sequence With the Voltage timing sequence The Euclidean distance between , ; From the matrix Find a line from arrive The path makes the path have the minimum cumulative Euclidean distance, where the minimum cumulative Euclidean distance is the similarity between the three-phase voltage time series data of the base point and the three-phase voltage time series data of the remaining monitoring nodes. The smaller the cumulative Euclidean distance, the higher the similarity.

[0040] In this embodiment, if Figure 2 As shown, the CNN+Transformer series model in step S3 includes a first three-dimensional convolutional (3DCNN) layer, a maximum pooling layer, a series module of multiple convolutional layers and Transformer encoders, a data flattening layer, a fully connected layer, and a Softmax layer connected in sequence.

[0041] like Figure 3As shown, the series module of the convolutional layer and the Transformer encoder includes an average pooling layer, a second 3DCNN layer, multiple stacked Transformer encoders and a third 3DCNN layer connected in sequence.

[0042] like Figure 4 As shown, the multiple stacked Transformer encoders each include a multi-head self-attention layer, a first residual connection and layer normalization layer, a feedforward fully connected layer, and a second residual connection and layer normalization layer connected in sequence.

[0043] The algorithmic complexity of the attention mechanism is proportional to the square of the input data size, and the complexity is Therefore, the multi-head attention mechanism of the multi-head self-attention layer in this embodiment is as follows: the T tokens of the input data of the multi-head self-attention layer are split into T / t groups, and each group calculates t tokens, which can effectively reduce the computational complexity to ; Use multiple stacked Transformer encoders to calculate T tokens of the input data of the multi-head self-attention layer respectively; Based on the calculation results of multiple stacked Transformer encoders, use the average pooling layer to perform feature fusion, while obtaining the global features of the input data, which can effectively reduce the complexity of the calculation; Among them, when splitting the T minimum processing units (tokens) of the input data of the multi-head self-attention layer into T / t groups, it is necessary to perform relative position encoding on each group of tokens. Specifically, a three-dimensional coordinate system is constructed according to the size of the spatiotemporal three-dimensional tensor. The three-dimensional coordinate system contains the coordinate information of the input data points of each multi-head self-attention layer; the length, width, and height coordinates of a certain point are subtracted from the length, width, and height coordinates of the remaining points to obtain the coordinate difference; the coordinate difference is added to obtain the distance relationship between the two coordinates, and finally the relative position encoding of all input data points is obtained. The specific calculation formula is:

[0044] in, represents the relative position encoding matrix, Represents the input data point Location and input data points The relative position information of , ; Attention of multi-head self-attention layer The calculation formula is:

[0045] in, 、 and Represents the mapping of the input data of the multi-head self-attention layer, represents the activation function, A mapping representing the input The transpose of Represents the vector dimension of each point in the input data of the multi-head self-attention layer, which is the number of channels here, Represents the relative position encoding bias of the input data of the multi-head self-attention layer.

[0046] In this embodiment, step S4 specifically includes the following formula: S41. Input the spatiotemporal three-dimensional tensor into the first 3DCNN layer for three-dimensional convolution processing to obtain the output value of the first 3DCNN layer , the specific calculation formula is:

[0047] in, Represents a three-dimensional space-time tensor through the convolution kernel The output obtained, that is, the output value after three-dimensional convolution processing, represents the space-time three-dimensional tensor, Represents the number of channels of the space-time three-dimensional tensor, , Represents the total number of channels of the space-time three-dimensional tensor, 、 and Respectively represent the length, width and height of the space-time three-dimensional tensor, 、 and Represent the length, width and height of the convolution kernel respectively, 、 and Respectively represent the step size of the convolution kernel in the length, width and height directions, Represents the convolution kernel The weight parameters in Represents the convolution kernel The bias on 、 and They respectively represent the range of movement of the three-dimensional convolution window in the length, width and height directions of the spatiotemporal three-dimensional tensor; S42. The output value of the first 3DCNN layer Input the maximum pooling layer for processing to obtain the output value of the maximum pooling layer , the specific calculation formula is:

[0048] in, Indicates the maximum value; S43. The output value of the maximum pooling layer Input is a series module of M convolutional layers and Transformer encoders, and the output is the fusion feature of the spatiotemporal three-dimensional tensor; S44. Flattening the fused features of the spatiotemporal three-dimensional tensor to obtain one-dimensional data; S45. Convert the number of one-dimensional data into the number of classifications for voltage sag state estimation through linear mapping.

[0049] In this embodiment, the voltage sag state estimation result in step S4 includes six voltage amplitude categories: category V-1, category V-2, category V-3, category V-4, category V-5 and category V-6, which correspond to different voltage amplitudes. Scope, specifically: Category V-1: ; Category V-2: ; Category V-3: ; Category V-4: ; Category V-5: ; Category V-6: .

[0050] The voltage sag state estimation method based on spatiotemporal three-dimensional tensor proposed in the present invention deeply explores the spatiotemporal mapping law of voltage sag data in the power grid, converts the voltage sag time series data into an image representation that can reflect the correlation information of different timestamp features in the time series through Gram angle field transformation, and obtains the coupling relationship between monitoring nodes by calculating the similarity of time series data. Based on this, the image data is stacked into a three-dimensional tensor to reflect the spatial characteristics of the system; a CNN+Transformer series model is adopted, which has excellent local and global feature extraction capabilities and strong robustness, and the attention mechanism therein is improved, which effectively reduces the computational complexity of the traditional attention mechanism and overcomes the problem of high-dimensional information loss in the calculation process of the traditional attention mechanism; the preprocessed data is input into the model for training to obtain the state estimation result. Through the above method, the accuracy and practicality of voltage sag state estimation can be effectively improved.

[0051] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A voltage sag state estimation method based on time-space three-dimensional tensor, characterized in that: The method comprises the following steps: S1. Collect actual detection data of the power grid and obtain the three-phase voltage time series data of the power grid; S2. Perform Gram angle field transformation and similarity calculation on the three-phase voltage time series data to obtain a three-dimensional space-time tensor; S3. Build a CNN+Transformer tandem model; S4. Input the spatiotemporal three-dimensional tensor into the CNN+Transformer series model and output the voltage sag state estimation result.

2. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 1, characterized in that: The step S2 specifically includes the following sub-steps: S21. Performing Gram angle field transform on the three-phase voltage time series data to obtain three-phase GAF two-dimensional image data; S22. Taking the outermost monitoring node in the grid topology as the base point, the similarity between the three-phase voltage time series data of the base point and the three-phase voltage time series data of the remaining monitoring nodes is calculated using the DTW algorithm; S23. Based on the similarity, the three-phase GAF two-dimensional image data are stacked into a three-channel spatiotemporal three-dimensional tensor.

3. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 2, characterized in that: The step S21 specifically includes the following steps: S211. Normalize the three-phase voltage time series data. The calculation formula for the normalization is: in, represents the normalized three-phase voltage time series data, represents the three-phase voltage time series before normalization, Represents the three-phase voltage timing sequence The Time series data, Represents the three-phase voltage timing sequence The maximum value in Represents the three-phase voltage timing sequence The minimum value in ; S212. Convert the normalized time series data into polar coordinates, where the cosine value of the polar angle is the time series data value, and the polar diameter of the polar coordinate is the timestamp. The specific calculation formula is: in, represents the polar angle, represents the inverse cosine function, represents the polar axis, Represents time series data The timestamp of the point, Indicates the number of all time points contained in the time series data; S213. Use the Gram matrix to perform Gram angular field transformation on the polar coordinate time series data to obtain three-phase GAF two-dimensional image data. The specific calculation formula is: in, represents three-phase GAF two-dimensional image data, Represents the three-phase voltage time series in polar coordinate form The transpose of Represents the three-phase voltage time series in polar coordinate form The Time series data, Represents the inner product of two vectors, that is, the correlation information of features with different timestamps.

4. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 2, characterized in that: The step S22 is specifically as follows: according to the voltage timing sequence of the base point And the voltage timing sequence of the remaining monitoring nodes , build a Matrix ,in, Indicates the first voltage in the base point timing sequence. Time series data, Indicates the voltage timing sequence of the remaining monitoring nodes. Time series data, matrix Chinese elements Indicates the Voltage timing sequence With the Voltage timing sequence The Euclidean distance between , ; From the matrix Find a line from arrive The path makes the path have the minimum cumulative Euclidean distance, and the minimum cumulative Euclidean distance is the similarity between the three-phase voltage time series data of the base point and the three-phase voltage time series data of the remaining monitoring nodes.

5. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 1, characterized in that: The CNN+Transformer series model in step S3 includes a first 3DCNN layer, a maximum pooling layer, a series module of multiple convolutional layers and Transformer encoders, a data flattening layer, a fully connected layer, and a Softmax layer connected in sequence.

6. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 5, characterized in that: The series module of the convolutional layer and the Transformer encoder includes an average pooling layer, a second 3DCNN layer, multiple stacked Transformer encoders and a third 3DCNN layer connected in sequence.

7. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 6, characterized in that: The multiple stacked Transformer encoders each include a multi-head self-attention layer, a first residual connection and layer normalization layer, a feedforward fully connected layer, and a second residual connection and layer normalization layer connected in sequence.

8. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 7, characterized in that: The multi-head attention mechanism of the multi-head self-attention layer is as follows: splitting the T tokens of the input data of the multi-head self-attention layer into T / t groups, calculating t tokens for each group; using multiple stacked Transformer encoders to calculate the T tokens of the input data of the multi-head self-attention layer respectively; Based on the calculation results of multiple stacked Transformer encoders, the average pooling layer is used to perform feature fusion to obtain the global features of the input data; Among them, when splitting the T tokens of the input data of the multi-head self-attention layer into T / t groups, the relative position encoding of each group of tokens is required. Specifically, a three-dimensional coordinate system is constructed according to the size of the spatiotemporal three-dimensional tensor. The three-dimensional coordinate system contains the coordinate information of the input data points of each multi-head self-attention layer; the length, width, and height coordinates of a certain point are subtracted from the length, width, and height coordinates of the remaining points to obtain the coordinate difference; the coordinate difference is added to obtain the distance relationship between the two coordinates, and finally the relative position encoding of all input data points is obtained. The specific calculation formula is: in, represents the relative position encoding matrix, Represents the input data point Location and input data points The relative position information of , .

9. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 8, characterized in that: The step S4 specifically includes the following formula: S41. Input the spatiotemporal three-dimensional tensor into the first 3DCNN layer for three-dimensional convolution processing to obtain the output value of the first 3DCNN layer , the specific calculation formula is: in, Represents the three-dimensional space-time tensor through the convolution kernel The output obtained, that is, the output value after three-dimensional convolution processing, represents the space-time three-dimensional tensor, Represents the number of channels of the three-dimensional space-time tensor, , Represents the total number of channels of the space-time three-dimensional tensor, 、 and Respectively represent the length, width and height of the space-time three-dimensional tensor, 、 and Represent the length, width and height of the convolution kernel respectively, 、 and Respectively represent the step size of the convolution kernel in the length, width and height directions, Represents the convolution kernel The weight parameters in Represents the convolution kernel The bias on 、 and They respectively represent the range of movement of the three-dimensional convolution window in the length, width and height directions of the spatiotemporal three-dimensional tensor; S42. The output value of the first 3DCNN layer Input the maximum pooling layer for processing to obtain the output value of the maximum pooling layer , the specific calculation formula is: in, Indicates the maximum value; S43. The output value of the maximum pooling layer Input is a series module of M convolutional layers and Transformer encoders, and the output is the fusion feature of the spatiotemporal three-dimensional tensor; S44. Flattening the fused features of the spatiotemporal three-dimensional tensor to obtain one-dimensional data; S45. Convert the number of one-dimensional data into the number of classifications for voltage sag state estimation through linear mapping.

10. The voltage sag state estimation method based on time-space three-dimensional tensor according to claim 1, characterized in that: The voltage sag state estimation result in step S4 includes six voltage amplitude categories: category V-1, category V-2, category V-3, category V-4, category V-5 and category V-6, which correspond to different voltage amplitudes. Scope, specifically: Category V-1: ; Category V-2: ; Category V-3: ; Category V-4: ; Category V-5: ; Category V-6: .

Images