Coastal region storm surge disaster prediction method based on multiple modes

Abstract

The invention discloses a coastal region storm surge disaster prediction method based on multiple modes, which combines multi-source meteorological environment data, three-dimensional geographic space data and storm surge historical occurrence data, and comprises the following steps: on one hand, performing long-term prediction and trend fitting on future storm surge conditions by using a multi-mode network; fitting to obtain a linear component and a nonlinear component related to storm surge prediction, combining the linear component and the nonlinear component, and converting the combined component into a storm surge prediction probability P1 through an activation function; on the other hand, for the three-dimensional geographic space data, respectively obtaining a predicted value of supervised learning of a labeled scene and a predicted value of hierarchical supervised learning of an unlabeled scene of the three-dimensional geographic space data; averaging the predicted values of the two, and converting the averaged predicted values into a storm surge prediction probability P2 through an activation function; and finally, weighting the storm surge prediction probability P1 and the storm surge prediction probability P2 to obtain a final storm surge prediction value.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological disaster risk prediction of coastlines, and in particular relates to a storm surge disaster prediction method for coastal areas based on multi-modality. Background Art

[0002] With the acceleration of urbanization, especially the rapid development of infrastructure in coastal areas, how to deal with the impact of natural disasters (such as storm surges, tidal fluctuations, etc.) on coastal infrastructure has become an urgent problem to be solved. Traditional disaster prevention and mitigation technologies have certain limitations in real-time data monitoring, environmental change response, infrastructure risk assessment, etc. Especially in dynamic environments (such as tides, storm surges and other frequently changing scenes), how to efficiently and accurately conduct disaster warnings, risk assessments and infrastructure monitoring has become the key to improving disaster response capabilities.

[0003] However, existing technologies mainly focus on static model construction and infrastructure management, lack sufficient consideration for dynamic risk assessment brought about by environmental changes, and are difficult to achieve deep integration and predictive analysis of real-time data. Therefore, how to build a system that can cope with dynamic environmental changes and accurately predict and assess storm surge disasters in coastal areas has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present invention mainly aims to overcome the deficiencies of the prior art and provides a method for predicting storm surge disasters in coastal areas based on multi-modality.

[0005] The present invention is achieved through the following technical solutions: A method for predicting storm surge disasters in coastal areas based on multi-modal, comprising the following steps: Step 1: Collect meteorological environment data, three-dimensional geographic spatial data and historical storm surge data of the target coastal area to be evaluated; Step 2: Based on the data collected in step 1, use a multimodal network to perform long-term prediction and trend fitting of future storm surge conditions; Step 2.1: Fit the linear relationship of the data collected in step 1 to obtain the linear component related to storm surge prediction ; Step 2.2: Fit the nonlinear relationship of the data collected in step 1 to obtain the nonlinear components related to storm surge prediction ; Step 2.3: Substitute the linear components of storm surge prediction obtained in step 2.1 into Nonlinear components related to the storm surge prediction obtained in step 2.2 Add together to get new storm surge prediction data ; Step 2.4: Substitute the storm surge prediction data obtained in step 2.3 into , converted into storm surge prediction probability through activation function ; Step 3: Combine the three-dimensional geospatial data obtained in step 1 and use supervised learning methods to predict storm surges; Step 3.1: Perform random shuffling data enhancement processing on the three-dimensional geospatial data; Step 3.2: Use the Jenks Natural Breaks algorithm (automatic segmentation algorithm) to generate dynamic thresholds for the three-dimensional geospatial data processed in step 3.1 to obtain the thresholds for each category. Three thresholds ; Step 3.3: Three thresholds based on step 3.2 , the pseudo labels of the mined objects in the unlabeled scenes in the 3D geospatial data are generated by hierarchical supervision, and the hierarchical pseudo labels are divided into three categories: high confidence, fuzzy and low confidence; Step 3.4 Training target loss function, total training loss = supervised learning loss for labeled scenes + Hierarchical supervised learning loss for unlabeled scenes ; By training the target loss function, we can get the predicted value of supervised learning for the labeled scene And the predicted values ​​of hierarchical supervised learning for unlabeled scenes ; Step 3.5: Set the predicted value Converted to storm surge prediction probability through activation function ; Step 4: Substitute the storm surge prediction probability obtained in step 2 And the storm surge prediction probability obtained in step 3 , weighted to obtain the final storm surge prediction value .

[0006] In the above technical solution, meteorological environment data include: wind speed, sea temperature, sea level, sea level pressure and air humidity with time attributes, forming a meteorological environment data set represented by a time series; three-dimensional geographic spatial data include: three-dimensional geographic spatial data of coastline changes, marine environment and infrastructure of target coastal areas with time attributes; storm surge historical occurrence data include: storm surge water height data with time attributes.

[0007] In the above technical solution, in step 3.3: Pseudo labels for high confidence classes , ; Pseudo-labeling of fuzzy categories , ; Pseudo labels for low confidence classes , ; represents the confidence score of the pseudo label x.

[0008] In the above technical solution, in step 3.4: 1) Supervised learning loss for labeled scenes: ; : total sample size; : Number of categories; the number of categories is 2, the first category is the occurrence of storm surge, represents the first category; the second category is when no storm surge occurs, represents the second category; :sample The true label of category c is one-hot encoded; Indicates that sample i belongs to category c, Indicates that sample i does not belong to category c; :sample The predicted probability of category c; 2) Hierarchical supervised learning loss for unlabeled scenes: ; : is the weight coefficient; High confidence loss: ; ; : is the loss function; : Pseudo label of sample i; : The predicted value of sample i: Blur loss: ; : Balance factor, used to adjust the importance of positive and negative samples; : Adjustment factor, used to reduce the influence of easy-to-classify samples and increase the attention of difficult-to-classify samples; : Increase the loss contribution of samples with smaller prediction probabilities; Low confidence loss: ; ; : Confidence threshold. A value less than this value indicates that the predicted value is close to the pseudo label and is considered valid. : Indicates discarding the gradient update of the sample or setting its loss to zero.

[0009] In the above technical solution, according to the storm surge prediction value , for disaster risk classification and decision-making: if , risk level = high risk; if , risk level = medium risk; if , risk level = low risk; =Threshold for high risk; = Low risk threshold.

[0010] The advantages and beneficial effects of the present invention are: The present invention combines multi-source meteorological environment data, three-dimensional geographic space data and historical storm surge occurrence data. On the one hand, a multimodal network is used to perform long-term prediction and trend fitting of future storm surge conditions, and linear components and nonlinear components related to storm surge prediction are obtained by fitting. The linear components and nonlinear components are combined and then converted into a storm surge prediction probability P1 through an activation function. On the other hand, for the three-dimensional geographic space data, the predicted values ​​of supervised learning of its labeled scenes and the predicted values ​​of hierarchical supervised learning of unlabeled scenes are obtained respectively; and the predicted values ​​of the two are averaged and converted into a storm surge prediction probability P2 through an activation function; finally, the storm surge prediction probability P1 and the storm surge prediction probability P2 are weighted to obtain a final storm surge prediction value. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of the storm surge disaster prediction method for coastal areas based on multi-modality of the present invention.

[0012] Figure 2 It is a trend diagram of the change of error performance such as mean error (ME), mean absolute error (MAE), root mean square error (RMSE), mean absolute scaled error (MASE) and U-theil coefficient of the multi-modal storm surge disaster prediction method in coastal areas of the present invention with the number of training rounds.

[0013] For ordinary technicians in this field, other relevant drawings can be obtained based on the above drawings without any creative work. DETAILED DESCRIPTION

[0014] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with specific embodiments.

[0015] The present invention designs a storm surge disaster prediction method for coastal areas based on multi-modality, and the specific steps are as follows: Step 1: Data collection.

[0016] Step 1.1: Collect meteorological environment data of the target coastal area to be evaluated through sensors, including wind speed, sea temperature, sea level, sea level pressure and air humidity with time attributes, to form a meteorological environment data set represented by a time series. : ; :Wind speed (unit: ); :Sea water temperature (unit: ); : sea level height (unit: m); : Sea level pressure (unit: hPa); : Air humidity (percentage).

[0017] Step 1.2: Use remote sensing technology (satellites, drones, etc.) to obtain three-dimensional geospatial data on the coastline changes, marine environment, and infrastructure of the target coastal area to be assessed, and form a three-dimensional geospatial dataset represented by a time series. : ; : Satellite remote sensing data; : UAV remote sensing data; : Remote sensing data fusion and modeling functions.

[0018] Step 1.3: Obtain the historical occurrence data of storm surges to form a storm surge historical occurrence dataset represented by a time series : ; i represents the historical occurrence data of the i-th storm surge, and N is the total number of historical occurrence data of storm surge; : Timestamp (time when the storm surge occurred); : Storm surge height (unit: m, indicating the height of water level rise).

[0019] Step 2: Based on the data collected in step 1, use a multimodal network to perform long-term prediction and trend fitting of future storm surge conditions.

[0020] Step 2.1: Fit the linear relationship of the data collected in step 1 (meteorological environment data, three-dimensional geographic spatial data, and historical storm surge data) to obtain the linear components related to storm surge prediction ; ; : Indicates at time The linear components related to storm surge prediction (e.g., long-term trend or average change of storm surge, linear part of periodic fluctuation, etc.); : Difference operator, which means to perform differential operation on the original time series. Order difference operation. is the difference order, which aims to make the non-stationary time series stationary to facilitate subsequent modeling and analysis; : The constant term, similar to the intercept, represents the average level or base value of the sequence. In this embodiment, it represents the base value of the long-term average storm surge occurrence in the region; : autoregressive order, is the autoregressive coefficient, Indicates the past Time series of moments The value of Indicates the initial value, representing the original collected data. and It reflects the influence of past storm surges on current storm surges, such as hour, express right The influence coefficient reflects the linear correlation degree between the occurrence of storm surge at the previous moment and the occurrence of storm surge at the current moment; : Moving average order, is the moving average coefficient, It's the past The moving average part is mainly used to deal with irregular components such as noise and short-term fluctuations in the time series, so that the model can better fit the actual observed storm surge historical data. Determines the weight distribution of each error term; : It’s time The white noise represents some uncontrollable and random factors that affect the occurrence of storm surges and cannot be explained by the model.

[0021] Step 2.2: Fit the nonlinear relationship of the data collected in step 1 (meteorological environment data, three-dimensional geographic spatial data, and historical storm surge data) to obtain the nonlinear components related to storm surge prediction. .

[0022] Optimizing the parameters of a time series neural network by simulating humpback whale hunting behavior, assuming is the hyperparameter combination vector of the time series neural network, and the parameters are updated after iteration: ; ; in, , Gradually decrease from 2 to 0, and yes Random vectors within ; : When searching for the best hyperparameters for a time series neural network, it reflects the distance metric between the hyperparameter combination and the optimal hyperparameter set, such as the overall difference between the hyperparameter combination and the optimal combination in terms of the number of neurons, learning uniformity, and batch length. : In the context of hyperparameter optimization, it is the hyperparameter combination vector currently found to make the time series neural network perform best (such as minimum), including parameter values ​​such as the number of hidden layer neurons, learning rate, batch length, etc., which is continuously updated with the algorithm iteration to approach the global optimum; : The time series neural network hyperparameter combination vector currently being evaluated is continuously adjusted and updated during the algorithm search process, and its initial value is randomly set within the hyperparameter value range. Interactive iterative optimization; and : coefficient vector, Controls the step size of the individual approaching the optimal solution, and its value is affected by and Impact, with iteration Reduce the step size from large to small, guiding the search to gradually focus; Mainly used to adjust search randomness, The combined effect affects the randomness of individual position updates, balancing exploration and utilization in hyperparameter search to avoid falling into local optimality, such as reasonably adjusting the search range and step size when searching for the number of neurons.

[0023] The final nonlinear output of time series data is: ; in, and is the activation function (Sigmoid and Tanh), is the hyperparameter combination vector of the time series neural network, At the moment The nonlinear components related to storm surge prediction (e.g. random changes affected by the meteorological environment, nonlinear parts of periodic fluctuations, etc.); Represents at the moment Nonlinear components associated with storm surge prediction.

[0024] Step 2.3: Substitute the linear components of storm surge prediction obtained in step 2.1 into Nonlinear components related to the storm surge prediction obtained in step 2.2 Add together to get new storm surge prediction data : .

[0025] Step 2.4: Substitute the storm surge prediction data obtained in step 2.3 into , converted to storm surge prediction probability through an activation function (such as Sigmoid function): ; : Storm surge prediction probability (i.e. predicted probability of occurrence of storm surge); : Sigmoid activation function, ensure output The value range is .

[0026] Step 3: Combine the three-dimensional geospatial data obtained in step 1 and use supervised learning methods to predict storm surges.

[0027] Step 3.1: Perform random shuffling and data augmentation processing on the three-dimensional geospatial data.

[0028] The main operation process is to crop and compress the point cloud scene into a bird's-eye view grid, further divide the bird's-eye view grid into small units, and achieve data enhancement by shuffling these small units. The shuffled patches are then passed through the backbone network to extract features, allowing the network to learn data features in different combinations and orders, and enhance its understanding of the data and feature extraction capabilities. For example, when detecting small objects such as small reefs on the coast, this shuffled patch method allows the network to better learn its feature representation and avoid learning limitations caused by conventional order.

[0029] The features are then extracted through the backbone network and the original position is restored. Among them, the point cloud scene is a data form that is collected by equipment such as lidar and contains many discrete points to digitally represent real scenes (such as three-dimensional spatial conditions such as coastal terrain and coastal infrastructure). Each point in it has its spatial coordinates and other related attribute information, which is the data basis for subsequent operations. The bird's-eye view grid is a two-dimensional grid form similar to a bird's-eye view from above, which is obtained by converting the three-dimensional point cloud scene through specific projection, division, etc., which is convenient for segmentation, shuffling and other operations. Each grid can correspond to the summary of point cloud information in a certain spatial area. For example, after converting the point cloud data of a coastal area into a bird's-eye view grid, the distribution of terrain, buildings, etc. in different locations can be seen more clearly (in the form of a grid).

[0030] Step 3.2: Use the Jenks Natural Breaks algorithm (automatic segmentation algorithm) to generate dynamic thresholds for the three-dimensional geospatial data processed in step 3.1 to obtain the thresholds for each category. c Three thresholds .

[0031] The principle of the Jenks Natural Breaks algorithm is to design a threshold generation strategy based on a set of confident scenes and their prediction pairs, taking into account the confidence score, objectness score, and consistency threshold (IoU); ; Among them, JNB stands for Jenks Natural Breaks automatic segmentation algorithm, which is used to divide data into several natural groups (i.e. categories); :category c The confidence score set of ; here the category c Refers to the two categories: storm surge occurred and no storm surge occurred; : high confidence threshold; : Intermediate confidence threshold; : Low confidence threshold.

[0032] Consistency threshold (the overlap ratio between the predicted storm surge impact area and the area affected by the actual storm surge in history, which is used to assist in judging the accuracy of the prediction and the degree of fit with the actual situation, and then assist in the generation of the threshold): ; : No. Prediction box With real box The intersection and union ratio of : IoU consistency threshold; : Total number of samples.

[0033] Step 3.3: Three thresholds based on step 3.2 , the pseudo labels of mining objects in unlabeled scenes in 3D geospatial data are generated through hierarchical supervision, and the hierarchical pseudo labels are divided into three categories: high confidence, fuzzy and low confidence.

[0034] The pseudo-label is not an accurate label that is actually manually annotated. It is a temporary identification that roughly divides the category to which it belongs based on the previous threshold, and is divided into categories corresponding to high confidence, fuzzy and low confidence. 1) High confidence category: represents relatively reliable and more trustworthy data category; ; Pseudo labels representing high confidence categories, represents the confidence score of the pseudo label x.

[0035] 2) Fuzzy category: A category divided based on the threshold, between high confidence and low confidence, representing data that is not sure of its accuracy but not completely unreliable; ; Pseudo labels representing ambiguous categories.

[0036] 3) Low confidence category: represents the data category with low credibility. The input data may contain a lot of interference or inaccuracy, which needs to be processed to avoid adverse effects on network training. For example, some sensors collect abnormal data, and remote sensing images are difficult to identify and unreliable due to occlusion and other reasons. ; Pseudo labels representing low confidence classes.

[0037] For pseudo labels of low confidence categories, they are removed as noise points.

[0038] Step 3.4: Train the target loss function.

[0039] Total training loss = supervised learning loss for labeled scenes + Hierarchical supervised learning loss for unlabeled scenes ,Right now: By training the target loss function, we can get the predicted value of supervised learning for the labeled scene. And the predicted values ​​of hierarchical supervised learning for unlabeled scenes .

[0040] The total training loss takes into account the labeled and unlabeled scenarios, and measures the overall degree of deviation between the network training effect and the ideal situation. When conducting machine learning training related to disaster risk assessment of coastal infrastructure, the accuracy of the final disaster risk assessment is improved by continuously optimizing and reducing the total training loss.

[0041] 1) Supervised learning loss for labeled scenes; ; : total sample size; : Number of categories; in the present invention, the number of categories is 2, the first category is the occurrence of storm surge, c=1 represents the first category; the second category is no storm surge, c=2 represents the second category; :sample The true label of category c is one-hot encoded; Representation sample Belongs to category c, Representation sample Does not belong to category c; :sample The predicted probability of class c.

[0042] 2) Hierarchical supervised learning loss for unlabeled scenes: ; : Corresponding to high, fuzzy, and low confidence losses respectively; : The corresponding weight coefficient.

[0043] High confidence loss (the loss function uses square error when the error is small, which is smoother; it uses absolute error when the error is large to reduce the risk of gradient explosion): ; ; : It is an existing loss function for deep learning; :sample Pseudo labels of :sample The predicted value of .

[0044] Fuzzy loss (for classification tasks with fuzzy pseudo-labels): ; : Balance factor, used to adjust the importance of positive and negative samples; : Adjustment factor, used to reduce the influence of easy-to-classify samples and increase the attention of difficult-to-classify samples; : Increase the loss contribution for samples with small prediction probability (difficult to classify).

[0045] Low confidence loss (for denoising tasks with low confidence pseudo labels): ; ; : Confidence threshold. A value less than this value indicates that the predicted value is close to the pseudo-label and can be considered valid. : Indicates discarding the gradient update of the sample or setting its loss to zero to avoid interfering with model training.

[0046] Step 3.5: Output the predicted storm surge probability.

[0047] The predicted value Converted to the probability output of storm surge prediction through an activation function (such as Sigmoid function): ; : Storm surge prediction probability; is the Sigmoid activation function, ensuring that the output value range is .

[0048] Step 4: Weight the predicted probabilities obtained in steps 2 and 3 to obtain the final storm surge forecast value for coastal areas. , to achieve quantitative assessment of disaster risks; ; in, : Weighted value.

[0049] Furthermore, according to the storm surge prediction value , disaster risk classification and decision-making can be carried out: if , risk level = high risk; if , risk level = medium risk; if , risk level = low risk; =Risk threshold (determined based on historical data and experience); =Threshold for high risk; = Low risk threshold.

[0050] Experimental verification: See attached Figure 2 The paper simulates the changing trends of error performances such as mean error (ME), mean absolute error (MAE), root mean square error (RMSE), mean absolute scaled error (MASE), and U-theil coefficient of the multimodal storm surge disaster prediction method for coastal areas proposed in the present invention with the number of training rounds.

[0051] from Figure 2 It can be observed that all error indicators show a downward trend as the number of training rounds increases. This shows that the proposed algorithm is continuously optimized during the training process, the error is gradually reduced, and the prediction ability of the model is improved.

[0052] ME (mean error) decreases the fastest, decreasing rapidly in the first 10 rounds of training, and then tends to be stable, indicating that the deviation of this method converges early. MAE (mean absolute error) and RMSE (root mean square error) decrease relatively evenly, but RMSE decreases more slowly than MAE, indicating that it is still affected by some large error samples (RMSE is more sensitive to large errors). MASE (mean absolute scaled error) decreases more slowly than ME, but also tends to be stable after 30 rounds.

[0053] In the early stage of training (the first 10 rounds), all error values ​​dropped rapidly, indicating that the model learned the main patterns and features in the early stage. In the middle stage of training (10-30 rounds), the downward trend was still obvious, but the rate was reduced, which means that the learning curve of the model began to stabilize. In the late stage of training (30-50 rounds), all error indicators basically stabilized, indicating that the algorithm converged well, and further training had little effect on the error reduction, entering the bottleneck period.

[0054] There are a few fluctuations in the error curve, which is due to the uncertainty factor in the random noise simulation experiment. These fluctuations are small, indicating that the algorithm training process is relatively stable.

[0055] The present invention is described above by way of example. It should be noted that, without departing from the core of the present invention, any simple deformation, modification or other equivalent replacement that can be made by those skilled in the art without inventive effort falls within the protection scope of the present invention.

Claims

1. A storm surge disaster prediction method for coastal areas based on multi-modality, characterized in that: The following steps are involved: Step 1: Collect meteorological environment data, three-dimensional geographic spatial data and historical storm surge data of the target coastal area to be evaluated; Step 2: Based on the data collected in step 1, use a multimodal network to perform long-term prediction and trend fitting of future storm surge conditions; Step 2.1: Fit the linear relationship of the data collected in step 1 to obtain the linear component related to storm surge prediction ; Step 2.2: Fit the nonlinear relationship of the data collected in step 1 to obtain the nonlinear components related to storm surge prediction ; Step 2.3: Substitute the linear components of storm surge prediction obtained in step 2.1 into Nonlinear components related to the storm surge prediction obtained in step 2.2 Add together to get new storm surge prediction data ; Step 2.4: Substitute the storm surge prediction data obtained in step 2.3 into , converted into storm surge prediction probability through activation function ; Step 3: Combine the three-dimensional geospatial data obtained in step 1 and use supervised learning methods to predict storm surges; Step 3.1: Perform random shuffling data enhancement processing on the three-dimensional geospatial data; Step 3.2: Use the Jenks Natural Breaks algorithm to generate dynamic thresholds for the three-dimensional geospatial data processed in step 3.1 to obtain the thresholds for each category. Three thresholds ; Step 3.3: Three thresholds based on step 3.2 , the pseudo labels of the mined objects in the unlabeled scenes in the 3D geospatial data are generated by hierarchical supervision, and the hierarchical pseudo labels are divided into three categories: high confidence, fuzzy and low confidence; Step 3.4 Training target loss function, total training loss = supervised learning loss for labeled scenes + Hierarchical supervised learning loss for unlabeled scenes ; By training the target loss function, we can get the predicted value of supervised learning for the labeled scene And the predicted values ​​of hierarchical supervised learning for unlabeled scenes ; Step 3.5: Set the predicted value Converted to storm surge prediction probability through activation function ; Step 4: Substitute the storm surge prediction probability obtained in step 2 And the storm surge prediction probability obtained in step 3 , weighted to obtain the final storm surge prediction value .

2. The method for predicting storm surge disasters in coastal areas based on multi-modality according to claim 1, characterized in that: Meteorological environment data, including: wind speed, sea temperature, sea level, sea level pressure and air humidity data with time attributes, forming a meteorological environment data set represented by a time series; three-dimensional geographic spatial data, including: three-dimensional geographic spatial data of coastline changes, marine environment and infrastructure in target coastal areas with time attributes; historical occurrence data of storm surges, including: storm surge water height data with time attributes.

3. The method for predicting storm surge disasters in coastal areas based on multi-modality according to claim 1, characterized in that: In step 3.3: Pseudo labels for high confidence classes , ; Pseudo-labeling of fuzzy categories , ; Pseudo labels for low confidence classes , ; represents the confidence score of the pseudo label x.

4. The method for predicting storm surge disasters in coastal areas based on multi-modality according to claim 1, characterized in that: In step 3.4: 1) Supervised learning loss for labeled scenes: ； : total sample size; : Number of categories; the number of categories is 2, the first category is the occurrence of storm surge, Represents the first category; The second category is when there is no storm surge. represents the second category; :sample The true label of category c is one-hot encoded; Indicates that sample i belongs to category c, Indicates that sample i does not belong to category c; :sample The predicted probability of category c; 2) Hierarchical supervised learning loss for unlabeled scenes: ； : is the weight coefficient; High confidence loss: ; ； : is the loss function; : Pseudo label of sample i; : The predicted value of sample i: Blur loss: ； : Balance factor, used to adjust the importance of positive and negative samples; : Adjustment factor, used to reduce the influence of easy-to-classify samples and increase the attention of difficult-to-classify samples; : Increase the loss contribution of samples with smaller prediction probabilities; Low confidence loss: ; ； : Confidence threshold. A value less than this value indicates that the predicted value is close to the pseudo label and is considered valid. : Indicates discarding the gradient update of the sample or setting its loss to zero.

5. The method for predicting storm surge disasters in coastal areas based on multi-modality according to claim 1, characterized in that: According to the storm surge prediction , for disaster risk classification and decision-making: if , risk level = high risk; if , risk level = medium risk; if , risk level = low risk; =Threshold for high risk; = Low risk threshold.

Images