An extreme weather influence on power grid power generation power evaluation method, system and device

Abstract

The present application relates to the technical field of power grid safety risk assessment, in particular to a method, system and device for evaluating the influence of extreme weather on power generation, which first acquires meteorological cloud maps and micro-meteorological data corresponding to extreme weather; secondly inputs the meteorological cloud maps into a VIT prediction model for prediction to obtain a first sub-decision; then inputs the micro-meteorological data into a Transformer prediction model for prediction to obtain a second sub-decision; finally inputs the first sub-decision and the second sub-decision into a multi-modal fusion decision maker for cascaded fusion to obtain the probability of the influence of extreme weather on power generation. The present application combines the prediction results of two modes of meteorological cloud maps and micro-meteorological data for decision fusion, comprehensively collects different dimensional information when extreme weather occurs, and accurately predicts the influence of extreme weather on new energy power grid power generation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid safety risk assessment, in particular to an extreme weather influence on power grid power generation power evaluation method, system and device. BACKGROUND

[0002] Global warming is becoming increasingly severe, and extreme weather is becoming more and more frequent. Some extreme weather will threaten the normal operation of new energy power grids, resulting in a decrease in power generation power in a short period of time. Especially for high-proportion new energy power grids, a significant decrease in new energy output may even affect the normal operation of the entire power system. Therefore, accurately predicting the influence of extreme weather on new energy power grid power generation power and judging whether the power grid is at risk are of great research significance for improving the power grid's ability to cope with extreme weather disasters and responding to emergency recovery in a timely manner.

[0003] At present, the prediction of the influence of extreme weather on new energy power grid power generation power is mainly divided into two methods: traditional modeling and machine learning. For the prediction of the influence of extreme weather on new energy power grid power generation power, traditional modeling is often very complex and has low prediction accuracy; most existing machine learning models only consider single-dimensional data, and the considered modalities are single, which cannot fully utilize the limited data and cannot accurately predict the influence of extreme weather on new energy power grid power generation power. SUMMARY

[0004] The purpose of the present application is to provide an extreme weather influence on power grid power generation power evaluation method, system and device, which evaluates the influence of two-dimensional extreme weather on power grid power generation power to improve the evaluation accuracy.

[0005] To achieve the above-mentioned purpose, the present application provides an extreme weather influence on power grid power generation power evaluation method, which comprises:

[0006] Obtaining meteorological cloud images and micro-meteorological data corresponding to extreme weather;

[0007] Inputting the meteorological cloud images into a VIT prediction model for prediction to obtain a first sub-decision;

[0008] Inputting the micro-meteorological data into a Transformer prediction model for prediction to obtain a second sub-decision;

[0009] Cascade fusion of the first sub-decision and the second sub-decision into a multi-modal fusion decision maker to obtain the probability of the influence on the power grid power generation power.

[0010] Optionally, the inputting of the meteorological cloud images into the VIT prediction model for prediction to obtain the first sub-decision specifically comprises:

[0011] The weather cloud picture is cropped and converted to obtain an input matrix corresponding to the weather cloud picture;

[0012] The classification category information vector is embedded in the input matrix corresponding to the weather cloud picture to obtain an encoding feature matrix;

[0013] The position information matrix and the encoding feature matrix are fused to obtain a position fusion feature matrix;

[0014] The position fusion feature matrix is input into a first Transformer encoder for learning feature extraction to obtain a first learning feature matrix;

[0015] The first learning feature matrix is input into a first MLP decision maker for decision extraction to obtain a first sub-decision.

[0016] Optionally, the position fusion feature matrix is input into a first Transformer encoder for learning feature extraction to obtain a first learning feature matrix, specifically including:

[0017] The position fusion feature matrix is input into a multi-head attention module to extract attention weights to obtain an attention weight matrix;

[0018] The attention weight matrix is input into a fully connected module for learning feature extraction to obtain a first learning feature matrix.

[0019] Optionally, the micro-meteorological data is input into a Transformer prediction model for prediction to obtain a second sub-decision, specifically including:

[0020] The micro-meteorological data is arranged into a micro-meteorological data sequence according to wind speed, wind direction, air pressure, humidity, temperature, and precipitation;

[0021] The micro-meteorological data sequence is input into a second Transformer encoder for learning feature extraction to obtain a second learning feature matrix;

[0022] The second learning feature matrix is input into a second MLP decision maker for decision extraction to obtain a second sub-decision.

[0023] Optionally, the first sub-decision and the second sub-decision are input into a multi-modal fusion decision maker for cascaded fusion to obtain a probability of affecting power generation of the power grid, specifically including:

[0024] The first sub-decision and the second sub-decision are input into a classifier for cascaded splicing operation to obtain a splicing sequence;

[0025] The splicing sequence is input into a multi-modal fusion decision maker for fusion to obtain a probability of affecting power generation of the power grid.

[0026] This invention also provides a system for assessing the impact of extreme weather on power grid generation, the system comprising:

[0027] The acquisition module is used to acquire meteorological cloud images and micro-meteorological data corresponding to extreme weather events.

[0028] The first prediction module is used to input the meteorological cloud image into the VIT prediction model for prediction and obtain the first sub-decision;

[0029] The second prediction module is used to input the micro-meteorological data into the Transformer prediction model for prediction, and obtain the second sub-decision.

[0030] The cascaded fusion module is used to input the first sub-decision and the second sub-decision into the multimodal fusion decision-maker for cascaded fusion to obtain the probability of the impact on the power generation of the power grid.

[0031] Optionally, the first prediction module specifically includes:

[0032] The cropping and conversion unit is used to crop and convert the meteorological cloud image to obtain the input matrix corresponding to the meteorological cloud image.

[0033] A vector embedding unit is used to embed a classification category information vector into the input matrix corresponding to the meteorological cloud image to obtain an encoded feature matrix.

[0034] A matrix fusion unit is used to fuse the location information matrix and the encoded feature matrix to obtain a location fusion feature matrix;

[0035] The first learning feature extraction unit is used to input the position fusion feature matrix into the first Transformer encoder to extract learning features and obtain the first learning feature matrix.

[0036] The first sub-decision extraction unit is used to input the first learned feature matrix into the first MLP decision generator to extract the first sub-decision.

[0037] Optionally, the learning feature extraction unit specifically includes:

[0038] The attention weight extraction subunit is used to input the position fusion feature matrix into the multi-head attention module to extract attention weights and obtain an attention weight matrix.

[0039] The learning feature extraction subunit is used to input the attention weight matrix into the fully connected module to extract learning features and obtain the first learning feature matrix.

[0040] Optionally, the second prediction module specifically includes:

[0041] The sorting unit is used to arrange the micrometeorological data into a micrometeorological data sequence according to wind speed, wind direction, air pressure, humidity, temperature, and precipitation.

[0042] The second learning feature extraction unit is used to input the micro-meteorological data sequence into the second Transformer encoder to extract learning features and obtain the second learning feature matrix.

[0043] The second sub-decision extraction unit is used to input the second learned feature matrix into the second MLP decision generator to extract the decision and obtain the second sub-decision.

[0044] The present invention also provides an apparatus for assessing the impact of extreme weather on power grid generation, the apparatus including a processor for running a program, wherein the program executes the method for assessing the impact of extreme weather on power grid generation as described in any one of the above.

[0045] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0046] This invention provides a method, system, and apparatus for assessing the impact of extreme weather on power grid generation. First, meteorological cloud maps and micrometeorological data corresponding to the extreme weather are acquired. Second, the meteorological cloud maps are input into a VIT (Vibration and Influence) prediction model for prediction, yielding a first sub-decision. Then, the micrometeorological data are input into a Transformer prediction model for prediction, yielding a second sub-decision. Finally, the first and second sub-decisions are input into a multimodal fusion decision-maker for cascaded fusion to obtain the probability of the impact on power grid generation. This invention combines the prediction results from both meteorological cloud maps and micrometeorological data for decision fusion, integrating information from different dimensions collected during extreme weather events to accurately predict the impact of extreme weather on the power generation of new energy power grids. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the method for assessing the impact of extreme weather on power grid generation in an embodiment of the present invention. Figure 1 ;

[0049] Figure 2 This is a flowchart of the method for assessing the impact of extreme weather on power grid generation in an embodiment of the present invention. Figure 2 ;

[0050] Figure 3 This is a diagram of the VIT prediction model according to an embodiment of the present invention;

[0051] Figure 4 This is a structural diagram of the Transformer encoder according to an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] The purpose of this invention is to provide a method, system, and device for assessing the impact of extreme weather on power grid generation, which assesses the impact of extreme weather on power grid generation based on a two-dimensional perspective, thereby improving the accuracy of the assessment.

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1

[0056] like Figures 1-2 As shown, this invention discloses a method for assessing the impact of extreme weather on power grid generation, the method comprising:

[0057] Step S1: Obtain meteorological cloud images and micro-meteorological data corresponding to extreme weather.

[0058] Step S2: Input the meteorological cloud image into the VIT prediction model to make a prediction and obtain the first sub-decision.

[0059] Step S3: Input the micro-meteorological data into the Transformer prediction model for prediction to obtain the second sub-decision.

[0060] Step S4: Input the first sub-decision and the second sub-decision into the multimodal fusion decision-maker for cascade fusion to obtain the probability of the impact on the power generation of the power grid.

[0061] The following is a detailed discussion of each step:

[0062] Step S1: Obtain meteorological cloud images and micrometeorological data corresponding to extreme weather; in this embodiment, extreme weather refers to weather and climate conditions that deviate significantly from their average state. Micrometeorological data includes wind speed, wind direction, air pressure, humidity, temperature, and precipitation.

[0063] Before step S1, the following step is included: Step S5: Constructing a VIT network based on historical meteorological cloud images and corresponding labels, specifically including:

[0064] Step S51: Obtain multiple historical meteorological cloud images of the extreme weather events that affected the power generation of new energy grids, and label each historical meteorological cloud image; if the label of the historical meteorological cloud image is 0, it means that the extreme weather did not have a significant impact on the power generation of new energy grids; if the label of the historical meteorological cloud image is 1, the extreme weather has had a significant impact on the power generation of new energy grids.

[0065] Step S52: Construct a training map set and a test map set based on multiple historical meteorological cloud images and the corresponding labels of each historical meteorological cloud image; train the VIT network using the training map set to obtain an initial VIT prediction model; test the initial VIT prediction model using the test map set to determine whether the VIT test termination condition is met; if the VIT test termination condition is met, use the initial VIT prediction model as the final VIT prediction model and output the VIT prediction model.

[0066] The VIT test termination condition in this embodiment can be set according to actual needs, and can be a preset number of times or the test accuracy.

[0067] Before step S1, the following step is included: Step S6: Constructing a Transformer prediction model based on micrometeorological data and corresponding labels, specifically including:

[0068] Step S61: Obtain multiple historical micro-meteorological data points for the extreme weather events that affected the power generation of new energy grids. Arrange the historical micro-meteorological data in the order of wind speed, wind direction, air pressure, humidity, temperature, and precipitation to form a data sequence. Label each historical micro-meteorological data sequence according to the sequence order. If the label of a historical micro-meteorological data sequence is 0, it means that the extreme weather event did not have a significant impact on the power generation of the new energy grid; if the label of a historical micro-meteorological data sequence is 1, it means that the extreme weather event has had a significant impact on the power generation of the new energy grid.

[0069] Step S62: Construct a training set and a test set based on multiple historical micro-meteorological data and the corresponding labels of each historical micro-meteorological data; train the Transformer encoder using the training set to obtain an initial Transformer prediction model; test the initial Transformer prediction model using the test set to determine whether the Transformer test termination condition is met; if the Transformer test termination condition is met, use the initial Transformer prediction model as the final initial Transformer prediction model and output the Transformer prediction model.

[0070] The termination condition for the Transformer test in this embodiment can be set according to actual needs, and can be a preset number of times or the test accuracy.

[0071] Figure 2 Steps S21-23 are omitted; please refer to the documentation for details. Figure 3 ,like Figures 2-3 As shown, the meteorological cloud image is input into the VIT prediction model for prediction, resulting in the first sub-decision, which specifically includes:

[0072] Step S21: Crop and transform the meteorological cloud image to obtain the input matrix corresponding to the meteorological cloud image, specifically including:

[0073] Step S211: Input the meteorological cloud image into the cropping layer for cropping processing to obtain multiple image patches. In this embodiment, the meteorological cloud images are input sequentially, one at a time. The input meteorological cloud images are cut into patches of the same size. The size of the patches can be set according to actual needs. The resulting image patches are three-dimensional matrices. For subsequent conversion, they must be two-dimensional matrices. Therefore, it is necessary to convert the three-dimensional matrix into a two-dimensional matrix. For details of the conversion, please refer to the following steps:

[0074] Step S212: Input all image patches into the embedding layer for convolution operation to obtain the one-dimensional feature vector corresponding to each image patch.

[0075] Step S213: Input the one-dimensional feature vectors corresponding to all image patches into the matrix transformation layer for concatenation processing to obtain the input matrix corresponding to the meteorological cloud image. In this embodiment, each image patch corresponds to a one-dimensional feature vector. To reflect the characteristics of the meteorological cloud image, it is necessary to concatenate the one-dimensional feature vectors corresponding to all image patches to form a two-dimensional matrix. This two-dimensional matrix can be used as the input matrix of the model. The two-dimensional matrix A is represented as follows:

[0076] A n×v1 (1);

[0077]

[0078] v1 = a 2 ×3(3);

[0079] In the formula, n is the number of one-dimensional vector sequences, also known as the number of image patches, v1 is the feature vector dimension, H and W are the height and width of the meteorological cloud image, and a is the size of the patch set by the user.

[0080] Step S22: Embed the classification category information vector (i.e., class token vector) into the input matrix corresponding to the meteorological cloud image to obtain the encoding feature matrix, which is represented as follows:

[0081] A' (n+1)×v1 (4);

[0082] In the formula, n is the number of tokens, that is, the number of one-dimensional vector sequences, and v1 is the dimension of the feature vector.

[0083] To facilitate subsequent training and loss value calculation, a special class token vector is added to the learnable embedding in step S22. This vector contains only two classification categories: 0 and 1. This vector serves as the encoded feature of the image in subsequent steps. Specifically, the class token vector is added to the input matrix corresponding to the meteorological cloud image using learnable embedding; this step is called Learnable Embedding.

[0084] Step S23: Fuse the location information matrix and the coding feature matrix to obtain the location fusion feature matrix.

[0085] Converting an image into a sequence results in the loss of natural positional information between blocks. To enable the model to understand block positions, positional information must be added to each block. The positional information of all blocks forms a positional information matrix with the same dimensions as the encoded feature matrix processed in step S22. The positional information matrix is ​​then directly superimposed on the encoded feature matrix processed in step S22 without changing the matrix's dimensions. This step is called Position Embedding.

[0086] The location fusion feature matrix is ​​specifically represented as follows:

[0087] A” (n+1)×v1 (5);

[0088] In the formula, n is the number of tokens, that is, the number of one-dimensional vector sequences, and v1 is the dimension of the feature vector.

[0089] Step S24: Input the location fusion feature matrix into the first Transformer encoder for learning feature extraction to obtain the first learned feature matrix, specifically including:

[0090] Step S241: Merge the location feature matrix ( Figure 4 (Represented by Tokens) is input into the multi-head attention module to extract attention weights and obtain the attention weight matrix.

[0091] The location fusion feature matrix is ​​input into the multi-head self-attention layer, which calculates the attention weight corresponding to each location in the location fusion feature matrix to capture the relationship between different locations.

[0092] To alleviate the vanishing gradient problem and accelerate the training process, residual connections and layer normalization operations are performed on the output of the multi-head attention layer. This step constitutes the MSA module and can be repeated multiple times. The vector representation after the MSA module is as follows:

[0093] z l =MSA(LN(z) l-1 ))+z l-1 (6);

[0094] In the formula, z l-1 Let z be the position fusion feature matrix input to the (l-1)th layer. l 'This is the attention weight matrix after processing by the MSA module. MSA() is the multi-head attention layer operation, and LN() is the layer normalization operation.

[0095] Step S242: Input the attention weight matrix into the fully connected module for feature extraction to obtain the first learned feature matrix. The specific formula is as follows:

[0096] z l =MLP(LN(z) l '))+z l '(7);

[0097] In the formula, z l ' is the attention weight matrix output by the l-th layer MSA module, z l This is the first learned feature matrix after processing by the MLP module. MLP() is the fully connected module operation, and LN() is the layer normalization operation.

[0098] Step S242: Input the attention weight matrix into the fully connected module for feature extraction to obtain the first learned feature matrix, specifically including:

[0099] Step S2421: Input the attention weight matrix into the feedforward neural network layer for nonlinear transformation mapping, learn local features, and obtain the local feature matrix.

[0100] The inter-layer transfer function of a feedforward neural network layer is expressed as:

[0101] FFN(x)=max(0,xW1+b1)W2+b2(8);

[0102] In the formula, x represents a vector matrix, W1 and W2 are weight matrices, and b1 and b2 are bias matrices.

[0103] Step S2422: Input the local feature matrix into the normalization layer for residual connection and layer normalization operations to obtain the first learned feature matrix.

[0104] Step S2421 uses the attention weight matrix output by the multi-head self-attention module as input to the feedforward neural network layer. This layer performs a non-linear transformation mapping on the features, enabling the model to learn local features. Step S2422 processes the output of the feedforward neural network layer using residual connections and layer normalization operations. The two steps combined form a feature extraction learning module using a fully connected layer (MLP).

[0105] Step S25: Input the first learned feature matrix into the first MLP decision maker to extract the decision and obtain the first sub-decision.

[0106] The learned feature matrix extracted by the Transformer encoder (also known as the Transformer Encoder module) in step S24 is input into the first MLP decision-maker, which converts the extracted image features into sub-decisions for meteorological cloud image prediction. The inter-layer transfer function of the first MLP decision-maker is the same as that in formula (8).

[0107] Step S3: Input the micrometeorological data into the Transformer prediction model for prediction, and obtain the second sub-decision, which specifically includes:

[0108] Step S31: Arrange the micrometeorological data into a micrometeorological data sequence according to wind speed, wind direction, air pressure, humidity, temperature and precipitation.

[0109] Step S32: Input the micro-meteorological data sequence into the second Transformer encoder for learning feature extraction to obtain the second learning feature matrix; In this embodiment, the steps of the first Transformer encoder and the second Transformer encoder are the same, and will not be repeated here. See step S24 for details.

[0110] The first and second Transformer encoders have the same composition, both consisting of multiple Transformer Blocks. Each Transformer Block comprises a Multi-head Self-Attention Layer (MSA) module and a fully connected module (MLP module), as detailed below. Figure 4 As shown.

[0111] Step S33: Input the second learned feature matrix into the second MLP decision maker to extract the decision and obtain the second sub-decision.

[0112] The present invention converts the extracted micro-meteorological data sequence feature representation into a sub-decision of micro-meteorological data prediction, and the inter-layer transfer function of the second MLP decision-maker is the same as that in formula (8).

[0113] The first MLP decision-maker and the second MLP decision-maker of this invention can be the same MLP decision-maker or different MLP decision-makers. In this embodiment, the same MLP decision-maker is selected.

[0114] Step S4: Input the first and second sub-decisions into the multimodal fusion decision-maker for cascaded fusion to obtain the probability of their impact on the power generation of the power grid, specifically including:

[0115] Step S41: Input the first sub-decision and the second sub-decision into the classifier for cascade splicing operation to obtain the spliced ​​sequence.

[0116] The classifier is denoted as {L1(x),L2(x)}, where L1(x) and L2(x) represent the first and second sub-decisions of the input, respectively, and the concatenated output sequence is denoted as {y1,y2}.

[0117] Step S42: Input the spliced ​​sequence into the multimodal fusion decision-maker for fusion to obtain the probability of its impact on the power generation of the power grid. Specifically, this invention uses a binomial logistic regression model as the multimodal fusion decision-maker. The binomial logistic regression model is as follows:

[0118]

[0119] In the formula, x is the concatenated sequence, w is the weight matrix, and P(Y=1|x) is the probability that the extreme weather represented by the current input x will affect the power grid. In this invention, Y=1 indicates that it will affect the power grid. For example, P(Y=1|x)=0.8 indicates that there is an 80% probability that the rainstorm will affect the power grid.

[0120] This invention fully considers the significant impact of extreme weather disasters on renewable energy power generation; therefore, grids with a high proportion of renewable energy power generation are more susceptible to such disasters. This invention combines prediction results from two modalities—meteorological cloud maps and micrometeorological data—for decision fusion, integrating information from different dimensions collected during extreme weather events, thus demonstrating effectiveness and rationality. The deep neural network-based attention mechanism-based method proposed in this invention for predicting the impact of extreme weather disasters on grids with a high proportion of renewable energy provides guidance for grid companies to accurately predict the impact of extreme weather on the grid and to implement emergency response measures.

[0121] Example 2

[0122] This invention also discloses a system for assessing the impact of extreme weather on power grid generation, the system comprising:

[0123] The acquisition module is used to acquire meteorological cloud images and micrometeorological data corresponding to extreme weather.

[0124] The first prediction module is used to input the meteorological cloud image into the VIT prediction model for prediction and obtain the first sub-decision.

[0125] The second prediction module is used to input the micrometeorological data into the Transformer prediction model for prediction, and obtain the second sub-decision.

[0126] The cascaded fusion module is used to input the first sub-decision and the second sub-decision into the multimodal fusion decision-maker for cascaded fusion to obtain the probability of the impact on the power generation of the power grid.

[0127] As an optional implementation, the first prediction module of the present invention specifically includes:

[0128] The cropping and conversion unit is used to crop and convert the meteorological cloud map to obtain the input matrix corresponding to the meteorological cloud map.

[0129] The vector embedding unit is used to embed classification category information vectors into the input matrix corresponding to the meteorological cloud map to obtain the encoded feature matrix.

[0130] The matrix fusion unit is used to fuse the location information matrix and the encoded feature matrix to obtain a location fusion feature matrix.

[0131] The first learning feature extraction unit is used to input the position fusion feature matrix into the first Transformer encoder to extract learning features and obtain the first learning feature matrix.

[0132] The first sub-decision extraction unit is used to input the first learned feature matrix into the first MLP decision generator to extract the first sub-decision.

[0133] As an optional implementation, the feature extraction unit of this invention specifically includes:

[0134] The attention weight extraction subunit is used to input the position fusion feature matrix into the multi-head attention module to extract attention weights and obtain an attention weight matrix.

[0135] The learning feature extraction subunit is used to input the attention weight matrix into the fully connected module to extract learning features and obtain the first learning feature matrix.

[0136] As an optional implementation, the second prediction module of the present invention specifically includes:

[0137] The sorting unit is used to arrange the micrometeorological data into a micrometeorological data sequence according to wind speed, wind direction, air pressure, humidity, temperature, and precipitation.

[0138] The second learning feature extraction unit is used to input the micro-meteorological data sequence into the second Transformer encoder to extract learning features and obtain the second learning feature matrix.

[0139] The second sub-decision extraction unit is used to input the second learned feature matrix into the second MLP decision generator to extract the decision and obtain the second sub-decision.

[0140] As an optional implementation, the cascaded fusion module of the present invention specifically includes:

[0141] The cascaded splicing unit is used to input the first sub-decision and the second sub-decision into the classifier for cascaded splicing operation to obtain the spliced ​​sequence;

[0142] The fusion unit is used to input the spliced ​​sequence into the multimodal fusion decision-maker for fusion to obtain the probability of its impact on the power generation of the power grid.

[0143] The parts that are the same as in Example 1 are specifically referred to in Example 1, and will not be repeated here.

[0144] Example 3

[0145] The present invention also provides an apparatus for assessing the impact of extreme weather on power grid generation, the apparatus including a processor for running a program, wherein the program executes the method for assessing the impact of extreme weather on power grid generation as described in Example 1.

[0146] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0147] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for assessing the impact of extreme weather on power grid generation, characterized in that, The method includes: Acquire meteorological cloud images and micrometeorological data corresponding to extreme weather events; The meteorological cloud image is input into the VIT prediction model for prediction, and the first sub-decision is obtained; The micrometeorological data is input into the Transformer prediction model for prediction, resulting in the second sub-decision. The first sub-decision and the second sub-decision are input into a multimodal fusion decision-maker for cascade fusion to obtain the probability of their impact on the power generation of the power grid. The step of inputting the meteorological cloud image into the VIT prediction model for prediction to obtain the first sub-decision specifically includes: The meteorological cloud image is cropped and transformed to obtain the input matrix corresponding to the meteorological cloud image. The classification category information vector is embedded into the input matrix corresponding to the meteorological cloud image to obtain the encoded feature matrix; The location information matrix and the encoded feature matrix are fused to obtain the location fusion feature matrix; The location fusion feature matrix is ​​input into the first Transformer encoder for learning feature extraction to obtain the first learning feature matrix; The first learned feature matrix is ​​input into the first MLP decision-maker to extract the first sub-decision; The step of inputting the position fusion feature matrix into the first Transformer encoder for learning feature extraction to obtain the first learned feature matrix specifically includes: The location fusion feature matrix is ​​input into the multi-head attention module to extract attention weights, thus obtaining the attention weight matrix; The attention weight matrix is ​​input into the fully connected module for learning feature extraction to obtain the first learning feature matrix; The step of inputting the micrometeorological data into the Transformer prediction model for prediction, and obtaining the second sub-decision, specifically includes: The micrometeorological data are arranged into a micrometeorological data sequence according to wind speed, wind direction, air pressure, humidity, temperature, and precipitation. The micro-meteorological data sequence is input into the second Transformer encoder for learning feature extraction to obtain the second learning feature matrix; The second learned feature matrix is ​​input into the second MLP decision-maker to extract the decision and obtain the second sub-decision. The step of inputting the first sub-decision and the second sub-decision into a multimodal fusion decision-maker for cascaded fusion to obtain the probability of their impact on the power generation of the power grid specifically includes: The first sub-decision and the second sub-decision are input into the classifier for cascaded splicing to obtain the spliced ​​sequence. The spliced ​​sequence is input into a multimodal fusion decision-maker for fusion to obtain the probability of its impact on the power generation of the power grid.

2. A system for assessing the impact of extreme weather on power grid generation, characterized in that, The system includes: The acquisition module is used to acquire meteorological cloud images and micro-meteorological data corresponding to extreme weather events. The first prediction module is used to input the meteorological cloud image into the VIT prediction model for prediction and obtain the first sub-decision; The second prediction module is used to input the micro-meteorological data into the Transformer prediction model for prediction, and obtain the second sub-decision. The cascaded fusion module is used to input the first sub-decision and the second sub-decision into the multimodal fusion decision-maker for cascaded fusion to obtain the probability of the impact on the power generation of the power grid; The step of inputting the meteorological cloud image into the VIT prediction model for prediction to obtain the first sub-decision specifically includes: The meteorological cloud image is cropped and transformed to obtain the input matrix corresponding to the meteorological cloud image. The classification category information vector is embedded into the input matrix corresponding to the meteorological cloud image to obtain the encoded feature matrix; The location information matrix and the encoded feature matrix are fused to obtain the location fusion feature matrix; The location fusion feature matrix is ​​input into the first Transformer encoder for learning feature extraction to obtain the first learning feature matrix; The first learned feature matrix is ​​input into the first MLP decision-maker to extract the first sub-decision; The step of inputting the position fusion feature matrix into the first Transformer encoder for learning feature extraction to obtain the first learned feature matrix specifically includes: The location fusion feature matrix is ​​input into the multi-head attention module to extract attention weights, thus obtaining the attention weight matrix; The attention weight matrix is ​​input into the fully connected module for learning feature extraction to obtain the first learning feature matrix; The step of inputting the micrometeorological data into the Transformer prediction model for prediction, and obtaining the second sub-decision, specifically includes: The micrometeorological data are arranged into a micrometeorological data sequence according to wind speed, wind direction, air pressure, humidity, temperature, and precipitation. The micro-meteorological data sequence is input into the second Transformer encoder for learning feature extraction to obtain the second learning feature matrix; The second learned feature matrix is ​​input into the second MLP decision-maker to extract the decision and obtain the second sub-decision. The step of inputting the first sub-decision and the second sub-decision into a multimodal fusion decision-maker for cascaded fusion to obtain the probability of their impact on the power generation of the power grid specifically includes: The first sub-decision and the second sub-decision are input into the classifier for cascaded splicing to obtain the spliced ​​sequence. The spliced ​​sequence is input into a multimodal fusion decision-maker for fusion to obtain the probability of its impact on the power generation of the power grid.

3. A device for assessing the impact of extreme weather on power grid generation, characterized in that, The device includes a processor for running a program, wherein the program executes the method for assessing the impact of extreme weather on power grid generation as described in claim 1.

Images