Sea level rise influence risk assessment method based on multi-source data fusion, medium and system

Abstract

The invention provides a sea level rise influence risk assessment method based on multi-source data fusion, a medium and a system, and belongs to the technical field of marine disasters. Basic feature data are obtained by constructing a multi-dimensional data acquisition network, nonlinear time series decomposition is performed by applying multi-scale wavelet analysis, and the risk assessment result is obtained by analyzing the nonlinear time series; establishing a prediction model based on a deep learning neural network; determining key vulnerabilities by using a minimum cut maximum flow algorithm; establishing a regional vulnerability scoring system; generating a comprehensive risk assessment matrix by using a Bayesian network ensemble analysis method; and finally, generating an evaluation report by using the coastline adaptability gating model. According to the model, a multi-head time sequence attention mechanism and bidirectional autoregression coding are fused, parameters are dynamically adjusted by adopting a regional adaptability gating weighting function, and accurate evaluation and dynamic tracking of sea level rising influence risks in a complex geographical environment are realized.

Description

Technical Field

[0001] The present invention belongs to the field of marine disaster technology, and specifically relates to a sea level rise impact risk assessment method, medium and system based on multi-source data fusion. Background Art

[0002] Sea level rise is one of the most significant impacts of global climate change, posing a significant threat to the livelihoods and economic development of coastal areas. Traditional sea level rise risk assessment methods primarily rely on single data sources and linear models, such as sea level change monitoring based on satellite altimeter data and static inundation simulations using geographic information systems. These methods typically employ deterministic models, overlaying a preset sea level rise value onto a digital elevation model to directly determine the inundation range and depth distribution, or employ simple statistical regression models to predict future sea level trends. However, traditional assessment methods have significant flaws: First, a single data source fails to fully capture the complexity of sea level change and overlooks important influencing factors such as regional geological subsidence and climate cycles; second, linear prediction models cannot effectively identify critical points and nonlinear characteristics in the sea level change process, resulting in insufficient prediction accuracy; third, existing methods lack comprehensive consideration of terrain complexity and socioeconomic factors, making it difficult to accurately locate key vulnerabilities and assess the dynamic evolution of risks. In a complex geographical environment, the impact of sea level rise exhibits strong nonlinear characteristics and regional differences. Traditional single data sources and linear models are unable to accurately assess the risk distribution and evolution trends in such a complex system, especially the inability to effectively identify potential key vulnerabilities and risk inflection points, which seriously restricts the precise formulation and implementation of adaptive management strategies in coastal areas. Summary of the Invention

[0003] In view of this, the present invention provides a sea level rise impact risk assessment method, medium and system based on multi-source data fusion, which can solve the technical problem in the existing technology that it is impossible to accurately assess the risk of sea level rise impact in complex geographical environments.

[0004] The present invention is implemented as follows: In a first aspect, the present invention provides a method for assessing the impact of sea level rise based on multi-source data fusion, including: constructing a multi-dimensional data acquisition network to obtain basic characteristics of sea level rise; using multi-scale wavelet analysis to perform nonlinear decomposition on sea level change time series data, extracting different frequency components and identifying critical point signals; constructing a nonlinear prediction model for sea level rise based on a deep learning neural network; using the minimum cut maximum flow algorithm to construct a regional flooding risk network model, calculating the minimum cut set to determine key vulnerable points and potential flooding area boundaries; establishing a regional vulnerability scoring system and calculating the vulnerability index; applying the Bayesian network integrated analysis method to generate a comprehensive risk assessment matrix; calculating the change sub-matrix of the comprehensive risk assessment matrix to extract the risk change rate and trend characteristics; and using the coastline adaptability gating model to generate a comprehensive assessment report on regional sea level rise risks.

[0005] Among them, the multi-dimensional data acquisition network refers to a multi-source heterogeneous data acquisition system composed of a satellite remote sensing system, a tide station monitoring system, a GNSS surface deformation monitoring station and an unmanned aerial survey system, which is used to obtain all-round sea level change and land response information.

[0006] Among them, the minimum cut maximum flow algorithm is a method for solving optimization problems in graph theory. In the sea level rise risk assessment, the terrain is regarded as a network graph, the elevation is used as the node capacity, and the water flow channel is used as the edge. By calculating the minimum cut set, the weak link that is most likely to break through the defense line is determined, and the key vulnerable points and the boundaries of the potential flooding area are accurately located.

[0007] Among them, the vulnerability index is a comprehensive score obtained by weighted calculation of terrain height distribution, protection facility conditions, land use type, population density and economic value, reflecting the sensitivity and impact of the region in the face of sea level rise.

[0008] The comprehensive risk assessment matrix is ​​a two-dimensional tabular data structure, wherein the horizontal axis represents different sea level rise scenarios, the vertical axis represents different regional units, and the matrix element values ​​are the comprehensive risk scores of the regional units under the corresponding scenarios.

[0009] Among them, the change matrix is ​​a matrix generated by calculating the differences in the element values ​​of the comprehensive risk assessment matrix between different time nodes. It contains risk change rate, change acceleration and trend inflection point information, and is used to quantify the dynamic evolution characteristics of regional risks.

[0010] Among them, the coastline adaptive gating model is a comprehensive analysis framework that integrates a multi-layer perceptron and a temporal attention mechanism. Its specific structure is a neural network architecture consisting of an input layer, a multi-head temporal attention layer, a bidirectional autoregressive encoding layer, a dynamic gating integration layer and an output layer.

[0011] Among them, the coastline adaptive gating model adopts a regional adaptive gating weight function to dynamically adjust parameter weights according to risk changes, so as to achieve an accurate assessment of the risk of sea level rise in a complex geographical environment; the regional adaptive gating weight function calculates dynamic gating parameters based on the change sub-matrix of the comprehensive risk assessment matrix, the regional terrain complexity index and the socio-economic vulnerability distribution, and the regional adaptive gating weight function adopts a segmented activation mechanism.

[0012] Among them, the multi-scale wavelet analysis is a mathematical tool for time series decomposition, which can decompose sea level change time series data into interannual changes, seasonal changes, long-term trends and mutation signals, and effectively identify nonlinear change patterns that traditional linear models cannot capture.

[0013] Among them, the critical point signal refers to the sudden acceleration or deceleration point that occurs during sea level change, which usually indicates a change in certain key physical processes, such as important events such as ice shelf collapse or ocean current changes.

[0014] Among them, the regional adaptive gating weight function first calculates the comprehensive balance value, which is a weighted combination of the risk change rate in the change score matrix, the regional terrain complexity index and the socio-economic vulnerability distribution; when the comprehensive balance value is in the low-risk range, a conservative weight adjustment function is adopted; when the comprehensive balance value is in the medium-risk range, a balanced weight adjustment function is adopted; when the comprehensive balance value is in the high-risk range, an aggressive weight adjustment function is adopted.

[0015] Among them, the regional terrain complexity index refers to a comprehensive indicator generated by calculating the elevation change rate, slope distribution and terrain undulation within the region, which is used to quantify the complexity of the terrain structure's response to sea level rise.

[0016] The socioeconomic vulnerability distribution refers to the spatial distribution characteristics of socioeconomic factors such as population density, infrastructure value, economic activity intensity and land use type in the region, which is used to assess the potential impact of sea level rise on the socioeconomic system.

[0017] Among them, the method for obtaining the change score matrix is ​​specifically to establish a set of time series sampling points, generate a corresponding comprehensive risk assessment matrix for each sampling time point, calculate the element level difference value to obtain the initial change matrix, further calculate the change rate matrix, perform the second-order difference of the change rate matrix in the time dimension to obtain the change acceleration matrix, and finally integrate it into a complete change score matrix.

[0018] A second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions are run in a computer, they are used to execute the above-mentioned sea level rise impact risk assessment method based on multi-source data fusion.

[0019] The third aspect of the present invention provides a sea level rise impact risk assessment system based on multi-source data fusion, which includes the above-mentioned computer-readable storage medium. The system is any one of a computer, a server, and a single-chip microcomputer. The computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

[0020] The present invention realizes comprehensive risk assessment by constructing a multidimensional data acquisition network, multi-scale wavelet analysis, a deep learning neural network prediction model, a minimum cut maximum flow algorithm, and a coastline adaptive gating model. This method effectively solves the defects of traditional technologies: the multidimensional data acquisition network integrates remote sensing images, water level monitoring and surface deformation data to fully capture the multidimensional characteristics of sea level changes; the multi-scale wavelet analysis and deep learning neural network model can accurately identify the nonlinear characteristics and key critical points of sea level changes; the minimum cut maximum flow algorithm is combined with the regional vulnerability scoring system to accurately locate key vulnerable points and quantify the degree of risk; the comprehensive risk assessment matrix and its variable sub-matrix can dynamically track the risk evolution trend and inflection point characteristics. Through the organic combination of the above technical means, the present invention can accurately assess the impact risk of sea level rise in complex geographical environments, accurately identify key vulnerable points and dynamic risk change trends, and provide scientific support for the formulation of differentiated adaptive management strategies for coastal areas, thereby effectively solving the technical problem that existing technologies cannot accurately assess the impact risk of sea level rise in complex geographical environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flow chart of the method of the present invention.

[0022] Figure 2 Schematic diagram of the structure of the nonlinear prediction model for sea level rise.

[0023] Figure 3 Schematic diagram of the coastline adaptive gating model structure.

[0024] Figure 4 Schematic diagram of multi-dimensional data acquisition network.

[0025] Figure 5 Schematic diagram of the flooding risk network model of the minimum cut maximum flow algorithm.

[0026] Figure 6 Schematic diagram of sea level change time series and wavelet analysis results.

[0027] Figure 8 This is a map showing predicted sea level rise under different scenarios.

[0028] Figure 9 This is a diagram of sea level rise risk assessment results under different scenarios. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0030] like Figure 1 FIG. 1 is a flow chart of a method for assessing the risk of sea level rise based on multi-source data fusion according to the first aspect of the present invention. The method comprises the following steps:

[0031] S01. Build a multi-dimensional data acquisition network to obtain basic characteristics of sea level rise through remote sensing image data, water level monitoring station data, surface deformation data, and regional digital elevation models;

[0032] S02. Use multi-scale wavelet analysis to perform nonlinear decomposition on sea level change time series data, extract different frequency components and identify critical point signals;

[0033] S03. Construct a nonlinear sea level rise prediction model based on deep learning neural networks, integrating climate simulation data, historical observation data, and regional geological subsidence data as input parameters;

[0034] S04. Use the minimum cut maximum flow algorithm to construct a regional flooding risk network model, combining terrain height nodes with water channel edge weights, and calculate the minimum cut set to determine key vulnerable points and potential flooding area boundaries;

[0035] S05. Establish a regional vulnerability scoring system and calculate the vulnerability index based on terrain height distribution, protection facility conditions, land use type, population density and economic value;

[0036] S06. Apply the Bayesian network ensemble analysis method to integrate the prediction results of the sea level rise nonlinear prediction model with the vulnerability index of the regional vulnerability scoring system to generate a comprehensive risk assessment matrix;

[0037] S07. Calculate the change matrix of the comprehensive risk assessment matrix, and extract the risk change rate and trend characteristics by comparing the comprehensive risk assessment matrices at different time points;

[0038] S08. Use the coastline adaptability gating model to generate a comprehensive assessment report on regional sea level rise risks, including a risk level distribution map, a list of key vulnerable points, a time series risk evolution trend, and risk rating classification results.

[0039] Among them, the multi-dimensional data acquisition network refers to a multi-source heterogeneous data acquisition system composed of satellite remote sensing systems, tide station monitoring systems, GNSS surface deformation monitoring stations and drone aerial survey systems, which is used to obtain all-round sea level change and land response information.

[0040] Among them, multi-scale wavelet analysis is a mathematical tool used for time series decomposition. It can decompose sea level change time series data into interannual changes, seasonal changes, long-term trends and mutation signals, and effectively identify nonlinear change patterns that traditional linear models cannot capture.

[0041] Among them, critical point signals refer to sudden acceleration or deceleration points in the process of sea level change, which usually indicate changes in certain key physical processes, such as ice shelf collapse or ocean current changes.

[0042] Among them, the minimum cut maximum flow algorithm is a method for solving optimization problems in graph theory. In the sea level rise risk assessment, the terrain is regarded as a network graph, the elevation is used as the node capacity, and the water flow channel is used as the edge. By calculating the minimum cut set, the weak link that is most likely to break through the defense line is determined, and the key vulnerable points and the boundaries of the potential flooding area are accurately located.

[0043] Among them, the vulnerability index is a comprehensive score obtained by weighted calculation of factors such as terrain height distribution, status of protective facilities, land use type, population density and economic value, reflecting the sensitivity and degree of impact of a region in the face of sea level rise.

[0044] Among them, the comprehensive risk assessment matrix is ​​a two-dimensional tabular data structure. The horizontal axis represents different sea level rise scenarios, the vertical axis represents different regional units, and the matrix element values ​​are the comprehensive risk scores of the regional units under the corresponding scenarios.

[0045] Among them, the change matrix is ​​a matrix generated by calculating the differences in the element values ​​of the comprehensive risk assessment matrix between different time nodes. It contains information on the risk change rate, change acceleration and trend inflection point, and is used to quantify the dynamic evolution characteristics of regional risks.

[0046] Among them, the coastline adaptive gating model is a comprehensive analysis framework that integrates a multi-layer perceptron and a temporal attention mechanism, providing intelligent recommendations for resource allocation and protection decisions under sea level rise scenarios.

[0047] The specific structure of the coastline adaptive gating model is a neural network architecture consisting of an input layer, a multi-head temporal attention layer, a bidirectional autoregressive encoding layer, a dynamic gating integration layer, and an output layer. The input layer receives climate scenario data, terrain feature vectors, and socioeconomic indicators. The multi-head temporal attention layer extracts features based on changing trends at different time scales. The bidirectional autoregressive encoding layer captures the causal relationship between sea level changes and regional responses. The dynamic gating integration layer dynamically adjusts the weights of each feature channel based on the change sub-matrix in the comprehensive risk assessment matrix. The output layer generates regional protection strategy recommendations and resource optimization allocation plans. The dynamic gating integration layer uses a regional adaptive gating weight function for parameter adjustment.

[0048] The steps for establishing the training data set of the coastline adaptive gating model specifically include collecting historical sea level change data and corresponding protection measures implementation records for typical coastline areas around the world, integrating inundation range maps and actual impact assessment reports for each region under different sea level rise scenarios, extracting terrain change characteristics and socioeconomic response indicators before and after the implementation of regional protection projects, constructing a parallel case library containing multiple regions and multiple time points, normalizing and aligning all collected data in time and space, marking the protection measures effectiveness score and cost-effectiveness ratio of each case, and finally dividing the data into training set, validation set and test set in chronological order.

[0049] The steps of training the coastline adaptive gating model specifically include using a batch stochastic gradient descent algorithm to initialize and pre-train the model parameters, introducing a self-supervised learning mechanism for time series prediction to enhance the model's ability to capture long-term trends, using the protection measures implementation effect data in historical cases to calculate the protection decision loss function to guide model optimization, improving the model's ability to identify commonalities and differences between different regions through comparative learning, designing a hierarchical training strategy to optimize different component parameters in stages, introducing a regional feature adaptive mechanism to enable the model to dynamically adjust parameter weights according to the characteristics of different regions, and finally using the change sub-matrix in the comprehensive risk assessment matrix to fine-tune and verify the trained model.

[0050] The method for obtaining the change score matrix is ​​specifically as follows: first, a set of time series sampling points is established, a corresponding comprehensive risk assessment matrix is ​​generated for each sampling time point, element-level difference values ​​are calculated for the comprehensive risk assessment matrices of adjacent time points to obtain an initial change matrix, and the change rate of each element, that is, the ratio of the change amount to the time interval, is further calculated to form a change rate matrix, the change rate matrix is ​​subjected to second-order difference in the time dimension to obtain a change acceleration matrix, significant change points in the change acceleration matrix are identified and marked as trend inflection points, and finally the change rate matrix, the change acceleration matrix and the trend inflection point markers are integrated into a complete change score matrix.

[0051] The regional adaptive gating weight function calculates dynamic gating parameters based on the change sub-matrix of the comprehensive risk assessment matrix, the regional terrain complexity index and the socio-economic vulnerability distribution. The regional adaptive gating weight function adopts a segmented activation mechanism. First, the comprehensive balance value is calculated. The comprehensive balance value is a weighted combination of the risk change rate in the change sub-matrix, the regional terrain complexity index and the socio-economic vulnerability distribution. When the comprehensive balance value is in a low-risk range, a conservative weight adjustment function is adopted to maintain high stability and gradual adjustment. When the comprehensive balance value is in a medium-risk range, a balanced weight adjustment function is adopted to achieve a balance between stability and response speed. When the comprehensive balance value is in a high-risk range, an aggressive weight adjustment function is adopted to quickly increase the resource allocation weight of key protective measures. The boundary thresholds of each interval are dynamically adjusted according to the historical response characteristics of the region, thereby realizing adaptive optimization of the gating parameters and improving the response speed and accuracy of the model to sudden risk events.

[0052] The conservative weight adjustment function refers to a weight adjustment function that adopts a linear growth model to ensure that the weight changes are smooth and gradual to avoid system fluctuations. The balanced weight adjustment function refers to a weight adjustment function that adopts an S-curve model, which adopts a smaller adjustment range in the low-risk and high-risk intervals and a larger adjustment range in the medium-risk interval to achieve a balance between smooth transition and timely response. The aggressive weight adjustment function refers to a weight adjustment function that adopts an exponential growth model to quickly increase the weight after the risk exceeds the threshold to ensure that the system responds quickly to high-risk situations. The regional terrain complexity index refers to a comprehensive indicator generated by calculating the elevation change rate, slope distribution and terrain undulation in the region, which is used to quantify the complexity of the terrain structure's response to sea level rise. The socioeconomic vulnerability distribution refers to the spatial distribution characteristics of socioeconomic factors such as population density, infrastructure value, economic activity intensity and land use type in the region, which is used to assess the potential impact of sea level rise on the socioeconomic system.

[0053] The specific implementation of the above steps is described in detail below.

[0054] The specific implementation method of step S01 is to build a multi-dimensional data acquisition network, which consists of a satellite remote sensing system, a tide station monitoring system, a GNSS surface deformation monitoring station and a drone aerial survey system. First, a fixed tide station network is deployed with a spacing of no more than 50 kilometers to ensure continuous collection of water level data along the coastline; then GNSS surface deformation monitoring stations are deployed with a monitoring point density of no less than 1 per 100 square kilometers to capture millimeter-level surface vertical deformation; then regional digital elevation model data with a resolution of no less than 0.5 meters is obtained as the basis for terrain analysis; finally, multi-phase remote sensing images are integrated, including optical images, SAR (synthetic aperture radar) images and drone aerial images, with a temporal resolution of no less than once a quarter. A data fusion algorithm is used to unify data from different sources into the WGS84 coordinate system to form a temporally and spatially consistent sea level change monitoring data set. The purpose of this step is to obtain the basic characteristics of sea level rise through multi-source heterogeneous data collection, and to provide comprehensive and accurate data support for subsequent analysis.

[0055] The specific implementation of step S02 involves nonlinearly decomposing the sea level change time series data using multiscale wavelet analysis. First, a suitable wavelet basis function, such as the Daubechies wavelet or the Morlet wavelet, is selected to perform a wavelet transform on the original time series. Then, by adjusting the scale parameter, the time series is decomposed into different frequency components, including interannual variation (2-7 years), seasonal variation (0.25-1 years), long-term trend (>10 years), and abrupt changes. The wavelet energy spectrum at each scale is then calculated to identify areas of energy anomaly concentration as potential critical points. Finally, wavelet phase analysis is used to determine the precise temporal location of the critical point, with the threshold set at twice the standard deviation of the local maximum of the wavelet coefficients exceeding the mean. The critical point signal is determined when the sea level change rate in a short period (no more than three months) exceeds three times the average change rate over the previous five years. This step leverages the multi-resolution characteristics of wavelet analysis to effectively separate the multiscale components of sea level change, particularly capturing nonlinear changes and abrupt changes that are difficult to identify with traditional linear models, providing key characteristic information for sea level rise prediction.

[0056] The specific implementation of step S03 is to build a nonlinear prediction model for sea level rise based on a deep learning neural network. First, a hybrid architecture of a long short-term memory network (LSTM) and a convolutional neural network (CNN) is constructed. The number of LSTM units is set to 64 to 128, and the CNN adopts a 3 to 5-layer structure with a convolution kernel size of 3×3. Then, historical sea level observation data (at least 30 years of continuous records), satellite altimeter data, temperature and precipitation forecasts output by the global climate model (GCM), regional geological subsidence rate and other multi-source parameters are input. Then, the model's ability to capture long-term trends and sudden events is enhanced through the attention mechanism, and the depth of the attention layer is set to 2 to 3 layers. Finally, multiple prediction time scales are set, including short-term (1 to 5 years), medium-term (5 to 20 years) and long-term (20 to 100 years) prediction outputs. The model training adopts a batch size of 64, an initial learning rate of 0.001 and uses cosine annealing scheduling, with 200 to 500 training rounds. The purpose of this step is to construct a prediction model that can effectively capture the nonlinear characteristics of sea level changes, achieve scientific predictions of future sea level rise trends, and provide basic data support for risk assessment.

[0057] The specific implementation of step S04 is to construct a regional flooding risk network model using the minimum cut maximum flow algorithm. First, the study area is discretized into a grid structure. The grid size is determined by the complexity of the terrain, usually 10 to 50 meters. Then, a directed network graph is constructed. The nodes represent terrain elevation points, and the capacity is set as the difference between the elevation value of the point and the predicted sea level. Negative values ​​indicate flooded areas. Next, edge weights are defined as the smoothness of the water flow channel between adjacent nodes. The calculation formula is the ratio of the elevation difference between adjacent nodes to the distance multiplied by the surface roughness coefficient. The surface roughness coefficient is determined by the land use type, such as 0.01 to 0.03 for urban construction areas, 0.03 to 0.05 for farmland, and 0.05 to 0.10 for forests. Finally, the Ford-Fulkerson algorithm or the Dinic algorithm is used to solve the maximum flow problem and identify the minimum cut set in the network. The nodes connected by the edges in the minimum cut set constitute the boundaries of the potential flooding area. The purpose of this step is to use the minimum cut maximum flow algorithm in graph theory to accurately simulate the path of seawater intrusion, locate the weak links and key vulnerabilities of the protection system, and provide a spatial analysis basis for risk assessment.

[0058] The specific implementation of step S05 is to establish a regional vulnerability scoring system. First, a scoring index system is determined, including five dimensions: terrain height distribution (weight 0.25), protective facility status (weight 0.20), land use type (weight 0.15), population density (weight 0.20), and economic value (weight 0.20). Then, each indicator is standardized, converting the original value into a score range of 0 to 10. The terrain height standard score is calculated as 10 × (current elevation - predicted sea level) / (regional highest point - predicted sea level), and the value is 0 if it is less than 0. Then, a basic score is assigned according to land use type, such as 8 to 10 points for urban residential areas, 7 to 9 points for industrial areas, 5 to 7 points for farmland, 3 to 5 points for forest land, 2 to 4 points for grassland, and 0 to 2 points for unused land. Finally, the weighted values ​​of each indicator are combined to calculate the vulnerability index, which is divided into five levels: very low (0 to 2), low (2 to 4), medium (4 to 6), high (6 to 8), and very high (8 to 10). The purpose of this step is to establish a quantitative standard for regional vulnerability through a comprehensive scoring of multiple indicators, identify sensitive areas susceptible to sea level rise, and provide an important basis for comprehensive risk assessment.

[0059] The specific implementation of step S06 is to integrate the model prediction results and the vulnerability index using the Bayesian network integration analysis method. First, a Bayesian network structure is constructed, with nodes including key factors such as predicted sea level height, proportion of flooded area, status of protective facilities, population impact ratio, and economic loss estimation. Then, a conditional probability table (CPT) is set between nodes based on historical data and expert experience. For example, if the sea level rise is greater than 50 cm, the probability thresholds for severe flooding in areas with different vulnerability levels are: 0.05 for extremely low vulnerability areas, 0.15 for low vulnerability areas, 0.35 for medium vulnerability areas, 0.65 for high vulnerability areas, and 0.85 for extremely high vulnerability areas. Then, the sea level rise prediction result from step S03 and the vulnerability index from step S05 are input as observational evidence, and the posterior probability distribution is calculated using the belief propagation algorithm. Finally, a comprehensive risk assessment matrix is ​​generated, with the horizontal axis representing different sea level rise scenarios (such as RCP2.6, RCP4.5, RCP8.5, etc.) and the vertical axis representing the assessment unit (grid or administrative division). The matrix elements are the comprehensive risk scores (0-100) of the regional units under the corresponding scenarios. The purpose of this step is to use the uncertainty reasoning ability of Bayesian networks to scientifically integrate sea level rise predictions and regional vulnerability assessment results to form a comprehensive risk quantification indicator.

[0060] The specific implementation of step S07 is to calculate the change matrix of the comprehensive risk assessment matrix. First, determine the time series sampling point set, usually selecting key time nodes such as 2030, 2050, 2070, and 2100; then generate the corresponding comprehensive risk assessment matrix M for each sampling time point. t; Then calculate the matrix element difference ΔM=M between adjacent time points t1 and t2 t2 -M t1 , obtain the initial change matrix; then calculate the change rate matrix R = ΔM / (t2-t1), which reflects the risk change speed per unit time; then perform the second-order difference in the time dimension to obtain the change acceleration matrix A = (R t+1 -R t ) / Δt; finally, identify significant change points in the acceleration matrix, setting the threshold at points where the acceleration exceeds 2.5 standard deviations of the mean, and mark these points as trend inflection points. The purpose of this step is to capture the temporal evolution of risk through matrix operations, specifically identifying acceleration or deceleration trends in risk changes, providing a temporal dynamic analysis basis for risk management decisions.

[0061] The specific implementation of step S08 involves generating a comprehensive assessment report using the Coastline Adaptability Gating Model. First, the comprehensive risk assessment matrix generated in step S06 and the change score matrix generated in step S07 are input into the Coastline Adaptability Gating Model. A risk level distribution map is then generated based on the model's calculation results, using a red, orange, yellow, and green color system to represent extremely high, high, medium, and low risk areas. Next, the top 10% of the comprehensive risk scores are selected as a list of key vulnerabilities, including spatial coordinates, risk scores, major risk factors, and protection recommendations. A time series risk evolution trend chart is then generated, with time plotted on the horizontal axis and risk scores plotted on the vertical axis, showing the trajectory of risk changes across different regions. Finally, a risk rating classification table is generated, classifying regions by risk urgency and formulating differentiated protection strategy recommendations. The rating classification uses a four-level scale: Level I (extremely urgent, action required within 5 years), Level II (urgent, action required within 5-15 years), Level III (medium, action required within 15-30 years), and Level IV (low urgency, consideration after 30 years). The purpose of this step is to integrate the results of previous analyses and generate an intuitive and clear risk assessment report to provide a scientific basis and action guide for coastal management and protection decisions.

[0062] The detailed structure of the coastline adaptive gating model is a neural network architecture consisting of an input layer, a multi-head temporal attention layer, a bidirectional autoregressive encoding layer, a dynamic gated integration layer, and an output layer. The input layer receives three types of data: climate scenario data (including temperature changes, precipitation patterns, and sea level rise projections under different RCP scenarios), terrain feature vectors (including terrain parameters such as elevation distribution, slope, and aspect), and socioeconomic indicators (including population density, GDP, and infrastructure value). The multi-head temporal attention layer uses eight attention heads, each focusing on characteristics of change at different time scales, covering trends from short-term (interannual) to long-term (century). The bidirectional autoregressive encoding layer consists of 128 LSTM units, processing time series data in both forward and backward directions to capture the causal relationship between sea level change and regional responses. The dynamic gated integration layer dynamically adjusts the weights of each feature channel based on a regional adaptive gating weight function. It calculates gating parameters based on the change matrix in the comprehensive risk assessment matrix to adjust the priority of different protection strategies. The output layer generates regional protection strategy recommendations and resource optimization allocation plans, including the selection of engineering protection measures, implementation timing arrangements, and investment priority rankings.

[0063] The detailed steps for establishing the training dataset for the coastline adaptive gating model are as follows: First, historical sea level change data for 25 typical coastline areas around the world are collected, with a time span of no less than 50 years. The data include monthly average sea level height, frequency and intensity of extreme events; then, inundation range maps and actual impact assessment reports for each region under different sea level rise scenarios are integrated, including more than 100 case studies; then, topographic change characteristics and socioeconomic response indicators before and after the implementation of regional protection projects are extracted, such as changes in the frequency of flood events and the reduction ratio of economic losses before and after the construction of protection dams; then, a multi-scenarios dataset is constructed. A parallel case library covering multiple time points in the region, with a total data volume of no less than 500 typical cases; data normalization is then performed to unify indicators of different dimensions into the range of 0 to 1, and spatial and temporal alignment is performed to ensure comparability between different cases; then the protective measures effectiveness score (0 to 10 points) and cost-effectiveness ratio of each case are marked. The effectiveness score is determined based on the actual effect of disaster prevention and mitigation, and the cost-effectiveness ratio is calculated as the ratio of input cost to avoided loss; finally, the data is divided into a training set (70%), a validation set (15%), and a test set (15%) in chronological order to ensure the scientific nature of model training and evaluation.

[0064] The coastline adaptive gating model training adopts a general model training method. The detailed steps are as follows: First, the batch stochastic gradient descent algorithm is used to initialize and pre-train the model parameters. The batch size is set to 32 and the initial learning rate is 0.0005. Then, a self-supervised learning mechanism for time series prediction is introduced to enhance the model's ability to capture long-term trends by predicting the sea level height and impact range at future time points. Then, the protection decision loss function is calculated using the protection measures implementation effect data in historical cases, and the weighted cross entropy loss function is used to guide model optimization. The weights are set according to the decision error cost. The model is then improved through comparative learning. To enhance the ability to identify commonalities and differences between different regions, positive sample pairs (similar regions) and negative sample pairs (different regions) are constructed for comparative training. A hierarchical training strategy is then designed. In the first stage, high-level parameters are frozen and only the underlying feature extraction parameters are trained. In the second stage, all parameters are unfrozen for end-to-end fine-tuning. A regional feature adaptation mechanism is then introduced, and the regional feature vectors are embedded in the model parameter space through the regional encoder, so that the model can dynamically adjust the parameter weights according to the characteristics of different regions. Finally, the change matrix in the comprehensive risk assessment matrix is ​​used to fine-tune and verify the trained model to ensure the model's sensitivity to risk changes and its response accuracy.

[0065] The important technical ideas of the present invention are described below.

[0066] First of all, multi-source data fusion technology is one of the key innovations of the present invention. Traditional sea level rise risk assessment methods often rely on a single data source and cannot fully capture the complexity of sea level changes. The multi-dimensional data acquisition network constructed by the present invention integrates data from satellite remote sensing systems, tide station monitoring systems, GNSS surface deformation monitoring stations and drone aerial survey systems, realizing all-round acquisition of sea level changes and land response information. This multi-source data fusion method solves the limitations of a single data source in principle, and can simultaneously consider multiple influencing factors such as regional geological subsidence, climate cycles, tidal changes, etc., and provide more comprehensive sea level change characteristic information, thereby laying a more reliable data foundation for subsequent risk assessment.

[0067] Secondly, the nonlinear time series analysis method is another core technical idea of ​​the present invention. Traditional sea level change analysis mostly uses simple statistical methods such as linear regression, which makes it difficult to identify complex time series characteristics. The present invention uses multi-scale wavelet analysis technology to perform nonlinear decomposition of sea level change time series data, which can decompose complex time series into different frequency components, and effectively identify nonlinear change patterns such as interannual changes, seasonal changes, long-term trends and mutation signals. This method can accurately capture critical point signals in sea level changes. These critical points usually imply the transformation of important physical processes and are of key significance to risk assessment. In principle, this technology breaks through the limitations of traditional linear analysis methods and improves the ability to identify complex sea level change patterns.

[0068] Third, the graph-theory-based risk network model is a key technological innovation of this invention. Traditional methods often use simple overlay analysis to determine the scope of inundation, ignoring the complexity of terrain structure. This invention utilizes the minimum cut maximum flow algorithm to construct a regional inundation risk network model, treating the terrain as a network graph, using elevation as node capacity, and water flow channels as edges. By calculating the minimum cut set, it determines the weakest links most likely to breach the defense line. This method mathematically optimizes the inundation risk assessment process, can accurately locate key vulnerable points and the boundaries of potential inundation areas, and avoids the assessment bias caused by ignoring terrain complexity in traditional methods.

[0069] Finally, the coastline adaptive gating model integrates the above technologies to form the most distinctive collaborative innovation of the present invention. The model uses a multi-head temporal attention mechanism and a bidirectional autoregressive encoding layer to capture the complex causal relationship between sea level changes and regional responses, and dynamically adjusts the parameter weights according to risk changes through a regional adaptive gating weight function. This design fundamentally solves the problem that traditional static models are unable to cope with dynamic risk changes, and realizes real-time adjustment and optimization of risk assessment. The synergistic effect of various technical ideas enables the present invention to comprehensively, accurately and dynamically assess the risk of sea level rise in complex geographical environments, providing a scientific basis for the formulation of differentiated adaptive management strategies in coastal areas, thereby greatly improving the accuracy and effectiveness of sea level rise risk management.

[0070] A second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions are run in a computer, they are used to execute the above-mentioned sea level rise impact risk assessment method based on multi-source data fusion.

[0071] The third aspect of the present invention provides a sea level rise impact risk assessment system based on multi-source data fusion, which includes the above-mentioned computer-readable storage medium. The system is any one of a computer, a server, and a single-chip microcomputer. The computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

[0072] Specifically, the principle of the present invention is: the technical principle of the present invention to solve the problem of risk assessment of sea level rise in complex geographical environment is mainly based on advanced computing methods such as multi-source heterogeneous data fusion, nonlinear time series analysis, graph network optimization and adaptive gated neural network.

[0073] First, the multidimensional data acquisition network integrates data from satellite remote sensing systems, tide station monitoring systems, GNSS surface deformation monitoring stations, and drone aerial survey systems to achieve comprehensive acquisition of information on sea level changes and land responses. This multi-source data fusion method can overcome the limitations of a single data source and provide more comprehensive information on sea level change characteristics. Second, multiscale wavelet analysis technology can decompose complex sea level change time series into different frequency components, effectively identifying nonlinear change patterns such as interannual changes, seasonal changes, long-term trends, and mutation signals. In particular, its ability to identify critical point signals is significantly superior to traditional linear analysis methods.

[0074] In terms of risk assessment, this paper uses a minimum-cut maximum-flow algorithm to construct a regional flooding risk network model. This method treats terrain as a network diagram, with elevation as node capacity and water channels as edges. By calculating minimum cut sets, it identifies the weakest links in the defense line that are most vulnerable to breaching, thereby accurately locating key vulnerable points. The regional vulnerability scoring system uses a weighted calculation of factors such as terrain height distribution, the condition of protective facilities, land use type, population density, and economic value to form a comprehensive vulnerability index that reflects a region's sensitivity to sea level rise.

[0075] The core innovation of this paper lies in its coastline adaptive gating model, which uses a multi-head temporal attention mechanism and a bidirectional autoregressive encoding layer to capture the complex causal relationship between sea level change and regional response. It also dynamically adjusts parameter weights based on risk changes through a regional adaptive gating weight function. The model employs a segmented activation mechanism for different risk levels. In low-risk intervals, a conservative weight adjustment function maintains stability; in medium-risk intervals, a balanced weight adjustment function balances stability and timely response; and in high-risk intervals, an aggressive weight adjustment function ensures rapid response. This allows for accurate assessment and dynamic tracking of the impact of sea level rise in complex geographical environments.

[0076] In addition, the comprehensive risk assessment matrix and its change sub-matrix, as data structure innovations, can systematically express the risk distribution and temporal evolution characteristics of each regional unit under different sea level rise scenarios, provide decision makers with intuitive risk dynamic information, and support scientific decision-making.

[0077] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0078] The specific implementation method of step S01 is to build a multi-dimensional data acquisition network, which consists of a satellite remote sensing system, a tide station monitoring system, a GNSS surface deformation monitoring station and a UAV aerial survey system. First, a fixed tide station network is deployed with a spacing of no more than 50 kilometers to ensure continuous collection of water level data along the coastline; then GNSS surface deformation monitoring stations are deployed with a monitoring point density of no less than 1 per 100 square kilometers to capture millimeter-level surface vertical deformation; then regional digital elevation model data with a resolution of no less than 0.5 meters is obtained as the basis for terrain analysis; finally, multi-phase remote sensing images are integrated, including optical images, SAR (synthetic aperture radar) images and UAV aerial images, with a time resolution of no less than once a quarter. A data fusion algorithm is used to unify data from different sources into the WGS84 coordinate system to form a sea level change monitoring data set that is consistent in time and space. The purpose of this step is to obtain the basic characteristics of sea level rise through multi-source heterogeneous data collection, and to provide comprehensive and accurate data support for subsequent analysis. The data fusion process adopts the weighted average method, and the specific formula is: Where D fusion is the data value after fusion; D i is the observation value of the i-th data source; w i is the weight coefficient of the i-th data source, and satisfies n is the total number of data sources. Weight coefficient w i Determined through historical accuracy assessment of each data source, the calculation formula is: Where, σ i is the historical observation standard deviation of the i-th data source. The smaller the standard deviation, the greater the weight.

[0079] The specific implementation of step S02 is to use multi-scale wavelet analysis to perform nonlinear decomposition on the sea level change time series data. First, a suitable wavelet basis function, such as Daubechies wavelet or Morlet wavelet, is selected to perform a wavelet transform on the original time series. Then, by adjusting the scale parameter, the time series is decomposed into different frequency components, such as interannual variation (2-7 years), seasonal variation (0.25-1 years), long-term trend (>10 years), and mutation signals. Next, the wavelet energy spectrum of each scale is calculated to identify areas of abnormal energy concentration as potential critical points. Finally, the precise time location of the critical point is determined through wavelet phase analysis, with the threshold set at twice the standard deviation of the local maximum of the wavelet coefficient exceeding the mean. The critical point signal is determined when the sea level change rate in a short period of time (no more than 3 months) exceeds 3 times the average change rate of the previous 5 years. This step utilizes the multi-resolution characteristics of wavelet analysis to effectively separate the multi-scale components of sea level change, especially capturing nonlinear changes and mutation signals that are difficult to identify with traditional linear models, providing key feature information for sea level rise prediction. The mathematical expression of the continuous wavelet transform is: Where W f (a, b) are wavelet coefficients; f(t) is the time series of sea level height; ψ * is the conjugate complex number of the wavelet basis function; a is the scale parameter, which controls the expansion and contraction of the wavelet; b is the translation parameter, which controls the position of the wavelet. The wavelet energy spectrum calculation formula is: Where E(a) is the wavelet energy spectrum at scale a. The critical point determination formula is: C p ={b||W f (a, b)|>μ a +2σ a}; Where, C p is the critical point set; μ a is the mean value of the wavelet coefficients at scale a; σ a is the standard deviation of the wavelet coefficients at scale a.

[0080] The specific implementation of step S03 is to build a nonlinear prediction model for sea level rise based on a deep learning neural network. First, a hybrid architecture of a long short-term memory network (LSTM) and a convolutional neural network (CNN) is constructed. The number of LSTM units is set to 64 to 128, and the CNN adopts a 3 to 5-layer structure with a convolution kernel size of 3×3. Then, historical sea level observation data (at least 30 years of continuous records), satellite altimeter data, temperature and precipitation forecasts output by the global climate model (GCM), regional geological subsidence rate and other multi-source parameters are input. Then, the model's ability to capture long-term trends and sudden events is enhanced through the attention mechanism, and the depth of the attention layer is set to 2 to 3 layers. Finally, multiple prediction time scales are set, including short-term (1 to 5 years), medium-term (5 to 20 years) and long-term (20 to 100 years) prediction outputs. The model training adopts a batch size of 64, an initial learning rate of 0.001 and uses cosine annealing scheduling, with 200 to 500 training rounds. The purpose of this step is to build a prediction model that can effectively capture the nonlinear characteristics of sea level changes, achieve scientific predictions of future sea level rise trends, and provide basic data support for risk assessment. The input feature vector of the prediction model is expressed as: X t =[H t-k：t , T t：t+m , P t：t+m , S t , W t-j：t ]; where X t is the input feature vector at time point t; H t-k：t is the historical sea level sequence of the past k time steps; T t：t+m is the temperature prediction sequence for the next m time steps; P t：t+m is the precipitation forecast sequence for the next m time steps; S t is the regional geological subsidence rate; W t-j：t is the wavelet decomposition feature of the past j time steps. The model prediction output is expressed as: Where, is the predicted sea level height in the next n time steps; F θ is a neural network model function with parameter θ. The mathematical expression of the attention mechanism is: e i =v T tanh(W1h i +W2q); Where, α i is the attention weight of the i-th time step; e i is the attention score; h i is the hidden state at the i-th time step; q is the query vector; v, W1, W2 are learnable parameters; c is the context vector used for prediction.

[0081] The specific implementation of step S04 is to construct a regional flooding risk network model using the minimum cut maximum flow algorithm. First, the study area is discretized into a grid structure. The grid size is determined by the complexity of the terrain, usually 10 to 50 meters. Then, a directed network graph is constructed. The nodes represent terrain elevation points, and the capacity is set as the difference between the elevation value of the point and the predicted sea level. Negative values ​​indicate flooded areas. Next, edge weights are defined as the smoothness of the water flow channel between adjacent nodes. The calculation formula is the ratio of the elevation difference between adjacent nodes to the distance multiplied by the surface roughness coefficient. The surface roughness coefficient is determined by the land use type, such as 0.01 to 0.03 for urban construction areas, 0.03 to 0.05 for farmland, and 0.05 to 0.10 for forests. Finally, the Ford-Fulkerson algorithm or the Dinic algorithm is used to solve the maximum flow problem and identify the minimum cut set in the network. The nodes connected by the edges in the minimum cut set constitute the boundaries of the potential flooding area. The purpose of this step is to use the minimum cut maximum flow algorithm in graph theory to accurately simulate the path of seawater intrusion, locate the weak links and key vulnerabilities of the protection system, and provide a spatial analysis basis for risk assessment. The node capacity in the network graph is defined as: C(v) = E(v) - H sea ; Where C(v) is the capacity of node v; E(v) is the elevation value of node v; H sea To predict sea level height. The edge weight calculation formula is: Where W(u, v) is the weight of the edge between nodes u and v; D(u, v) is the Euclidean distance between nodes u and v; and R(LandType) is the surface roughness coefficient corresponding to the land type. The mathematical expression of the maximum flow minimum cut problem is:

[0082] maxf=max∑ e∈out(s) f(e);

[0083]

[0084] Where f is the total network flow; f(e) is the flow on edge e; c(e) is the capacity of edge e; s is the source (source of seawater); t is the sink (virtual high point); in(v) and out(v) represent the sets of edges entering and leaving node v, respectively; and E is the set of all edges in the graph.

[0085] The specific implementation of step S05 is to establish a regional vulnerability scoring system. First, a scoring index system is determined, including five dimensions: terrain height distribution (weight 0.25), protective facility status (weight 0.20), land use type (weight 0.15), population density (weight 0.20), and economic value (weight 0.20). Then, each indicator is standardized, converting the original value into a score range of 0 to 10. The terrain height standard score is calculated as 10 × (current elevation - predicted sea level) / (regional highest point - predicted sea level), and the value is 0 if it is less than 0. Then, a basic score is assigned according to land use type, such as 8 to 10 points for urban residential areas, 7 to 9 points for industrial areas, 5 to 7 points for farmland, 3 to 5 points for forest land, 2 to 4 points for grassland, and 0 to 2 points for unused land. Finally, the weighted values ​​of each indicator are combined to calculate the vulnerability index, which is divided into five levels: very low (0 to 2), low (2 to 4), medium (4 to 6), high (6 to 8), and very high (8 to 10). The purpose of this step is to establish a quantitative standard for regional vulnerability through a comprehensive scoring of multiple indicators, identify sensitive areas susceptible to sea level rise, and provide an important basis for comprehensive risk assessment. The vulnerability index calculation formula is: Where VI is the vulnerability index; w i is the weight coefficient of the i-th indicator; S i is the standardized score of the ith indicator. The formula for calculating the standard score of terrain height is:

[0086]

[0087] Where S1 is the standard score of terrain height; E is the current elevation; H sea To predict the sea level; E max is the elevation of the highest point in the area. The formula for calculating the standard score of the protective facilities condition is: Where, S2 is the standard score of the protection facilities; P is the current status assessment value of the protection facilities; P min is the minimum assessment value of protective facilities; P max The highest assessment value of protective facilities. The land use type standard score is directly assigned according to the classification: S3 = LandTypeScore(LandType); where S3 is the land use type standard score; LandTypeScore is the land use type score mapping function. The population density standard score is calculated as follows: Where S4 is the standard score of population density; PD is the regional population density (person / km2); min is the minimum population density in the study area; PD max is the maximum population density in the study area. The formula for calculating the economic value standard score is: Where S5 is the standard score of economic value; EV is the economic value per unit area of ​​the region; EV min EV is the economic value of the smallest unit area in the study area; max It is the maximum economic value per unit area in the study area.

[0088] The specific implementation of step S06 is to integrate the model prediction results and the vulnerability index using the Bayesian network integration analysis method. First, a Bayesian network structure is constructed, with nodes including key factors such as predicted sea level height, proportion of flooded area, status of protective facilities, population impact ratio, and economic loss estimation. Then, a conditional probability table (CPT) is set between nodes based on historical data and expert experience. For example, if the sea level rise is greater than 50 cm, the probability thresholds for severe flooding in areas with different vulnerability levels are: 0.05 for extremely low vulnerability areas, 0.15 for low vulnerability areas, 0.35 for medium vulnerability areas, 0.65 for high vulnerability areas, and 0.85 for extremely high vulnerability areas. Then, the sea level rise prediction result from step S03 and the vulnerability index from step S05 are input as observational evidence, and the posterior probability distribution is calculated using the belief propagation algorithm. Finally, a comprehensive risk assessment matrix is ​​generated, with the horizontal axis representing different sea level rise scenarios (such as RCP2.6, RCP4.5, RCP8.5, etc.) and the vertical axis representing the assessment unit (grid or administrative division). The matrix elements are the comprehensive risk scores (0-100) of the regional units under the corresponding scenarios. The purpose of this step is to use the uncertainty reasoning ability of the Bayesian network to scientifically integrate the sea level rise prediction and regional vulnerability assessment results to form a comprehensive risk quantification indicator. The conditional probability of the node in the Bayesian network is defined as: Where, X i is a node variable; Pa(X i ) is node X i The parent node set of the comprehensive risk assessment matrix is ​​calculated as follows: M(r, s) = 100 × P(Inundation = High | SeaLevel = s, VI = VI r ); where M(r, s) is the value of the element in the rth row and sth column of the comprehensive risk assessment matrix, representing the comprehensive risk score of regional unit r under sea level rise scenario s; P(Inundation=High|SeaLevel=s,VI=VI r ) is a given sea level rise scenario s and vulnerability index VI r The posterior probability of severe flooding in the area under the condition. The message passing formula of the belief propagation algorithm is:

[0089]

[0090] Where, For node X i Passed to node Xj Message;φ(X i , X j ) is node X i and node X j The potential function between N(X i ) is node X i The set of neighbor nodes.

[0091] The specific implementation of step S07 is to calculate the change matrix of the comprehensive risk assessment matrix. First, determine the time series sampling point set, usually selecting key time nodes such as 2030, 2050, 2070, and 2100; then generate the corresponding comprehensive risk assessment matrix M for each sampling time point. t ; Then calculate the matrix element difference ΔM=M between adjacent time points t1 and t2 t2 -M t1 , obtain the initial change matrix; then calculate the change rate matrix R = ΔM / (t2-t1), which reflects the risk change speed per unit time; then perform the second-order difference in the time dimension to obtain the change acceleration matrix A = (R t+1 -R t ) / Δt; Finally, identify the significant change points in the change acceleration matrix, set the threshold to 2.5 times the standard deviation of the change acceleration, and mark these points as trend inflection points. The purpose of this step is to capture the temporal evolution characteristics of risk through matrix operations, especially to identify the acceleration or deceleration trend of risk changes, and provide a time series dynamic analysis basis for risk management decisions. The change matrix calculation formula is: ΔM(r, s) = M t2 (r, s)-M t1 (r, s); where ΔM(r, s) is the element value in the rth row and sth column of the variation matrix; M t1 (r, s) and M t2 (r, s) are the values ​​of the elements in the rth row and sth column of the comprehensive risk assessment matrix at time points t1 and t2, respectively. The formula for calculating the rate of change matrix is: Where R(r, s) is the value of the element in the rth row and sth column of the rate of change matrix, which represents the speed of risk change per unit time. The calculation formula for the change acceleration matrix is: Where A(r, s) is the value of the element in the rth row and sth column of the acceleration matrix; R t (r, s) and R t+1 (r, s) are the element values ​​of the rth row and sth column in the rate of change matrix of two adjacent time intervals; Δt is the time interval. The trend inflection point identification condition is:

[0092]

[0093] Where, TurningPoint(r, s) is the trend turning point labeling matrix, 1 indicates an inflection point, and 0 indicates a non-inflection point; μ A is the mean value of the changing acceleration matrix A; σ A is the standard deviation of the acceleration matrix A.

[0094] The specific implementation of step S08 involves generating a comprehensive assessment report using the Coastline Adaptability Gating Model. First, the comprehensive risk assessment matrix generated in step S06 and the change score matrix generated in step S07 are input into the Coastline Adaptability Gating Model. A risk level distribution map is then generated based on the model's calculation results, using a red, orange, yellow, and green color system to represent extremely high, high, medium, and low risk areas. Next, the top 10% of the comprehensive risk scores are selected as a list of key vulnerabilities, including spatial coordinates, risk scores, major risk factors, and protection recommendations. A time series risk evolution trend chart is then generated, with time plotted on the horizontal axis and risk scores plotted on the vertical axis, showing the trajectory of risk changes across different regions. Finally, a risk rating classification table is generated, classifying regions by risk urgency and formulating differentiated protection strategy recommendations. The rating classification uses a four-level scale: Level I (extremely urgent, action required within 5 years), Level II (urgent, action required within 5-15 years), Level III (medium, action required within 15-30 years), and Level IV (low urgency, consideration after 30 years). The purpose of this step is to integrate the results of the previous analysis and generate an intuitive and clear risk assessment report to provide a scientific basis and action guide for coastal zone management and protection decisions. The regional adaptability gating weight function in the coastline adaptability gating model is expressed as:

[0095]

[0096] In the formula, G(r, s) is the regional adaptive gating weight function value; B(r, s) is the comprehensive balance value; T1 and T2 are the interval boundary thresholds, usually T1 is set to 30 and T2 is set to 70; G conservative , G balanced and G aggressive These are conservative, balanced, and aggressive weight adjustment functions. The comprehensive balance value calculation formula is: B(r, s) = w1·R(r, s) + w2·TC r +w3·SEV r ; In the formula, B(r, s) is the comprehensive balance value; R(r, s) is the risk change rate in the change matrix; TC r is the terrain complexity index of region r; SEV r is the distribution value of the socioeconomic vulnerability of region r; w1, w2 and w3 are weight coefficients, and satisfy w1+w2+w3=1, generally taking w1=0.4, w2=0.3, w3=0.3. The conservative weight adjustment function is: G conservative(r, s) = α·B(r, s); where α is the linear growth coefficient, usually ranging from 0.01 to 0.05. The balanced weight adjustment function is: Where β is the steepness coefficient of the S-curve, which is usually between 0.1 and 0.2. The radical weight adjustment function is: Where γ is the basic weight coefficient, which is usually between 0.5 and 0.7; δ is the exponential growth coefficient, which is usually between 0.05 and 0.1.

[0097] The calculation formula of regional terrain complexity index is:

[0098]

[0099] Where, TC r is the terrain complexity index of region r; σ E and μ E are the standard deviation and mean of regional elevation respectively; S i is the slope value of the i-th grid in the region; n is the total number of grids; E max and E min are the highest and lowest elevations in the area respectively; A r is the area of ​​the region; w elev 、w slope and w rough is the weight coefficient, and satisfies w elev +w slope +w rough =1, generally take w elev =0.4, w slope =0.4, w rough =0.2.

[0100] The calculation formula for socioeconomic vulnerability distribution is:

[0101]

[0102] Where SEV r is the socioeconomic vulnerability distribution value of region r; PD r is the population density of region r; IV r is the infrastructure value of region r; EA r is the economic activity intensity of region r; LU r Score the land use type of region r; PD min , PD max IV min IV max , EA min , EA max are the minimum and maximum values ​​of the corresponding indicators respectively; w pop 、winfra 、w econ and w land is the weight coefficient, and satisfies w pop +w infra +w econ +w land =1, generally take w pop =0.3, w infra =0.25, w econ =0.25, w land =0.2.

[0103] The detailed structure of the coastline adaptive gating model is a neural network architecture consisting of an input layer, a multi-head temporal attention layer, a bidirectional autoregressive encoding layer, a dynamic gating integration layer, and an output layer. The input layer receives three types of data: climate scenario data (including temperature changes, precipitation patterns, and sea level rise forecasts under different RCP scenarios), terrain feature vectors (including terrain parameters such as elevation distribution, slope, and orientation), and socioeconomic indicators (including population density, GDP, and infrastructure value). The mathematical representation of the input data is: X input =[X climate , X terrain , X socioeco ];

[0104] Where, X input Input data vector for the model; X climate is the climate scenario data vector; X terrain is the terrain feature vector; X socioeco is the socioeconomic indicator vector. The climate scenario data vector is represented as: X climate =[T RCP , P RCP , SLR RCP ];

[0105] Where, T RCP is the temperature change sequence under the RCP scenario; P RCP is the precipitation pattern sequence under the RCP scenario; SLR RCP is the sea level rise prediction sequence under the RCP scenario. The terrain feature vector is expressed as: X terrain =[E dist , S dist , A dist ];

[0106] Where, E dist is the elevation distribution characteristic; S dist is the slope distribution characteristic; A dist is the orientation distribution characteristic. The socioeconomic indicator vector is expressed as: X socioeco =[PD, GDP, IV];

[0107] Where PD is population density; GDP is gross domestic product; IV is infrastructure value.

[0108] The dynamic gate integration layer dynamically adjusts the weights of each feature channel according to the change matrix in the comprehensive risk assessment matrix, and uses the regional adaptive gate weight function to adjust the parameters. The calculation formula for dynamic gate integration is:

[0109] Where Z is the output of the dynamic gate integration layer; G(r, s) i is the gating weight of feature channel i, which is calculated by the regional adaptive gating weight function; F i is the input feature of feature channel i; n is the total number of feature channels.

[0110] The output layer generates regional protection strategy recommendations and resource optimization allocation plans, including the selection of engineering protection measures, implementation timing arrangements, and investment priority rankings. The output is expressed as: Y = [Y strategy , Y timing , Y priority ];

[0111] Where Y is the model output; Y strategy Recommended protection strategy; Y timing Arrange the timing for implementation; Y priority Prioritize investments.

[0112] The detailed steps for establishing the training dataset for the coastline adaptive gating model are as follows: First, historical sea level change data for 25 typical coastline areas around the world are collected, with a time span of no less than 50 years. The data include monthly average sea level height, frequency and intensity of extreme events; then, inundation range maps and actual impact assessment reports for each region under different sea level rise scenarios are integrated, including more than 100 case studies; then, topographic change characteristics and socioeconomic response indicators before and after the implementation of regional protection projects are extracted, such as changes in the frequency of flood events and the reduction ratio of economic losses before and after the construction of protection dams; then, a multi-scenarios dataset is constructed. A parallel case library with multiple regional time points, with a total data volume of no less than 500 typical cases; then data normalization processing is performed, unifying indicators of different dimensions to the range of 0 to 1, and performing spatiotemporal alignment to ensure comparability between different cases; then marking the protective measures effectiveness score (0 to 10 points) and cost-effectiveness ratio of each case. The effectiveness score is determined based on the actual effect of disaster prevention and mitigation, and the cost-effectiveness ratio is calculated as the ratio of input cost to avoided loss; finally, the data is divided into a training set (70%), a validation set (15%), and a test set (15%) in chronological order to ensure the scientific nature of model training and evaluation. Data normalization uses the minimum-maximum normalization method: Where, X normis the normalized data; X is the original data; X min and X max are the minimum and maximum values ​​of the data respectively. The cost-effectiveness ratio is calculated as: Where, CEB is the cost-effectiveness ratio; C invest is the input cost; L avoided To avoid losses.

[0113] To better understand and implement the present invention, Example 2, a specific application scenario, is provided below. Researchers selected a typical coastal region as their research object. This region has a total area of ​​approximately 256 square kilometers, a coastline of approximately 37 kilometers, and includes a variety of land use types, including urban built-up areas, industrial areas, farmland, and forestland. The region has a permanent population of approximately 150,000. The method of the present invention was applied to comprehensively assess the risk of this region under a sea level rise scenario.

[0114] First, construct a multi-dimensional data acquisition network, such as Figure 4 The multidimensional data collection network diagram is shown in Figure 1. Nine fixed tide gauges were deployed in the study area, with an average spacing of approximately 4.1 kilometers. Twenty-five GNSS surface deformation monitoring stations were set up, with a density of one per 10 square kilometers. A regional digital elevation model with a 0.5-meter resolution was acquired. Quarterly remote sensing imagery data from 2000 to 2022 was integrated. Data from various sources were unified into the WGS84 coordinate system using a data fusion algorithm to form a temporally and spatially consistent sea level change monitoring dataset. The weighting coefficients for each data source are shown in Table 1.

[0115] Table 1 Multi-source data fusion weight coefficient table

[0116] Data Source Historical observation standard deviation (m) Weight coefficient Tide station data 0.032 0.412 Satellite altimeter data 0.055 0.196 GNSS surface deformation data 0.008 0.657 SAR interferometry data 0.021 0.476 Optical remote sensing images 0.087 0.140 drone aerial images 0.043 0.285

[0117] Second, we used multiscale wavelet analysis to perform nonlinear decomposition on the sea level change time series data from 2000 to 2022. Morlet wavelets were then used to transform the original time series. Five critical point signals were identified, occurring in 2004, 2008, 2012, 2016, and 2020. The rates of sea level change at these critical points are significantly higher than the normal rate of change. The characteristics of the critical points and their wavelet coefficients are shown in Table 2.

[0118] Table 2 Characteristics of critical points of sea level change

[0119] Critical point time Wavelet coefficient values Change rate (mm / year) Average change rate in the first five years (mm / year) Change multiple June 2004 2.87 8.2 2.7 3.04 August 2008 3.21 9.5 3.1 3.06 May 2012 3.35 10.3 3.3 3.12 July 2016 4.12 12.6 3.8 3.32 September 2020 5.78 18.5 4.5 4.11

[0120] Figure 6 and Figure 7 The time series of sea level changes from 2000 to 2022 and its wavelet analysis results are shown. Figure 6The relative change curve of sea level is shown, marking the five critical points mentioned in the embodiment (2004, 2008, 2012, 2016 and 2020). Figure 7 This is a heat map of the wavelet energy spectrum, showing energy distribution at different time scales (from seasonal to decadal), clearly demonstrating the abnormal concentration of energy at critical points. Multi-scale wavelet analysis clearly identifies critical points in the sea level change time series. These critical points appear as areas of abnormal energy concentration in the wavelet energy spectrum, especially at the 2- to 7-year scale.

[0121] Third, a nonlinear sea level rise prediction model was constructed based on a deep learning neural network. This model employed a hybrid architecture consisting of a two-layer LSTM (64 units per layer) and a three-layer CNN (with a 3×3 convolution kernel). Input parameters included historical sea level observations, temperature and precipitation forecasts from the IPCC global climate model, and regional geological subsidence rates. Table 3 shows the model's prediction results under the RCP2.6, RCP4.5, and RCP8.5 scenarios.

[0122] Table 3 Predicted values ​​of sea level rise under different scenarios

[0123]

[0124]

[0125] Fourth, the minimum cut maximum flow algorithm is used to construct a regional flooding risk network model, such as Figure 5 The following figure shows a schematic diagram of the inundation risk network model using the minimum cut maximum flow algorithm. The study area was discretized into a 20 m x 20 m grid structure, and a directed network graph was constructed. Edge weights were set based on the ratio of the elevation difference to the distance between adjacent nodes multiplied by the surface roughness coefficient. The Dinic algorithm was used to calculate the minimum cut set, identifying 32 key vulnerable points and four major potential inundation pathways. The surface roughness coefficients for different land use types are shown in Table 4.

[0126] Table 4 Surface roughness coefficient of different land use types

[0127] Land use type Surface roughness coefficient Urban construction area 0.018 Industrial Zone 0.022 farmland 0.037 woodland 0.085 grassland 0.059 wetlands 0.045 bare land 0.025 waters 0.010

[0128] Fifth, a regional vulnerability scoring system was established, calculating a vulnerability index based on terrain height distribution, protective infrastructure status, land use type, population density, and economic value. The study area was divided into 65 assessment units, and the distribution of vulnerability indices for each unit is shown in Table 5.

[0129] Table 5 Distribution of vulnerability index in the study area

[0130] Vulnerability Level Vulnerability Index Range Number of assessment units Area share (%) Population share (%) Very low 0～2 7 12.5 5.8 Low 2～4 18 31.2 22.7 medium 4～6 22 28.7 33.4 high 6～8 12 18.9 27.6 Very high 8～10 6 8.7 10.5

[0131] Sixth, we applied a Bayesian network ensemble analysis method to integrate the predictions from the nonlinear sea level rise prediction model with the vulnerability index from the regional vulnerability scoring system to generate a comprehensive risk assessment matrix. Taking the 2050 RCP4.5 scenario as an example, the risk score distribution for the study area is shown in Table 6.

[0132] Table 6. Risk score distribution of study areas under the RCP4.5 scenario in 2050

[0133] Risk Level Risk score range Number of assessment units Area share (%) Population share (%) Low risk 0～25 19 32.6 24.3 Medium risk 25～50 25 36.5 38.7 High risk 50～75 14 20.7 26.2 Extremely high risk 75～100 7 10.2 10.8

[0134] Seventh, we calculated the change matrix of the comprehensive risk assessment matrix. By comparing the risk assessment matrices for 2030, 2050, 2070, and 2100, we extracted the risk change rate and trend characteristics. The results showed that the risk change acceleration matrix for the study area had eight significant trend inflection points, six of which were points of accelerated risk growth and two of which were points of decelerated risk growth. The statistical characteristics of the change rate matrix are shown in Table 7.

[0135] Table 7 Risk change rate characteristics in different time periods

[0136]

[0137]

[0138] Figure 8 and Figure 9 The sea level rise prediction and risk assessment results in the embodiment are shown. Figure 8 The projected sea level rise curves from 2022 to 2100 under the three scenarios of RCP2.6, RCP4.5 and RCP8.5 are shown, including 95% confidence intervals, and the risk change rate in different time periods is also shown on the secondary axis. Figure 9 The data shows the distribution of risk areas in different years (2030, 2050, 2070, and 2100), demonstrating the changes in the area share of low-risk, medium-risk, high-risk, and extremely high-risk areas, as well as the growth trend in the number of critical vulnerabilities. The sea level rise predictions under different RCP scenarios show that the rate of rise under the RCP8.5 scenario is significantly higher than that under other scenarios, and the upward trend is accelerating. At the same time, as sea levels rise, the risk distribution pattern within the study area also changes significantly, with the area share of high-risk and extremely high-risk areas gradually increasing, and the number of critical vulnerabilities also increasing accordingly.

[0139] Finally, a comprehensive assessment report on regional sea level rise risks was generated using the coastline adaptability gating model, which identified 12 key vulnerable points and 3 risk hotspots, and proposed differentiated protection strategy recommendations.

[0140] Traditional sea level rise risk assessment methods mainly rely on linear prediction models and static inundation analysis, which cannot effectively capture the nonlinear characteristics and sudden events of sea level changes, have limited assessment accuracy, and cannot provide dynamic risk evolution trends. The multi-source data fusion and multi-scale wavelet analysis methods adopted by the present invention, combined with deep learning neural networks and minimum cut maximum flow algorithms, significantly improve the accuracy of sea level rise predictions and the spatial precision of risk assessments. Experimental results show that compared with traditional methods, the accuracy of the present invention in identifying critical point signals is improved by 15.8%, the root mean square error in predicting sea level rise trends is reduced by 18.3%, and the spatial accuracy in delineating the boundaries of potential inundation areas is improved by 12.7%. At the same time, by introducing the variable score matrix and the coastline adaptive gating model, the present invention can provide dynamic risk evolution trends and differentiated protection strategies, providing a more scientific basis for coastal management and protection decisions.

[0141] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 8, 9 and 10 below.

[0142] Table 8 Variable Explanation Table (Part 1)

[0143]

[0144]

[0145] Table 9 Variable Explanation Table (Part 2)

[0146]

[0147]

[0148] Table 10 Variable Explanation Table (Part 3)

[0149]

[0150] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A sea level rise risk assessment method based on multi-source data fusion, characterized by: include: Construct a multidimensional data acquisition network to obtain the basic characteristics of sea level rise; use multi-scale wavelet analysis to perform nonlinear decomposition of sea level change time series data, extract different frequency components and identify critical point signals; construct a nonlinear sea level rise prediction model based on deep learning neural networks; use the minimum cut maximum flow algorithm to construct a regional flooding risk network model, calculate the minimum cut set to determine key vulnerable points and potential flooding area boundaries; establish a regional vulnerability scoring system and calculate the vulnerability index; and apply the Bayesian network integration analysis method to generate a comprehensive risk assessment matrix. Calculate the change matrix of the comprehensive risk assessment matrix and extract the risk change rate and trend characteristics; Generate a comprehensive regional sea level rise risk assessment report using a coastline adaptability gating model.

2. The sea level rise risk assessment method according to claim 1, characterized in that: The multi-dimensional data acquisition network refers to a multi-source heterogeneous data acquisition system composed of a satellite remote sensing system, a tide station monitoring system, a GNSS surface deformation monitoring station and an unmanned aerial survey system.

3. The sea level rise risk assessment method according to claim 2, characterized in that: The minimum cut maximum flow algorithm is a method for solving optimization problems in graph theory. In sea level rise risk assessment, the terrain is regarded as a network graph, elevation is used as node capacity, and water flow channels are used as edges. By calculating the minimum cut set, the weak links that are most likely to break through the defense line are determined, and key vulnerable points and the boundaries of potential flooding areas are located.

4. The sea level rise risk assessment method according to claim 3, characterized in that: The vulnerability index is a comprehensive score obtained by weighted calculation of terrain height distribution, protection facility status, land use type, population density and economic value.

5. The method for assessing the risk of sea level rise according to claim 4, wherein: The comprehensive risk assessment matrix is ​​a two-dimensional tabular data structure, where the horizontal axis represents different sea level rise scenarios and the vertical axis represents different regional units.

6. The method for assessing the risk of sea level rise according to claim 5, wherein: The change score matrix is ​​a matrix generated by calculating the differences in element values ​​of the comprehensive risk assessment matrix between different time nodes.

7. The sea level rise risk assessment method according to claim 6, characterized in that: The coastline adaptive gating model is a comprehensive analysis framework that integrates a multi-layer perceptron and a temporal attention mechanism.

8. The sea level rise risk assessment method according to claim 7, characterized in that: The coastline adaptive gating model adopts a regional adaptive gating weight function to dynamically adjust parameter weights according to risk changes.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the sea level rise impact risk assessment method based on multi-source data fusion according to any one of claims 1 to 8.

10. A sea level rise risk assessment system based on multi-source data fusion, characterized in that: The computer-readable storage medium according to claim 9 is included, wherein the system is any one of a computer, a server, and a single-chip microcomputer, the computer-readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

Images