Storm surge inundation area grading warning method and system

Abstract

The application discloses a storm surge inundation area grading early warning method and system, and relates to the technical field of data processing. The method comprises the following steps: S1, multi-source data preprocessing: through multi-source heterogeneous data collection, space-time reference unification and quality grading processing, a standardized data set is obtained, wherein the quality grading result is used for weight distribution of subsequent feature extraction; S2, dynamic feature extraction; S3, initial inundation prediction; S4, deviation self-adaptive correction; S5, grading threshold adaptation; S6, grading early warning release and feedback. Through the sub-steps of multi-source heterogeneous data collection, space-time reference unification and quality grading processing, the application realizes effective integration of multi-source data. The unified space-time reference ensures the consistency of data, the three-level quality grading standard clearly defines the use priority of data, improves the utilization rate and reliability of data, and lays a solid foundation for subsequent feature extraction and model prediction.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a graded early warning method and system for storm surge inundation areas. Background Technology

[0002] Storm surges, a common natural disaster in coastal areas, are characterized by their suddenness and destructive power, posing a serious threat to the lives and property of coastal residents, economic development, and the ecological environment.

[0003] Traditional methods have weak integration capabilities for multi-source heterogeneous data. Data from different sources (such as radar echo data, tide gauge data, typhoon path data, etc.) often suffer from inconsistent spatiotemporal benchmarks and varying data quality, resulting in low data utilization and difficulty in fully leveraging the synergistic effects of multi-source data. Furthermore, the lack of clear data quality grading standards negatively impacts subsequent feature extraction and model prediction. In the feature extraction and model prediction stages, existing technologies fail to dynamically adjust the weights and time windows of real-time observation features and historical correlation features based on the warning timeline. This leads to insufficient reliance on real-time data in short-term warnings and inadequate utilization of historical data in long-term warnings, affecting prediction accuracy. In addition, the prediction models used lack feature learning capabilities in near-shore sensitive areas and have poor generalization ability to extreme weather cases, resulting in significant deviations between preliminary predictions and actual conditions. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a graded early warning method for storm surge inundation areas, comprising the following steps:

[0005] S1, Multi-source data preprocessing: Through multi-source heterogeneous data collection, spatiotemporal benchmark unification and quality grading, a standardized dataset is obtained, in which the quality grading results are used for weight allocation in subsequent feature extraction.

[0006] S2, Dynamic Feature Extraction: Based on the standardized dataset, extract real-time observation features and historical correlation features, construct a feature time series matrix, and dynamically adjust the input weights and time window lengths of the two types of features according to the warning timeliness.

[0007] S3, Initial Inundation Prediction: The weighted features are input into the pre-trained deep learning early warning model, and the model's multi-scale feature fusion module outputs the preliminary prediction results of the storm surge inundation area. The preliminary prediction results include the spatiotemporal distribution data of the inundation range, depth and duration.

[0008] S4, Deviation Adaptive Correction: Call the historical storm surge case library, calculate the deviation coefficient between the preliminary prediction result and the historical cases through the case similarity matching algorithm, and combine the residual feedback of real-time monitoring data to dynamically correct the preliminary prediction result and obtain the corrected flooding data;

[0009] S5, Graded Threshold Adaptation: Combining the corrected inundation data with the preset regional disaster-bearing body characteristic parameters, the warning thresholds for different sub-regions are determined by fuzzy hierarchical analysis, and the warning levels are divided.

[0010] S6, Tiered Early Warning Issuance and Feedback: Generate differentiated early warning information based on the early warning level, and issue it through a communication method adapted to the receiving object. The issuance process includes threshold judgment logic for communication method adaptation, and tracks the reach rate and response data of early warning information in real time for model iteration and optimization.

[0011] Preferably, step S1 includes the following sub-steps:

[0012] S11, Multi-source heterogeneous data acquisition: Data acquisition categorized by data type, specifically including:

[0013] Real-time observation data: collect radar echo reflectivity factor data (spatial resolution ≤ 1km, temporal resolution ≤ 6 minutes), minute-level coupled data of tide level and current velocity from tide gauge stations, real-time tracking data of typhoon path latitude and longitude and intensity (wind speed, air pressure), and 0.1m resolution submerged boundary images of the nearshore area taken by UAV.

[0014] Historical baseline data: hourly records of storm surge events over the past thirty years (including highest tide level and surge magnitude), 1m precision elevation data of coastal topography (including three-dimensional information of man-made structures such as dikes and roads), and historical disaster loss ledgers (including correlation data between inundation depth and economic losses).

[0015] Environmental data: Collect monthly average sea level change trend data over the past five years, coastal engineering protection level and disaster resistance assessment data (such as seawall design standards and seawall failure risk coefficient), and quarterly remote sensing data of vegetation coverage at 500m resolution (used to assess surface runoff resistance).

[0016] S12, Spatiotemporal reference unification: Convert all collected data to a unified coordinate system (such as WGS-84 coordinate system) and timestamp (UTC time), calibrate the time synchronization error of data from different sources (error ≤30s), and use a weighted fusion algorithm to process data in spatially overlapping areas;

[0017] S13, Quality Grading Processing: The data undergoes a three-level quality assessment, specifically:

[0018] Grade A: Real-time data with data integrity ≥90%, timestamp error ≤30s, and normal sensor calibration status is marked as core input data;

[0019] Grade B: Historical data with a missing rate of 10%-30% but complete key parameters (such as tide level and typhoon intensity) are marked as auxiliary input data after being completed by spatiotemporal interpolation (error ≤5%).

[0020] Grade C: Data that exceeds the effective time limit (satellite remote sensing exceeding 6 hours, UAV imagery exceeding 2 hours) or has a key parameter missing rate >30% is marked as reference constraint data.

[0021] Preferably, the dynamic allocation of input weights and time window settings for the two types of features in step S2 includes:

[0022] S21, Feature Extraction and Matrix Construction: Extract real-time observation features (including radar echo intensity, tidal level change rate, typhoon movement vector, etc.) and historical correlation features (including historical same-period water increase coefficient, inundation pattern of typhoons with similar paths, etc.) from the standardized dataset, and construct a 10×10 dimension feature matrix according to the time series.

[0023] S22, Dynamic weight allocation:

[0024] Short-term warning (0-3 hours): Real-time observation features weight ≥70%, of which radar echo and tide data weights account for no less than 50%; Historical correlation features weight ≤30%, selecting feature data of similar weather types in the same period of the past three years;

[0025] Mid-term warning (3-12 hours): The weights of real-time observation features and historical correlation features are allocated in a 1:1 ratio. The weight of typhoon path data in real-time features is increased to 40%, and the historical features incorporate the water increase pattern data of typhoons with similar paths in the past five years.

[0026] Long-term warning (12-24 hours): Historical correlation feature weight ≥60%, focusing on the water increase process curve data of similar historical typhoons; Real-time observation feature weight ≤40%, mainly based on typhoon intensity and movement speed data;

[0027] S23, Time Window Adjustment: The short-term warning window is set to 1 hour, the medium-term to 3 hours, and the long-term to 6 hours, with a window sliding step of 30 minutes for all.

[0028] Preferably, the training process of the pre-trained deep learning early warning model in step S3 includes the following sub-steps:

[0029] S31, Sample Dataset Construction: Multi-source data of historical storm surge events are used as training samples. Input features include typhoon intensity (maximum wind speed, central pressure), initial tide level (including astronomical tide prediction), topographic parameters (elevation, slope, levee height) and trend features over the past 3 hours. Output labels are the vector boundary of the measured inundation range, inundation depth values ​​and duration of different areas.

[0030] S32, Model Architecture Design: An improved U-Net network architecture is adopted, with an attention gating module added at the encoding end to enhance feature learning of near-shore sensitive areas (such as dam breach points and low-lying residential areas), and feature fusion weights of skip connections are introduced at the decoding end (dynamically adjusted according to the importance of the area).

[0031] S33, Model Training Optimization: Combining the simulation results of numerical models (such as ADCIRC models) as auxiliary supervision signals, the simulation data of extreme weather cases (such as once-in-a-century storm surges) are transferred to the training set through transfer learning, and the Adam optimizer is used for training.

[0032] Preferably, the adaptive correction of deviation in step S4 includes the following sub-steps:

[0033] S41, Historical Case Screening: Extract candidate cases from the historical case database that have a spatial similarity of ≥80% to the current typhoon path (calculated by Hausdorf distance), a pressure gradient difference of ≤5hPa, and occur in the same season. The number of cases screened should be controlled between 5 and 10.

[0034] S42, Optimal Case Matching: Calculate the terrain difference coefficient (focusing on comparing dike height and coastline morphology) and tidal phase difference (matching degree of the rising and falling phases of astronomical tides) between the candidate case and the current prediction scenario, and determine the optimal matching case through the weighted Euclidean distance formula (terrain weight 0.6, tidal weight 0.4);

[0035] S43, Deviation Calculation and Correction: Deviation coefficient = (Measured water increase of the optimal case - Simulated water increase of the model) / Measured water increase of the optimal case. The deviation coefficient is dynamically corrected by combining the 5-minute residual (measured value - predicted value) of the real-time monitoring data. The corrected deviation is used to adjust the flooding depth and range of the preliminary prediction results. The correction range is controlled within ±10%.

[0036] Preferably, the determination of the warning threshold using the fuzzy hierarchical analysis method in step S5 includes the following sub-steps:

[0037] S51, Indicator System Construction: The corrected inundation data (inundation depth, duration) and regional disaster-bearing body characteristic parameters (population density, economic vulnerability, emergency response capability) are used as primary indicators. Each primary indicator is further divided into 3-5 secondary indicators (such as inundation depth divided into interval indicators such as <0.5m, 0.5-1.0m, etc.).

[0038] S52, Weighting and Fuzzy Evaluation: The weights of each indicator are determined using the 1-9 scaling method. A fuzzy judgment matrix is ​​constructed using triangular fuzzy numbers, and the comprehensive weight value of the indicators is calculated. The weight of the disaster-bearing body characteristic parameters accounts for no less than 40%.

[0039] S53, Threshold Interval Division: Based on the comprehensive evaluation results, the warning threshold intervals for different sub-regions are divided using the principle of maximum membership. The threshold difference for the same warning level in different sub-regions is controlled within ±20%.

[0040] Preferably, the threshold judgment logic for the regional disaster-bearing body characteristic parameters in step S5 is as follows:

[0041] S51, construct the disaster-bearing body characteristic baseline vector V0, including baseline population density (500 people / square kilometer), baseline economic vulnerability (10 million yuan / square kilometer), and baseline emergency response capability (3 shelters / 10 square kilometers and average evacuation distance ≤2 kilometers).

[0042] S52, calculate the cosine similarity S between the current disaster-bearing body feature vector V and the baseline vector V0;

[0043] S53, perform sliding window (1-hour window) statistics on the real-time disaster-bearing body data of the current area, and generate characteristic change rate curve R;

[0044] S54, calculate the extreme value D of the first derivative of the rate of change curve R, which reflects the degree of abrupt change in the characteristic parameter;

[0045] S55, set the vulnerability correlation coefficient K = 0.3 × population density weight + 0.4 × economic vulnerability weight + 0.3 × emergency response capacity weight;

[0046] S56. If the cosine similarity S≥0.8, the extreme value of the derivative of the rate of change D≤5% / hour, the vulnerability correlation coefficient K≤0.4, and the population density mutation ΔP<10%, then the current disaster-bearing body is determined to be in a low-sensitivity state, and the basic early warning threshold is adopted; otherwise, proceed with S57.

[0047] S57. If the region is determined to be in a highly sensitive state, the density clustering algorithm is used to divide the region into sub-units. S52-S55 are repeated for each sub-unit to remove abnormally fluctuating sub-units (such as population density mutations caused by temporary large sets). The weighted average of the feature parameters of the remaining sub-units is used as the basis for the final threshold calculation.

[0048] Preferably, the specific rules for classifying the warning levels in step S5 are as follows:

[0049] A three-dimensional hierarchical model of inundation depth, duration, and vulnerability index of disaster-bearing bodies was established and stored in the early warning system database. The triggering conditions for each level of early warning adopted a dynamic threshold verification mechanism with a verification accuracy of ±5% of the measured value. This three-dimensional hierarchical model was constructed in the following way:

[0050] When the vulnerability index of the disaster-bearing body is <0.3, a low-risk early warning sub-model is matched, and the early warning level is set as follows: when the flooding depth is <0.5m and the duration is <3h, a blue warning is triggered; when the flooding depth is ≤0.5m and the duration is <1.0m and the duration is ≤3h and the yellow warning preparatory level is triggered (internal prompt only).

[0051] When 0.3 ≤ vulnerability index of disaster-bearing body < 0.6, the medium-risk early warning sub-model is matched, and the classification strategy is as follows: when the flooding depth is < 0.5m but the duration is ≥ 3h, or when 0.5m ≤ flooding depth < 1.0m and the duration is < 3h, a yellow warning is triggered; when 1.0m ≤ flooding depth < 2.0m and 3h ≤ duration < 6h, an orange warning preparatory level is triggered.

[0052] When the vulnerability index of the disaster-bearing body is ≥0.6, the high-risk early warning enhancement strategy is activated, specifically: when the inundation depth is <1.0m but the duration is ≥6h, or when 1.0m≤inundation depth<2.0m and the duration is ≥3h, an orange warning is triggered; when the inundation depth is ≥2.0m or the duration is ≥12h, or when the vulnerability index is ≥0.8 and the real-time inundation depth is ≥0.5m, a red warning is triggered directly regardless of the duration; when the inundation depth is ≥1.5m and the vulnerability index is ≥0.7, the warning level is automatically upgraded by one level (e.g., the original orange warning is upgraded to a red warning).

[0053] Preferably, the threshold judgment logic for adapting the communication method of the receiving object in step S6 is as follows:

[0054] S61, construct the communication object feature matrix M, including object type (general public / flood control command department / sea operation vessel), historical response time (seconds), equipment coverage (%), and information complexity requirements (low / medium / high).

[0055] S62, calculate the matching degree P between different communication methods (SMS / electronic screen / encrypted data packet / BeiDou voice, etc.) and the object feature matrix. The matching degree for the general public is: P = 0.4 × response timeliness compliance rate + 0.3 × equipment compatibility rate + 0.3 × information carrying adaptability. The matching degree for ships at sea is: P = 0.6 × response timeliness compliance rate + 0.4 × equipment compatibility rate.

[0056] S63, set the communication priority threshold T, where T≥0.85 for red warning, T≥0.75 for orange warning, T≥0.65 for yellow warning, and T≥0.55 for blue warning;

[0057] S64. If the matching degree P of a certain communication method is greater than or equal to the priority threshold T of the corresponding warning level, then it is included in the candidate communication method set.

[0058] S65 performs a secondary screening of the candidate communication method set based on real-time channel quality (packet loss rate ≤5% is excellent, 5%-10% is medium, and >10% is poor), retaining communication methods with excellent or medium channel quality. When the real-time channel quality >10% packet loss rate, it automatically switches to the backup communication channel. The switching priority is: Beidou short message > maritime satellite > 5G emergency base station.

[0059] S66. If there is only one communication method after filtering, then use that method to publish directly; if there are multiple methods, sort them from high to low according to the matching degree P, and select the first two for collaborative publishing.

[0060] A graded early warning system for storm surge inundation areas, used to implement a graded early warning method for storm surge inundation areas, the system comprising:

[0061] Data acquisition and preprocessing module: used to execute step S1, including multi-source data acquisition unit, spatiotemporal benchmark unification unit and quality grading unit. Multi-source data acquisition unit collects real-time observation data, historical basic data and environmental related data according to data type classification. Spatiotemporal benchmark unification unit converts the collected data to a unified coordinate system and timestamp and processes spatially overlapping data. Quality grading unit performs three-level quality assessment on the data.

[0062] Feature extraction module: used to execute step S2, including feature extraction unit, matrix construction unit, weight allocation unit and time window adjustment unit. The feature extraction unit extracts real-time observation features and historical correlation features from the standardized dataset. The matrix construction unit constructs a feature time series matrix according to the time series. The weight allocation unit dynamically adjusts the input weights of the two types of features according to the warning timeliness. The time window adjustment unit sets the time window and sliding step size corresponding to different warning timeliness.

[0063] Initial prediction module: used to execute step S3, including model storage unit and prediction execution unit. The model storage unit stores the pre-trained deep learning early warning model, which adopts an improved U-Net network architecture and has been trained and optimized by sample dataset. The prediction execution unit inputs the weighted features into the model and outputs the preliminary prediction results through the multi-scale feature fusion module.

[0064] Deviation correction module: used to execute step S4, including case library unit, similarity calculation unit and deviation correction unit. The case library unit stores historical storm surge case library. The similarity calculation unit filters candidate cases and determines the optimal matching case through case similarity matching algorithm. The deviation correction unit calculates the deviation coefficient and combines real-time monitoring data residual feedback to dynamically correct the preliminary prediction results.

[0065] The graded threshold module is used to execute step S5. It includes a parameter storage unit, a threshold calculation unit, and a grade division unit. The parameter storage unit stores the characteristic parameters of the disaster-bearing body in the region. The threshold calculation unit determines the warning threshold of different sub-regions by combining the fuzzy hierarchical analysis method with the corrected flooding data. The grade division unit divides the warning level according to the three-dimensional graded model and includes the threshold judgment logic of the characteristic parameters of the disaster-bearing body in the region.

[0066] Early warning release and feedback module: used to execute step S6, including information generation unit, communication adaptation unit and feedback tracking unit. The information generation unit generates differentiated early warning information according to the early warning level. The communication adaptation unit releases early warning information by adapting the communication method of the receiving object and includes threshold judgment logic for communication method adaptation. The feedback tracking unit tracks the reach rate and response data of early warning information in real time for model iterative optimization.

[0067] This invention discloses a graded early warning method and system for storm surge inundation areas, which has the following beneficial effects:

[0068] (1) This invention achieves effective integration of multi-source data through sub-steps of multi-source heterogeneous data acquisition, spatiotemporal benchmark unification, and quality grading processing. The unified spatiotemporal benchmark ensures data consistency, and the three-level quality grading standard clarifies the data usage priority, improving data utilization and reliability, and laying a solid foundation for subsequent feature extraction and model prediction;

[0069] (2) This invention dynamically adjusts the input weights and time window lengths of real-time observation features and historical correlation features based on the warning timeliness through the dynamic feature extraction stage, making short-term warnings more reliant on real-time data and long-term warnings more focused on historical data, thus enhancing the pertinence of feature extraction. The pre-trained deep learning warning model adopts an improved U-Net network architecture, adds an attention gating module, strengthens feature learning for near-shore sensitive areas, and, combined with numerical model simulation results and transfer learning, improves the model's generalization ability under extreme weather cases, effectively improving the accuracy of the initial prediction results;

[0070] (3) This invention combines the corrected inundation data with the regional disaster-bearing body characteristic parameters through a graded threshold adaptation process, and determines the warning threshold for different sub-regions and classifies the warning levels through fuzzy hierarchical analysis. The construction of the three-dimensional graded model takes into account factors such as inundation depth, duration and vulnerability index of disaster-bearing bodies, making the classification of warning levels more scientific and reasonable, and able to accurately reflect the risk status of different regions. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0072] Figure 1 This is a schematic block diagram of the graded early warning method of the present invention;

[0073] Figure 2 This is a schematic block diagram of the graded early warning system of the present invention. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0075] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0076] This invention provides, for example Figure 1The illustrated graded early warning method for storm surge inundation areas includes the following steps: S1, Multi-source data preprocessing: A standardized dataset is obtained through multi-source heterogeneous data acquisition, spatiotemporal benchmark unification, and quality grading. The quality grading results are used for weight allocation in subsequent feature extraction; S2, Dynamic feature extraction: Based on the standardized dataset, real-time observation features and historical correlation features are extracted to construct a feature time series matrix. The input weights and time window lengths of the two types of features are dynamically adjusted according to the early warning timeliness; S3, Initial inundation prediction: The weighted features are input into a pre-trained deep learning early warning model. The model's multi-scale feature fusion module outputs preliminary prediction results for the storm surge inundation area. The preliminary prediction results include spatiotemporal distribution data of inundation range, depth, and duration; S4, Adaptive bias correction: A historical storm surge case library is called. The algorithm calculates the deviation coefficient between the preliminary prediction results and historical cases using a case similarity matching algorithm. Combined with residual feedback from real-time monitoring data, the preliminary prediction results are dynamically corrected to obtain the corrected flooding data. S5, Tiered Threshold Adaptation: Combining the corrected flooding data with preset regional disaster-bearing body characteristic parameters, fuzzy hierarchical analysis is used to determine the warning thresholds for different sub-regions, classifying warning levels. S6, Tiered Warning Issuance and Feedback: Differentiated warning information is generated based on the warning level and issued through communication methods adapted to the receiving objects. The issuance process includes threshold judgment logic for communication method adaptation and real-time tracking of the reach rate and response data of the warning information for model iteration and optimization. Multi-source data preprocessing solves the problems of difficulty in integrating heterogeneous multi-source data, inconsistent spatiotemporal benchmarks, and uneven data quality affecting subsequent analysis in traditional technologies. It achieves the acquisition of standardized datasets, providing a reliable data foundation for accurate feature extraction. Dynamic feature extraction overcomes the problem of insufficient warning accuracy caused by the inability to flexibly adjust feature weights and time windows according to the warning timeliness in previous warning systems, constructing a reasonable feature time series matrix. This study utilizes a pre-trained deep learning-based early warning model for initial inundation prediction, improving upon the poor accuracy of traditional prediction models in forecasting storm surge inundation areas. The model outputs preliminary predictions including the spatiotemporal distribution of inundation range, depth, and duration. Adaptive bias correction addresses the issue of significant discrepancies between preliminary predictions and actual conditions, which are difficult to effectively correct. Corrected inundation data is obtained through historical case comparisons and real-time monitoring feedback. Graded threshold adaptation addresses the problem of unreasonable early warning threshold settings that fail to fully consider regional characteristics, enabling the determination of early warning thresholds and the classification of early warning levels based on regional disaster-bearing body characteristic parameters. Graded early warning dissemination and feedback address the issues of untargeted early warning information dissemination, low reach and response rates, and difficulties in iterative model optimization. Information is disseminated through adapted communication methods, and feedback data is tracked for model improvement.

[0077] Step S1 includes the following sub-steps: S11, Multi-source heterogeneous data acquisition: Data is collected according to data type, specifically including: Real-time observation data: Radar echo reflectivity factor data (spatial resolution ≤ 1km, temporal resolution ≤ 6 minutes), minute-level tide level and current velocity coupling data from tide gauge stations, real-time tracking data of typhoon path latitude and longitude and intensity (wind speed, air pressure), and near-shore 0.1m resolution inundation boundary images taken by UAVs; Historical basic data: Hourly records of storm surge rise processes over the past thirty years (including highest tide level and rise amplitude), 1m precision elevation data of coastal topography (including three-dimensional information of man-made structures such as dikes and roads), and historical disaster loss ledgers (including correlation data between inundation depth and economic losses); Environmental correlation data: Monthly average sea level change trend data over the past five years, coastal engineering protection level and disaster resistance capacity assessment data (such as seawall design standards and dike breach risk coefficients), and quarterly remote sensing data of vegetation coverage at 500m resolution (used to assess surface runoff resistance); S12, Spatiotemporal reference system 1. All collected data are converted to a unified coordinate system (e.g., WGS-84 coordinate system) and timestamp (UTC time). Time synchronization errors of data from different sources are calibrated (error ≤ 30s). Data in spatially overlapping areas are processed using a weighted fusion algorithm. S13. Quality grading: Data is assessed in three levels: A-level: Real-time data with ≥90% data integrity, timestamp error ≤ 30s, and normal sensor calibration status is marked as core input data; B-level: Historical data with a missing rate of 10%-30% but complete key parameters (e.g., tide level, typhoon intensity) is completed through spatiotemporal interpolation (error ≤ 5%) and marked as auxiliary input data; C-level: Data exceeding the effective timeframe (satellite remote sensing exceeding 6 hours, UAV imagery exceeding 2 hours) or with a key parameter missing rate > 30% is marked as reference constraint data. Through multi-source heterogeneous data acquisition, the problems of incomplete data collection and unclear classification are solved. Real-time observation, historical baseline, and environmental correlation data are comprehensively collected, providing rich data sources for subsequent analysis. Unified spatiotemporal references resolve the issue of inconsistent spatiotemporal references and the inability to conduct collaborative analysis of data from different sources. All data is converted to a unified coordinate system and timestamps, and errors are calibrated and spatially overlapping data is processed. Quality grading addresses the problem of unclear data quality standards and the resulting inefficiency. A three-level quality assessment clarifies core inputs, auxiliary inputs, and reference constraints, improving data utilization efficiency.

[0078] Specifically, step S2 involves dynamically allocating the input weights and setting the time window for the two types of features, including: S21, Feature Extraction and Matrix Construction: Extracting real-time observation features (including radar echo intensity, tidal level change rate, typhoon movement vector, etc.) and historical correlation features (including historical concurrent water level increase coefficient, inundation patterns of typhoons with similar paths, etc.) from the standardized dataset, and constructing a 10×10 dimension feature matrix according to the time series; S22, Dynamic Weight Allocation: Short-term warning (0-3 hours): Real-time observation feature weight ≥70%, of which radar echo and tidal level data weight accounts for no less than 50%; Historical correlation feature weight ≤30%, selecting feature data of similar weather types from the same period in the past three years; Medium-term warning (3-12 hours): Real-time observation features Historical correlation features are weighted at a 1:1 ratio. In real-time features, the weight of typhoon path data is increased to 40%, and historical features incorporate water increase patterns from typhoons with similar paths over the past five years. For long-term warnings (12-24 hours), historical correlation features have a weight ≥60%, primarily selecting water increase curves from similar historical typhoons. Real-time observation features have a weight ≤40%, mainly focusing on typhoon intensity and movement speed data. S23, Time Window Adjustment: Short-term warning windows are set to 1 hour, medium-term to 3 hours, and long-term to 6 hours, with a 30-minute sliding step for each. Feature extraction and matrix construction address the issues of incomplete and unsystematic feature extraction, comprehensively extracting real-time observation and historical correlation features from standardized datasets and constructing a feature matrix. Dynamic weight allocation solves the problem of unreasonable weight distribution for features with different timeframes during the warning process, rationally allocating the weights of real-time observation and historical correlation features according to the different needs of short-term, medium-term, and long-term warnings. Time window adjustment addresses the problem of time window settings failing to adapt to changes in warning timeframes, setting reasonable time windows and sliding steps according to different warning timeframes, making warnings more aligned with actual conditions.

[0079] The training process of the pre-trained deep learning early warning model in step S3 includes the following sub-steps: S31, Sample dataset construction: Multi-source data of historical storm surge events are used as training samples. Input features include typhoon intensity (maximum wind speed, central pressure), initial tide level (including astronomical tide prediction), topographic parameters (elevation, slope, levee height), and trend features over the past 3 hours. Output labels are the vector boundary of the measured inundation range, inundation depth values ​​and duration in different areas; S32, Model architecture design: An improved U-Net network architecture is adopted (the improved U-Net network retains the traditional "encoder-decoder" symmetrical structure, with the left encoder responsible for feature extraction and the right decoder...). The system is responsible for feature reconstruction, fusing features of the same scale through skip connections. An attention gating module is added to the encoder to enhance feature learning for sensitive near-shore areas (such as dikes and low-lying areas) through weight adjustment. The decoder introduces a dynamic weight allocation mechanism in the skip connections, adjusting the fusion weights based on regional importance to make the features more aligned with the prediction needs of different regions. The overall architecture improves the prediction accuracy for key areas of storm surge inundation, adapting to the requirements of tiered early warning systems. An attention gating module is added to the encoder (this module is used in the improved U-Net network encoder to enhance feature learning for sensitive near-shore areas. It receives the feature map of the current convolutional block in the encoder and the feature map of the corresponding layer in the decoder). The feature map is mapped to the same channel dimension through a 1×1 convolution and then processed by an activation function to generate an attention weight map. In the weight map, the weight values ​​of the corresponding positions in near-shore sensitive areas (such as dikes and low-lying areas) are higher, while the weight values ​​of non-sensitive areas are lower. Multiplying this weight map element-wise with the current feature map at the encoder can enhance the expression of features in sensitive areas, suppress features in irrelevant areas, and make the network more focused on feature extraction of key areas for early warning, thereby improving the predictive accuracy. This strengthens the feature learning of near-shore sensitive areas (such as dike breach points and low-lying residential areas), and introduces feature fusion weights of skip connections at the decoder (dynamically adjusted according to the importance of the area); S33, Model training optimization: combining numerical patterns Simulation results (such as those from the ADCIRC model) are used as auxiliary supervision signals. Through transfer learning, simulated data from extreme weather events (such as once-in-a-century storm surges) are transferred to the training set. The Adam optimizer is employed (used in the training process of the improved U-Net network model; its role is to accelerate model convergence and improve training stability by dynamically adjusting the update step size of model parameters. This optimizer combines the advantages of momentum gradient descent and adaptive learning rate, recording the first-order moment estimate (momentum) and second-order moment estimate (the basis for adaptive learning rate adjustment) of parameter updates, and correcting these two estimates through a bias correction mechanism, thereby dynamically determining the update magnitude of each parameter).In this invention, the Adam optimizer is used to train the model, enabling it to converge to a better state more quickly, reducing oscillations during training, and ensuring greater stability and efficiency in learning features related to storm surge inundation areas, thus improving the model's predictive performance. The training sample dataset construction addresses the issues of unrepresentative training samples and unreasonable feature selection, using historical storm surge multi-source data as samples and selecting appropriate input features and output labels to provide high-quality data for model training. The model architecture design addresses the problems of insufficient feature learning and unreasonable feature fusion in traditional models for nearshore sensitive areas. An improved U-Net network architecture is adopted, adding an attention gating module to enhance feature learning in nearshore sensitive areas and dynamically adjusting the feature fusion weights of skip connections at the decoding end. Model training optimization addresses the problems of poor generalization ability and difficulty in handling extreme weather cases. Numerical model simulation results are used as auxiliary supervision signals, and extreme weather case data are transferred to the training set through transfer learning. The Adam optimizer is used to improve training effectiveness.

[0080] The adaptive correction of deviation in step S4 includes the following sub-steps: S41, Historical case screening: Extracting cases from the historical case database with a spatial similarity ≥80% (calculated using Hausdorff distance, which measures the similarity between two point sets and is used in this invention to calculate the spatial similarity between the current typhoon path and historical typhoon paths. The calculation method is as follows: For two point sets A and B, first calculate the distance from each point in A to the nearest point in B, and take the maximum value of these distances, denoted as h(A, B); then calculate the distance from each point in B to the nearest point in A, and take the maximum value, denoted as h(B, A); the Hausdorff distance is the larger of h(A, B) and h(B, A). In step S41 of this invention, the spatial similarity between the current typhoon path and historical typhoon paths is calculated using Hausdorff distance, thereby screening out candidate cases with similar paths, providing reliable historical case references for subsequent deviation correction, and improving the accuracy of similar case matching.) and pressure gradient difference ≤5h. Candidate cases with the same occurrence season are selected, with the number controlled to 5-10; S42, Optimal case matching: Calculate the terrain difference coefficient (focusing on comparing dike height and coastline morphology) and tidal phase difference (matching degree of astronomical tide rise and fall stages) between the candidate cases and the current prediction scenario, and determine the optimal matching case through the weighted Euclidean distance formula (terrain weight 0.6, tidal weight 0.4); S43, Deviation calculation and correction: Deviation coefficient = (measured water increase of the optimal case - simulated water increase of the model) / measured water increase of the optimal case, and dynamically correct the deviation coefficient by combining the 5-minute residual (measured value - predicted value) of real-time monitoring data. The corrected deviation is used to adjust the inundation depth and range of the preliminary prediction results, and the correction range is controlled within ±10%. Through historical case screening, the problem of accurately screening similar cases from a large number of historical cases is solved. By setting strict screening conditions, such as spatial similarity of typhoon paths, pressure gradient differences and occurrence seasons, a suitable number of candidate cases are selected. Optimal case matching addresses the difficulty of determining the best-matching case when considering only a single factor. It comprehensively considers factors such as terrain difference coefficient and tidal phase difference, and determines the optimal matching case using a weighted Euclidean distance formula. Deviation calculation and correction solves the problems of inaccurate deviation coefficient calculation and untimely correction. It dynamically corrects the deviation coefficient by combining real-time monitoring data residuals, accurately adjusting the inundation depth and range of the preliminary prediction results, and controlling the correction magnitude to ensure the stability of the results.

[0081] The step S5, which uses fuzzy hierarchical analysis to determine the early warning threshold, includes the following sub-steps: S51, Indicator system construction: The corrected inundation data (inundation depth, duration) and regional disaster-bearing body characteristic parameters (population density, economic vulnerability, emergency response capability) are used as primary indicators. Each primary indicator has 3-5 secondary indicators (e.g., inundation depth is subdivided into <0.5m, 0.5-1.0m, etc.); S52, Weight assignment and fuzzy evaluation: The 1-9 scaling method is used to determine the weight of each indicator. A fuzzy judgment matrix is ​​constructed using triangular fuzzy numbers, and the comprehensive weight value of the indicators is calculated. The weight of the disaster-bearing body characteristic parameters accounts for no less than 40%; S53, Threshold interval division: Based on the comprehensive evaluation results, the early warning threshold intervals of different sub-regions are divided using the maximum membership principle. The threshold difference of the same early warning level in different sub-regions is controlled within ±20%. Through the construction of the indicator system, the problems of incomplete indicators and unclear hierarchy in determining the early warning threshold are solved. The corrected inundation data and regional disaster-bearing body characteristic parameters are used as primary indicators, and reasonable secondary indicators are set up to form a complete indicator system. Weight assignment and fuzzy evaluation address the issues of strong subjectivity and unscientific evaluation in determining indicator weights. A 1-9 scale method is used to determine weights, and a fuzzy judgment matrix is ​​constructed using triangular fuzzy numbers to scientifically calculate the comprehensive weight value of indicators while ensuring a reasonable weight ratio for disaster-bearing body characteristic parameters. Threshold interval division resolves the problems of unreasonable early warning threshold interval division and insufficient reflection of regional differences. Based on the comprehensive evaluation results, the principle of maximum membership is used to divide early warning threshold intervals for different sub-regions, and the threshold differences for the same early warning level in different sub-regions are controlled.

[0082] Specifically, the threshold judgment logic for the regional disaster-bearing body characteristic parameters in step S5 is as follows: S51, construct the disaster-bearing body characteristic baseline vector V0, including baseline population density (500 people / square kilometer), baseline economic vulnerability (10 million yuan / square kilometer), and baseline emergency response capacity (3 shelters / 10 square kilometers and average evacuation distance ≤ 2 kilometers); S52, calculate the cosine similarity S between the current regional disaster-bearing body characteristic vector V and the baseline vector V0; S53, perform sliding window (1-hour window) statistics on the real-time disaster-bearing body data of the current region to generate the characteristic change rate curve R; 54. Calculate the extreme value D of the first derivative of the rate of change curve R, reflecting the degree of abrupt change in the characteristic parameters; S55. Set the vulnerability correlation coefficient K = 0.3 × population density weight + 0.4 × economic vulnerability weight + 0.3 × emergency response capacity weight; S56. If the cosine similarity S ≥ 0.8, the extreme value D of the rate of change derivative ≤ 5% / hour, the vulnerability correlation coefficient K ≤ 0.4, and the population density abrupt change ΔP < 10%, then the current disaster-bearing body is determined to be in a low-sensitivity state, and the basic early warning threshold is adopted; otherwise, proceed to S57; S57. If it is determined to be in a high-sensitivity state, density clustering is adopted. The density clustering algorithm is used to group data based on the density of data points. Its core idea is to group closely connected points into the same cluster, making it suitable for identifying cluster structures with complex shapes. The main process is as follows: Core points are identified by setting a radius and a minimum number of points; then, all points within the core point and its radius are grouped into the same cluster, iterating until all connectable points are assigned to their respective clusters. The remaining unassigned points are considered noise points. In step S57 of this invention, the density clustering algorithm is used to divide the region into sub-units, which can group regions with similar features into the same sub-unit. The system uses a unit-based approach to facilitate targeted analysis of disaster-bearing body characteristic parameters. It eliminates sub-units with abnormal fluctuations caused by temporary large gatherings, ensuring the reliability of subsequent threshold calculations. The region is divided into sub-units, and steps S52-S55 are repeated for each sub-unit to eliminate abnormally fluctuating sub-units (such as sudden population density changes caused by temporary large gatherings). The weighted average of the characteristic parameters of the remaining sub-units is used as the final threshold calculation basis. By constructing a disaster-bearing body characteristic benchmark vector, the system addresses the lack of a benchmark for judging the characteristic state of disaster-bearing bodies, clarifying benchmark indicators such as population density, economic vulnerability, and emergency response capabilities. Cosine similarity calculation addresses the difficulty in measuring the degree of difference between the current region's disaster-bearing body characteristics and the benchmark, quantifying the difference through cosine similarity. Sliding window statistics and derivative extrema calculation address the inability to effectively monitor the changing trends and abrupt changes of disaster-bearing body characteristic parameters, generating characteristic change rate curves and calculating first-order derivative extrema to reflect abrupt changes. A vulnerability correlation coefficient is established to address the inability to comprehensively measure the vulnerability of disaster-bearing bodies, comprehensively calculating the vulnerability correlation coefficient through reasonable weight allocation.The threshold judgment logic solves the problem of not being able to accurately determine the characteristic state of the disaster-bearing body and reasonably determine the basis for early warning thresholds. It determines the characteristic state of the disaster-bearing body based on a series of conditions, and uses a reasonable algorithm to process highly sensitive states and determine the basis for the final threshold calculation.

[0083] The specific rules for classifying the warning levels in step S5 are as follows: A three-dimensional classification model of inundation depth, duration, and vulnerability index of the disaster-bearing body is established and stored in the warning system database. The triggering conditions for each warning level adopt a dynamic threshold verification mechanism with a verification accuracy of ±5% of the measured value. This three-dimensional classification model is constructed as follows: When the vulnerability index of the disaster-bearing body is <0.3, a low-risk warning sub-model is matched, and the warning level is set as follows: a blue warning is triggered when the inundation depth is <0.5m and the duration is <3h; a yellow warning preparatory level (internal notification only) is triggered when 0.5m ≤ inundation depth <1.0m and 3h ≤ duration <6h; when 0.3 ≤ vulnerability index of the disaster-bearing body <0.6, a medium-risk warning sub-model is matched, and the classification strategy is as follows: a yellow warning is triggered when the inundation depth is <0.5m but the duration is ≥3h, or when 0.5m ≤ inundation depth <1.0m and the duration is <3h. When the inundation depth is between 1.0m and 2.0m and the duration is between 3h and 6h, an orange alert (preparatory level) is triggered. When the vulnerability index of the disaster-bearing body is ≥0.6, a high-risk alert enhancement strategy is activated, specifically: when the inundation depth is <1.0m but the duration is ≥6h, or when the inundation depth is between 1.0m and 2.0m and the duration is ≥3h, an orange alert is triggered; when the inundation depth is ≥2.0m or the duration is ≥12h, or when the vulnerability index is ≥0.8 and the real-time inundation depth is ≥0.5m, a red alert is triggered regardless of the duration; when the inundation depth is ≥1.5m and the vulnerability index is ≥0.7, the alert level is automatically upgraded by one level (e.g., the original orange alert is upgraded to a red alert). By establishing a three-dimensional hierarchical model, the problem of a single standard for classifying alert levels and the inability to comprehensively consider multiple factors is solved. A comprehensive hierarchical model is constructed by inundation depth, duration, and the vulnerability index of the disaster-bearing body. The dynamic threshold verification mechanism solves the problem of inflexible alert triggering conditions and inability to adapt to actual changes. Dynamic threshold verification ensures verification accuracy and makes alert triggering more closely match the actual situation. Different risk warning sub-models and strategies solve the problem of inaccurate warning strategies for different risk areas. Different risk levels are divided according to the vulnerability index of disaster-bearing bodies, and corresponding warning sub-models and hierarchical strategies are matched to improve the targeting of warnings.

[0084] Specifically, the threshold judgment logic for adapting the communication method of the receiving object in step S6 is as follows: S61, construct the communication object feature matrix M, including object type (general public / flood control command department / sea-based vessels), historical response time (seconds), equipment coverage (%), and information complexity requirements (low / medium / high); S62, calculate the matching degree P between different communication methods (SMS / electronic screen / encrypted data packet / BeiDou voice, etc.) and the object feature matrix. The matching degree for the general public is: P = 0.4 × response time compliance rate + 0.3 × equipment compatibility rate + 0.3 × information carrying adaptability; the matching degree for sea-based vessels is: P = 0.6 × response time compliance rate + 0.4 × equipment compatibility rate; S63, set the communication priority threshold T, where T ≥ 0.85 for red alerts, T ≥ 0.75 for orange alerts, T ≥ 0.65 for yellow alerts, and T ≥ 0.55 for blue alerts; S64, if If the matching degree P of a certain communication method is greater than or equal to the priority threshold T of the corresponding warning level, then it is included in the candidate communication method set; S65, the candidate communication method set is further filtered according to real-time channel quality (packet loss rate ≤5% is excellent, 5%-10% is medium, and >10% is poor), and communication methods with excellent or medium channel quality are retained. When the real-time channel quality >10% packet loss rate, it is automatically switched to the backup communication channel, with the switching priority as follows: Beidou short message > maritime satellite > 5G emergency base station; S66, if there is only one communication method after filtering, then that method is used directly for publication; if there are multiple methods, they are sorted from high to low according to the matching degree P, and the first two are selected for collaborative publication (such as SMS + community electronic screen, encrypted data packet + emergency command system pop-up). By constructing a communication object feature matrix, the problem of incomplete understanding of the characteristics of communication objects is solved, and factors such as object type, historical response timeliness, equipment coverage and information complexity requirements are fully considered. The system calculates communication method matching degree, addressing the problem of inaccurately assessing the compatibility between communication methods and target audiences. It calculates matching degree by setting appropriate weights for response timeliness compliance, equipment compatibility, and information carrying capacity adaptability based on different target audience types. Setting communication priority thresholds resolves the issue of unclear priority in early warning information dissemination by setting corresponding communication priority thresholds according to different early warning levels. Candidate communication method filtering addresses the problems of unreasonable communication method selection and unreliable channel quality by filtering based on matching degree and real-time channel quality, ensuring the effectiveness of information dissemination. A communication method dissemination strategy addresses the problem of inappropriate selection when multiple communication methods are available by rationally determining single or collaborative dissemination methods based on the filtering results, improving the effectiveness of early warning information dissemination.

[0085] This invention provides, for example Figure 2The illustrated storm surge inundation area graded early warning system includes: a data acquisition and preprocessing module for executing step S1, comprising a multi-source data acquisition unit, a spatiotemporal benchmark unification unit, and a quality grading unit. The multi-source data acquisition unit collects real-time observation data, historical baseline data, and environmentally related data according to data type classification. The spatiotemporal benchmark unification unit converts the collected data to a unified coordinate system and timestamp and processes spatially overlapping data. The quality grading unit performs a three-level quality assessment of the data. A feature extraction module for executing step S2 includes a feature extraction unit, a matrix construction unit, a weight allocation unit, and a time window adjustment unit. The feature extraction unit extracts real-time observation data from the standardized dataset. The features are correlated with historical features. The matrix construction unit constructs a feature time series matrix according to the time series. The weight allocation unit dynamically adjusts the input weights of the two types of features according to the warning timeliness. The time window adjustment unit sets the time window and sliding step size corresponding to different warning timeliness. The initial prediction module is used to execute step S3 and includes a model storage unit and a prediction execution unit. The model storage unit stores the pre-trained deep learning warning model, which adopts an improved U-Net network architecture and has been trained and optimized with sample datasets. The prediction execution unit inputs the weighted features into the model and outputs the preliminary prediction results through the multi-scale feature fusion module. The bias correction module is used to execute step S4 and includes a case library unit. The system comprises a similarity calculation unit and a deviation correction unit. The case library unit stores historical storm surge case studies. The similarity calculation unit filters candidate cases using a case similarity matching algorithm and determines the optimal matching case. The deviation correction unit calculates the deviation coefficient and dynamically corrects the preliminary prediction results using feedback from real-time monitoring data residuals. The tiered threshold module, used to execute step S5, includes a parameter storage unit, a threshold calculation unit, and a tiered classification unit. The parameter storage unit stores regional disaster-bearing body characteristic parameters. The threshold calculation unit determines the warning thresholds for different sub-regions using fuzzy hierarchical analysis combined with corrected inundation data. The tiered classification unit classifies warning levels based on a three-dimensional tiered model and includes regional disaster-bearing body characteristics. The threshold judgment logic for parameters; the early warning release and feedback module: used to execute step S6, including an information generation unit, a communication adaptation unit, and a feedback tracking unit. The information generation unit generates differentiated early warning information according to the early warning level. The communication adaptation unit releases early warning information by adapting the communication method of the receiving object and includes threshold judgment logic for communication method adaptation. The feedback tracking unit tracks the reach rate and response data of early warning information in real time for model iterative optimization. Among them, the data acquisition and preprocessing module, through the collaborative work of the multi-source data acquisition unit, the spatiotemporal benchmark unification unit, and the quality grading unit, solves the problems of data acquisition, integration, and quality assessment, and realizes the acquisition of high-quality standardized datasets.The feature extraction module, utilizing feature extraction, matrix construction, weight allocation, and time window adjustment units, solves the problems of unreasonable feature extraction, matrix construction, and weight and time window settings, constructing a feature time series matrix that meets the timeliness requirements of early warning. The initial prediction module, with the help of model storage and prediction execution units, solves the problems of model storage and prediction execution, outputting reliable preliminary prediction results using a high-quality model. The deviation correction module, through case library, similarity calculation, and deviation correction units, solves the problems of historical case selection, matching, and deviation calculation correction, improving the accuracy of prediction results. The tiered threshold module, relying on parameter storage, threshold calculation, and tier classification units, solves the problems of parameter storage, threshold calculation, and tier classification, achieving scientific and reasonable early warning tiering. The early warning release and feedback module, with information generation, communication adaptation, and feedback tracking units, solves the problems of early warning information generation, release, and feedback tracking, improving the effectiveness of early warning information release and model iterative optimization capabilities. Through the coordinated operation of these modules, the entire system comprehensively solves a series of problems in the tiered early warning process for storm surge inundation areas, improving the accuracy, relevance, and effectiveness of early warnings.

[0086] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for graded warning of storm surge inundation areas, characterized by, Comprise the following steps: S1, multi-source data preprocessing: through multi-source heterogeneous data collection, space-time reference unification and quality grading processing, obtain standardized data set, wherein the quality grading result is used for weight distribution of subsequent feature extraction; S2, dynamic feature extraction: based on the standardized data set, real-time observation features and historical correlation features are extracted, a feature time series matrix is constructed, and the input weight and time window length of the two types of features are dynamically adjusted according to the early warning timeliness, and the input weight and time window setting of the two types of features are dynamically allocated, including: S21, feature extraction and matrix construction: real-time observation features and historical correlation features are extracted from the standardized data set, and a 10*10 dimensional feature matrix is constructed according to time series; S22, weight dynamic distribution: Short-term warning: real-time observation feature weight≥70%, wherein radar echo and tide level data weight ratio is not less than 50%; historical correlation feature weight≤30%, selecting similar weather type feature data of the same period in the past three years; Medium-term warning: real-time observation feature and historical correlation feature weight is allocated according to 1:1, and typhoon path data weight in real-time feature is increased to 40%, and historical feature introduces the water level increase rule data of similar path typhoon in the past five years; Long-term warning: historical correlation feature weight≥60%, focusing on selecting the water level increase process curve data of the same typhoon in history; real-time observation feature weight≤40%, mainly with typhoon intensity and moving speed data; S23, time window adjustment: short-term warning window is set to 1 hour, medium-term is 3 hours, and long-term is 6 hours, and window sliding step is 30 minutes; S3, initial inundation prediction: input the weighted features into the pre-trained deep learning warning model, pass through the multi-scale feature fusion module, and output the preliminary prediction result; S4, deviation adaptive correction: call the historical storm surge case library, correct the preliminary prediction by combining the real-time monitoring data with the deviation coefficient, and obtain the corrected inundation data; S5, graded threshold adaptation: combining the corrected inundation data with the preset regional disaster-bearing body characteristic parameters, the warning threshold of different sub-regions is determined by fuzzy analytic hierarchy process, and the warning levels are divided; S6, graded warning release and feedback: generate differentiated warning information according to the warning level, release through the communication mode adapted to the receiving object, the release process includes threshold judgment logic of communication mode adaptation, and real-time tracking of warning information reach rate and response data is used for model iteration optimization.

2. The graded warning method of storm surge inundation area according to claim 1, characterized in that, Step S1 includes the following sub-steps: S11, multi-source heterogeneous data collection: classified collection according to data type; S12, space-time reference unification: convert all collected data to a unified coordinate system and time stamp, calibrate the time synchronization error of different source data, and use weighted fusion algorithm to process the data in the spatial overlapping area; S13, quality grading processing: three-level quality assessment is performed on the data.

3. The graded warning method of storm surge inundation area according to claim 1, characterized in that, The step S3 is specifically: input the weighted features into the pre-trained deep learning warning model, output the preliminary prediction result of the storm surge inundation area through the multi-scale feature fusion module of the model, the preliminary prediction result contains the spatio-temporal distribution data of the inundation range, depth and duration, and the training process of the pre-trained deep learning warning model includes the following sub-steps: S31, sample data set construction: taking the multi-source data of historical storm surge events as training samples, the input features include typhoon intensity, initial tidal level, terrain parameters and trend features in the last 3 hours, and the output label is the vector boundary of the measured inundation range, the inundation depth value of different regions and the duration; S32, model architecture design: using an improved U-Net network architecture, adding an attention gate module at the encoding end to strengthen feature learning in sensitive nearshore areas, and introducing a feature fusion weight of skip connection at the decoding end; S33, model training and optimization: combining the simulation results of the numerical model as auxiliary supervision signals, migrating the simulation data of extreme weather cases to the training set through transfer learning, and training with the Adam optimizer.

4. The graded warning method of storm surge inundation area according to claim 1, characterized in that, The step S4 is specifically: calling the historical storm surge case library, calculating the deviation coefficient of the preliminary prediction result and the historical cases through the case similarity matching algorithm, combining the residual feedback of real-time monitoring data to dynamically correct the preliminary prediction result, and obtaining the corrected inundation data, and the deviation adaptive correction includes the following sub-steps: S41, historical case screening: extracting candidate cases with spatial similarity≥80%, pressure gradient difference≤5hPa and same season from the historical case library, and the number of screening is controlled within 5-10; S42, optimal case matching: calculating the topographic difference coefficient and tidal phase difference between the candidate cases and the current prediction scenario, and determining the optimal matching case through the weighted Euclidean distance formula; S43, deviation calculation and correction: deviation coefficient=(measured water level rise of the optimal case-model simulated water level rise) / measured water level rise of the optimal case, dynamically correcting the deviation coefficient combined with the 5-minute residual of real-time monitoring data, and the corrected deviation is used to adjust the inundation depth and range of the preliminary prediction result.

5. The graded warning method of storm surge inundation area according to claim 1, characterized in that, The determination of the warning threshold by the fuzzy analytic hierarchy process in step S5 includes the following sub-steps: S51, index system construction: taking the corrected inundation data and regional hazard-bearing body characteristic parameters as first-level indicators, and setting 3-5 second-level indicators under each first-level indicator; S52, weight assignment and fuzzy evaluation: determining the weight of each indicator by using the 1-9 scale method, constructing a fuzzy judgment matrix by using a triangular fuzzy number, and calculating the comprehensive weight value of the indicator, wherein the weight proportion of the hazard-bearing body characteristic parameters is not less than 40%; S53, threshold interval division: according to the comprehensive evaluation result, the warning threshold interval of different sub-regions is divided by using the maximum membership degree principle, and the threshold difference of the same warning level in different sub-regions is controlled within ±20%.

6. The graded warning method of storm surge inundation area according to claim 1, characterized in that, The threshold judgment logic of the regional hazard-bearing body characteristic parameters in step S5 is specifically: S51, construct a hazard-bearing body characteristic reference vector V0, including reference population density, reference economic vulnerability and reference emergency capacity; S52, calculate the cosine similarity S between the current area disaster-bearing body feature vector V and the reference vector V0; S53, sliding window statistics of real-time disaster-bearing body data of the current area is generated to generate a feature change rate curve R; S54, calculate the first derivative extreme value D of the change rate curve R, reflecting the mutation degree of the feature parameter; S55, set the vulnerability correlation coefficient K = 0.3 × population density weight + 0.4 × economic vulnerability weight + 0.3 × emergency capability weight; S56, if the cosine similarity S is greater than or equal to 0.8, and the change rate derivative extreme value D is less than or equal to 5% / hour, and the vulnerability correlation coefficient K is less than or equal to 0.4, and the population density mutation ΔP is less than 10%, it is determined that the current disaster-bearing body feature is in a low sensitive state, and the basic warning threshold is adopted; otherwise, S57 is executed; S57, if it is determined that the high sensitive state is adopted, the density clustering algorithm is used to divide the area into sub-units, and S52-S55 are repeatedly executed for each sub-unit, the abnormal fluctuation sub-unit is removed, and the weighted mean of the feature parameters of the remaining sub-units is taken as the final threshold calculation basis.

7. The method of graduated warning of storm surge inundation areas as claimed in claim 1, wherein, The specific rules for dividing the warning level in step S5 are: A three-dimensional grading model of submergence depth-duration-disaster-bearing body vulnerability index is established and stored in the warning system database, and a dynamic threshold checking mechanism is used for each warning trigger condition, with a checking accuracy of ±5% measured value. The three-dimensional grading model is constructed by the following methods: When the disaster-bearing body vulnerability index is less than 0.3, match the low-risk warning sub-model, and set the warning level as follows: when the submergence depth is less than 0.5m and the duration is less than 3h, trigger the blue warning; When the submergence depth is 0.5m to less than 1.0m and the duration is 3h to less than 6h, trigger the yellow warning standby level; When the disaster-bearing body vulnerability index is 0.3 to less than 0.6, match the medium-risk warning sub-model, and the grading strategy is as follows: when the submergence depth is less than 0.5m but the duration is greater than or equal to 3h, or the submergence depth is 0.5m to less than 1.0m and the duration is less than 3h, trigger the yellow warning; When the submergence depth is 1.0m to less than 2.0m and the duration is 3h to less than 6h, trigger the orange warning standby level; When the disaster-bearing body vulnerability index is greater than or equal to 0.6, activate the high-risk warning intensification strategy, specifically: when the submergence depth is less than 1.0m but the duration is greater than or equal to 6h, or the submergence depth is 1.0m to less than 2.0m and the duration is greater than or equal to 3h, trigger the orange warning; when the submergence depth is greater than or equal to 2.0m or the duration is greater than or equal to 12h, or when the vulnerability index is greater than or equal to 0.8 and the real-time submergence depth is greater than or equal to 0.5m, regardless of the duration, directly trigger the red warning; when the submergence depth is greater than or equal to 1.5m and the vulnerability index is greater than or equal to 0.7, the warning level is automatically upgraded by one level.

8. The graded warning method of storm surge inundation area according to claim 1, characterized in that, The threshold judgment logic of the communication mode adapted to the receiving object in step S6 is as follows: S61, construct a communication object feature matrix M, including object type, historical response time, device coverage, information complexity requirement; S62, calculate the matching degree P of different communication modes and object feature matrix, the matching degree of the general public: P=0.4×response timeliness compliance rate+0.3×device compatibility rate+0.3×information bearing adaptation degree; the matching degree of the sea vessel: P=0.6×response timeliness compliance rate+0.4×device compatibility rate; S63, set the communication priority threshold T, wherein T≥0.85 corresponds to the red warning, T≥0.75 corresponds to the orange warning, T≥0.65 corresponds to the yellow warning, and T≥0.55 corresponds to the blue warning; S64, if the matching degree P of a communication mode is greater than or equal to the priority threshold T corresponding to the warning level, the communication mode is included in the candidate communication mode set; S65, the candidate communication mode set is subjected to secondary screening according to real-time channel quality, and the communication mode with optimal or good channel quality is retained; S66, if there is only one communication mode after screening, the mode is directly used for publishing; if there are multiple modes, the top two modes are selected for collaborative publishing according to the matching degree P from high to low.

9. A graded warning system for storm surge inundation areas, characterized by, The system comprises: A data acquisition and preprocessing module for performing step S1, including a multi-source data acquisition unit, a space-time reference unification unit, and a quality grading unit, the multi-source data acquisition unit acquires real-time observation data, historical basic data, and environmental correlation data according to data types, the space-time reference unification unit converts the collected data to a unified coordinate system and time stamp and processes spatially overlapping data, and the quality grading unit performs three-level quality assessment on the data; A feature extraction module for performing step S2, including a feature extraction unit, a matrix construction unit, a weight allocation unit, and a time window adjustment unit, the feature extraction unit extracts real-time observation features and historical correlation features from the standardized data set, the matrix construction unit constructs a feature time series matrix according to the time sequence, the weight allocation unit dynamically adjusts the input weights of the two types of features according to the warning time, and the time window adjustment unit sets the time window and sliding step corresponding to different warning times; An initial prediction module for performing step S3, including a model storage unit and a prediction execution unit, the model storage unit stores a pre-trained deep learning warning model, the model uses an improved U-Net network architecture and is optimized through a sample data set, and the prediction execution unit inputs the weighted features into the model and outputs the preliminary prediction result through a multi-scale feature fusion module; A bias correction module for performing step S4, including a case library unit, a similarity calculation unit, and a bias correction unit, the case library unit stores a historical storm surge case library, the similarity calculation unit selects candidate cases and determines the optimal matching case through a case similarity matching algorithm, and the bias correction unit calculates the bias coefficient and dynamically corrects the preliminary prediction result combined with real-time monitoring data residual feedback; The hierarchical threshold module is used for executing step S5, and includes a parameter storage unit, a threshold calculation unit and a grade division unit. The parameter storage unit stores regional hazard-bearing body characteristic parameters. The threshold calculation unit determines early warning thresholds of different sub-regions by combining the modified inundation data with a fuzzy analytic hierarchy process. The grade division unit divides early warning grades according to a three-dimensional hierarchical model, and contains threshold judgment logic of the regional hazard-bearing body characteristic parameters. The early warning release and feedback module is used for executing step S6, and includes an information generation unit, a communication adaptation unit and a feedback tracking unit. The information generation unit generates differentiated early warning information according to early warning grades. The communication adaptation unit releases the early warning information by adapting to communication modes of receiving objects, and contains threshold judgment logic of the communication mode adaptation. The feedback tracking unit tracks a touch rate of the early warning information and response data in real time, and is used for model iteration optimization.

Images