AI-based marine disaster prevention monitoring and early warning system

Abstract

The invention relates to the technical field of marine disaster monitoring and early warning, in particular to an AI-based marine disaster prevention monitoring and early warning system. Comprising a multi-source data acquisition module for acquiring multi-dimensional data through a sensor network, remote sensing data access and a meteorological data interface; the data preprocessing and fusion module is used for carrying out cleaning, space-time calibration and feature fusion on the collected data; the artificial intelligence analysis module is used for realizing accurate analysis and prediction of different levels of storm surge by using a multi-scale prediction model cluster unit; the risk assessment and early warning generation module is used for calculating risk indexes according to a related index system and dividing early warning grades; and the interaction and release module is used for realizing visual display and multi-channel release of the early warning information. According to the invention, advantages of multi-source data and artificial intelligence technology are integrated, and real-time and accurate monitoring and early warning can be carried out on storm surge and coast submerging conditions.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of marine disaster monitoring and early warning, in particular to an AI-based marine disaster prevention monitoring and early warning system. BACKGROUND

[0002] Storm surge and coastal inundation are one of the main natural disasters faced by coastal areas, which are strong in suddenness and destructive, and seriously threaten the safety of coastal cities, infrastructure and the lives and property of residents. Traditional storm surge warning systems mainly rely on physical model simulation or single data source analysis: The physical model needs to accurately set the boundary conditions, has high computational complexity and poor real-time performance, and is difficult to adapt to complex terrain and sudden weather changes; Single data source (such as water level station, weather station) has limited coverage and insufficient data dimension, and cannot fully reflect the storm surge evolution process; Manual analysis or traditional statistical model has insufficient recognition ability for nonlinear and high dynamic disaster characteristics, and the warning accuracy and advance cannot be guaranteed; The existing system lacks fine simulation of the dynamic process of coastal inundation and cannot provide regional risk distribution prediction.

[0003] With the development of sensor technology, remote sensing technology and big data technology, multi-source heterogeneous data (such as satellite remote sensing, radar monitoring, Internet of Things sensors, and meteorological forecast data) provide rich information for disaster monitoring, but traditional technology cannot efficiently process massive data and mine potential correlations, and it is urgent to introduce artificial intelligence technology to improve the warning capability. SUMMARY

[0004] In view of the problems in the above background technology, the application provides an AI-based marine disaster prevention monitoring and early warning system.

[0005] The AI-based marine disaster prevention monitoring and early warning system comprises: A multi-source data acquisition module comprising an Internet of Things sensor network, a remote sensing data access interface and a meteorological data interface, for real-time acquisition of hydrological, meteorological, topographical and remote sensing image data; A data preprocessing and fusion module comprising a data cleaning unit, a space-time calibration unit and a feature fusion unit, for cleaning, calibration and feature-level fusion of multi-source data; An artificial intelligence analysis module comprising a multi-scale prediction model cluster unit, a model optimization unit and an uncertainty analysis unit, the multi-scale prediction model cluster comprising a short-term prediction model based on LSTM or TCN, a spatial diffusion model based on CNN or GNN, and a disaster recognition model based on a target detection algorithm, the model optimization unit updating the model online using real-time monitoring data and adapting different regional geographical features through transfer learning; The risk assessment and early warning generation module is configured to combine the prediction results, terrain data, and regional disaster-bearing body information, assess the flooding risk, and generate graded early warning information. The interaction and release module is configured to display the early warning results on a visual display platform and push the early warning information to users.

[0006] Preferably, the Internet of Things sensor network adopts a distributed self-organizing architecture, including: The edge computing node is deployed at a coastal monitoring site and is responsible for local data preprocessing and feature extraction. The regional aggregation node uses Mesh network technology to realize self-organizing network communication between adjacent nodes and ensure data transmission reliability. The central control node receives regional data and performs global fusion analysis. When a certain monitoring node communication is interrupted, the system automatically reconstructs the network topology, relays the data transmission through adjacent nodes, and ensures the robustness of the monitoring network.

[0007] Preferably, the multi-scale prediction model cluster unit includes sub-models that work cooperatively, specifically including: The short-term prediction model predicts water level and wind speed parameters in the future 1-6 hours based on historical hydro-meteorological data. The spatial diffusion model simulates the spatial diffusion process of storm surges in the coastal zone in combination with terrain data and coastline features. The disaster identification model performs real-time identification and segmentation of flooded areas in remote sensing images.

[0008] Preferably, the risk assessment and early warning generation module is configured to: Establish a regional risk assessment index system including flooding depth, population density, and infrastructure importance. Generate a risk index through weighted calculation: ; wherein, is the risk index, is the flooding depth, is the population density, is the facility importance coefficient, , , is a weight coefficient determined through historical disaster data training; According to the comparison between the risk index and the preset threshold, generate blue, yellow, orange, and red graded early warnings.

[0009] Preferably, the uncertainty analysis unit quantifies the uncertainty of the prediction results based on Monte Carlo simulation or Bayesian inference, generates a confidence interval, and provides a probability reference for early warning decision-making.

[0010] Preferably, the visualization platform displays the storm surge evolution process, flooding risk distribution and warning area in real time through GIS maps, supports multi-scale zooming, time slider interaction and dynamic deduction of the flooding process.

[0011] The AI-based marine disaster prevention monitoring and warning method comprises the following steps: Real-time collection of multi-source data: hydrological, meteorological, topographical and remote sensing image data are obtained through a sensor network, remote sensing equipment and a meteorological interface; Data preprocessing and fusion: the collected data are cleaned, time and space calibrated and feature fused to construct a standardized data set; Artificial intelligence analysis and prediction: The preprocessed data are input into a multi-scale prediction model cluster to obtain short-term prediction results of water level and wind speed related parameters and storm surge spatial diffusion simulation; Real-time analysis of remote sensing images is performed by using a disaster recognition model to extract current flooding area features; Model parameters are updated by a model optimization unit to improve prediction accuracy; Risk assessment and warning generation: The prediction results, topographical data and regional disaster-bearing body information are combined to calculate regional risk indexes; Warning information is generated according to the risk indexes to determine the warning level and the affected range; Warning release and interaction: the warning results are displayed through a visualization platform, and the warning information is released through multiple channels to support user interactive queries.

[0012] Preferably, in the step of artificial intelligence analysis and prediction, the short-term prediction model adopts an LSTM or TCN network structure, the spatial diffusion model adopts a CNN or GNN network structure, and the disaster recognition model adopts a YOLO, SSD or improved UNet model.

[0013] Preferably, in the step of warning release and interaction, the warning information is synchronously released through multiple channels such as short messages, broadcasts and emergency platforms, and differentiated warning contents are customized for different user groups.

[0014] Preferably, the method further comprises a system self-learning and optimization step: Historical warning data and actual disaster conditions are collected to construct a verification data set; The prediction results are compared with the actual conditions to evaluate the model performance; Model parameters and risk assessment index systems are adjusted according to the evaluation results to continuously improve the system warning accuracy.

[0015] Compared with the prior art, the present application has the following advantages: Propose a deep fusion architecture of multi-source data (Internet of Things sensors, remote sensing images, and weather forecasts) and artificial intelligence models to break through the limitations of a single data source; Build a multi-scale prediction model cluster, combining time series prediction and spatial diffusion models to achieve fine-grained prediction of storm surge in the "time-space" dimension; Introduce real-time model optimization and uncertainty analysis mechanisms to improve the system's adaptability to complex geographical environments and sudden disasters; Based on the dynamic risk assessment method of regional disaster-bearing body characteristics, realize the precision and individualization of early warning information. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The system architecture diagram of the AI-based marine disaster prevention monitoring and early warning system proposed in the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0018] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

[0019] Embodiment 1: Reference Figure 1 The AI-based marine disaster prevention monitoring and early warning system provided in the present embodiment includes a multi-source data acquisition module, a data preprocessing and fusion module, an artificial intelligence analysis module, a risk assessment and early warning generation module, and an interaction and publishing module. Each module cooperates with each other to build a real-time, high-precision, and self-adaptive marine disaster monitoring and early warning system. The specific implementation of each module is described in detail below.

[0020] The multi-source data acquisition module includes an Internet of Things sensor network, a remote sensing data access interface, and a meteorological data interface, which are used to acquire hydrological, meteorological, topographical, and remote sensing image data in real time.

[0021] Specifically, the multi-source data acquisition module is deployed and data is collected, including: Sensor network deployment: A pressure water level gauge is deployed every 5-10 kilometers along the coastline to measure the tidal level change in real time, with a measurement accuracy of ±0.5 cm and a sampling frequency of 1 minute / time. Wave buoys are deployed in offshore areas, equipped with three-axis accelerometers and GPS positioning systems, to measure significant wave height, period, and direction, with a 5-minute data transmission interval. Laser rangefinders and video monitoring equipment are deployed in key protection areas to assist in monitoring shoreline changes and inundation areas through image recognition technology. It should be noted that all sensor equipment has self-calibration function, which detects abnormal values by comparing adjacent site data.

[0022] Remote sensing data access: Satellite remote sensing data: Receive Sentinel-1 SAR satellite images, use C-band synthetic aperture radar to penetrate clouds, and achieve all-weather sea surface monitoring; At the same time, access Gaofen-3 satellite data for fine inundation area identification; Unmanned aerial vehicle data: During disaster warning, deploy fixed-wing unmanned aerial vehicle groups to fly along preset routes, equipped with multispectral cameras and LiDAR equipment, to obtain 10cm resolution orthophoto and DEM data, flight height 500-1000m, coverage range 50km²; Data receiving frequency: satellite images every day 4 times, unmanned aerial vehicle data every 2 hours update once.

[0023] Weather data interface: Integrate global forecast system and ECMWF data to obtain 72-hour forecast data of typhoon path, pressure field, and wind field, with 1-hour time resolution and 0.25°x0.25° spatial resolution; Access local Doppler radar data to monitor typhoon eye position, intensity changes, and precipitation distribution in real time, with 10-minute data update frequency.

[0024] The data preprocessing and fusion module includes a data cleaning unit, a space-time calibration unit, and a feature fusion unit, which are used for cleaning, calibrating, and feature-level fusion of multi-source data.

[0025] Specifically, the data preprocessing and fusion module cleans, calibrates, and feature-level fuses multi-source data, including: Data cleaning unit: Outlier detection: Use the Isolation Forest algorithm to identify outliers in water level data, set the detection threshold to 3 times the standard deviation; Missing value processing: For short-term missing data (<30 minutes), use cubic spline interpolation to complete; For long-term missing data, combine adjacent site data and estimate by spatial and temporal Kriging interpolation method; Denoising: Apply wavelet transform to wind speed data to remove high-frequency noise and retain true wind speed variation characteristics.

[0026] Space-time calibration unit: Time synchronization: synchronize all sensor clocks using NTP protocol, error controlled within ±10ms; Spatial registration: register satellite images with 1:10000 scale basic geographic information data, using SIFT feature point matching algorithm, registration accuracy controlled within 2 pixels; Projection conversion: uniformly use WGS84 coordinate system and UTM projection to ensure the consistency of multi-source data in space.

[0027] Feature fusion unit: Data layer fusion: Kalman filter fusion of satellite remote sensing inverted sea surface height and tide station measured data, state vector containing water level and change rate, observation noise covariance matrix dynamically adjusted according to sensor accuracy; Feature layer fusion: extract texture features (gray level co-occurrence matrix GLCM), spectral features (NDWI index) of SAR images and flow field features output by numerical model, and construct multi-dimensional feature vector; Decision layer fusion: use Dempster-Shafer evidence theory to fuse the prediction results of different models, and calculate the trust degree of each evidence through basic probability assignment function (BPA).

[0028] The artificial intelligence analysis module includes a multi-scale prediction model cluster unit, a model optimization unit, and an uncertainty analysis unit.

[0029] Specifically, the multi-scale prediction model cluster unit includes a short-term prediction model based on LSTM or TCN, a spatial diffusion model based on CNN or GNN, and a disaster identification model based on target detection algorithm: The short-term prediction sub-model predicts water level and wind speed parameters in the future 1-6 hours based on historical hydro-meteorological data. In a preferred embodiment, the short-term prediction sub-model specifically includes: Network structure: use a bidirectional LSTM network, the input layer dimension is 12 (including the water level, wind speed, and air pressure data of the previous 12 hours), the hidden layer contains 3 LSTM layers (128 neurons per layer), and the output layer is the water level prediction value in the future 6 hours; Training optimization: use Adam optimizer, learning rate set to 0.001, batch size 64, and early stopping strategy to prevent overfitting; Training data: collect historical data during typhoons in the past 10 years, and divide them into training set, validation set, and test set in the ratio of 8:1:1; The spatial diffusion model simulates the spatial diffusion process of storm surge in the coastal zone by combining terrain data and coastline features. In a preferred embodiment, the spatial diffusion model specifically includes: Network structure: Based on graph convolutional network (GCN), the study area is discretized into a triangular mesh (average edge length 200 m), each node contains attributes such as elevation, roughness, initial water level, etc., and the edge weight represents the water flow connectivity between nodes; Training strategy: Two-stage training method of unsupervised pre-training (autoencoder learning node feature representation) and supervised fine-tuning (using historical inundation data); Physical constraints: Add continuity equation constraint term to the loss function to ensure that the model prediction conforms to the law of conservation of mass; Output results: Generate 100m×100m resolution inundation depth raster map, prediction time step is 15 minutes.

[0030] The disaster identification model identifies and segments the inundated area in the remote sensing image in real time, and in a preferred embodiment, the disaster identification model specifically includes: Network structure: Improved U-Net architecture, encoder uses ResNet50 pre-trained model to extract features, decoder introduces attention mechanism to enhance the recognition ability of the inundation boundary; Data augmentation: Rotate, flip, add noise, etc. Augment SAR images to expand training samples; Loss function: Combine cross-entropy loss and Dice loss to balance the class imbalance problem.

[0031] Specifically, the model optimization unit updates the model online using real-time monitoring data, and adapts to different regional geographic features through transfer learning, including: Online learning mechanism: Collect new monitoring data every hour, use sliding window (window size 72 hours) to update the training set, and incrementally train the model; Transfer learning strategy: For new warning areas, use the model parameters trained in existing areas for initialization, and only fine-tune the last few layers of the network; Adaptive weight adjustment: According to the real-time quality evaluation results of different data sources, dynamically adjust the weight coefficients of each sub-model in the fusion prediction.

[0032] Specifically, the uncertainty analysis unit quantifies the uncertainty of the prediction results based on Monte Carlo simulation or Bayesian inference, generates a confidence interval, and provides a probability reference for warning decision-making, including: Monte Carlo dropout: Turn on the dropout layer during prediction, perform 50 forward propagations, and estimate the prediction uncertainty by the variance of the output; Bayesian neural network: Bayesian transformation of the LSTM model, using variational inference to estimate the posterior distribution of the weights, generating a probability prediction interval; Uncertainty visualization: The inundation probability is represented by color transparency on the GIS platform, for example, the inundation area with 90% confidence interval is displayed with opaque red color, and the area with 50% confidence interval is displayed with semi-transparent red color.

[0033] The risk assessment and early warning generation module is used to combine the prediction results, terrain data and regional disaster-bearing body information, assess the inundation risk and generate graded early warning information, including: Risk assessment index system: Inundation depth grading: The inundation depth is divided into 4 levels (D1: 0-0.5m, D2: 0.5-1.5m, D3: 1.5-3m, D4: >3m); Population density: The population distribution data of 100m x 100m grid is adopted, which is derived from the latest population census and the fusion results of mobile phone signaling data; Facility importance: An evaluation system containing 12 types of infrastructure is established, and importance coefficients are assigned for different facility types; Risk index calculation: Weight determination: The analytic hierarchy process (AHP) is used to construct a judgment matrix, and 10 disaster experts are invited to score, and the weight coefficients are determined through consistency check; Final risk index: R = 0.45D + 0.35P + 0.20I, where D, P and I are the normalized inundation depth, population density and facility importance indexes respectively.

[0034] Early warning level division: Blue warning: R ∈ [0.2, 0.4), the predicted inundation depth is ≤0.5m, and it may affect low-lying areas; Yellow warning: R ∈ [0.4, 0.6), the predicted inundation depth is 0.5-1.5m, and it may inundate part of the roads and building ground floors; Orange warning: R ∈ [0.6, 0.8), the predicted inundation depth is 1.5-3m, which may cause large-scale flooding and residents in low-lying areas need to be relocated; Red warning: R ≥ 0.8, the predicted inundation depth is >3m, which may cause major disasters and immediate full-scale emergency response is required.

[0035] The interaction and release module is used to display the early warning results and push early warning information to users, including: Visualization display platform: Three-dimensional scene construction: WebGL three-dimensional map is developed based on Cesium.js, integrating high-precision DEM data, three-dimensional building models and inundation prediction results; Dynamic deduction function: Time slider control is supported, and users can view the dynamic change process of the inundation range within the next 24 hours; Information query function: Click on any location on the map to display detailed information such as inundation time, maximum depth, risk level, etc. Thematic map production: Automatically generate thematic products such as inundation risk zoning map, population impact distribution map, and infrastructure threat assessment map.

[0036] Multi-channel release unit: SMS push: Cooperate with operators to send customized warning SMS to mobile users in the warning area based on geographic location positioning, including risk level, expected impact time, and risk avoidance suggestions; Emergency broadcast system: Link with local radio stations to update warning information every 15 minutes, focusing on reporting changes and emergency response measures; API interface: Provide standardized data interfaces to government emergency management platforms, traffic management systems, maritime departments, etc., supporting secondary development and system integration.

[0037] Human-computer interaction interface: Threshold customization: Allow administrators to adjust the parameters and warning thresholds of the risk assessment model according to actual conditions; Historical query: Provide historical warning event query and playback functions, supporting comparative analysis of warning effects of different typhoon processes; Report generation: Automatically generate PDF format warning analysis reports, including prediction results, risk assessment, response suggestions, etc., supporting one-key export.

[0038] Example 2: In another embodiment, an AI-based marine disaster prevention and monitoring warning method is provided, comprising: Real-time collection of multi-source data: Obtain hydrological, meteorological, topographic, and remote sensing image data through sensor networks, remote sensing devices, and meteorological interfaces; Data preprocessing and fusion: Clean, time and space calibration, and feature fusion of collected data to build a standardized data set; Artificial intelligence analysis and prediction: Input preprocessed data into a multi-scale prediction model cluster to obtain short-term prediction results of water level, wind speed-related parameters, and storm surge spatial diffusion simulation; Use disaster recognition models to analyze remote sensing images in real time to extract current inundation area features; Update model parameters through the model optimization unit to improve prediction accuracy; Risk assessment and warning generation: Combine prediction results, terrain data, and regional disaster-bearing body information to calculate regional risk indices; Generate warning information based on risk indices to determine warning levels and impact ranges; Early warning release and interaction: display early warning results through a visual platform, and release early warning information through multiple channels to support user interactive queries.

[0039] Specifically, in the artificial intelligence analysis and prediction step, the short-term prediction sub-model adopts an LSTM or TCN network structure, the spatial diffusion model adopts a CNN or GNN network structure, and the disaster identification model adopts a YOLO, SSD or improved UNet model.

[0040] Specifically, in the early warning release and interaction step, early warning information is released through multiple channels such as SMS, APP, broadcast and emergency platform, and different early warning contents are customized for different user groups.

[0041] Specifically, it further includes a system self-learning and optimization step: Collect historical early warning data and actual disaster conditions to build a verification data set; Compare and analyze the prediction results and actual conditions to evaluate the model performance; Adjust the model parameters and risk assessment index system according to the evaluation results to continuously improve the system early warning accuracy.

[0042] In the description of the present specification, the description of the terms "one embodiment", "example", "specific example" and the like means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0043] The preferred embodiments of the application disclosed above are only used to help explain the application. The preferred embodiments do not describe all the details and do not limit the application to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the present specification. The present specification selects and describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. AI-based marine disaster prevention monitoring and early warning system, characterized by: include: Multi-source data acquisition module, including IoT sensor network, remote sensing data access interface and meteorological data interface, for real-time acquisition of hydrological, meteorological, topographic and remote sensing image data; Data preprocessing and fusion module, including data cleaning unit, spatiotemporal calibration unit and feature fusion unit, is used to clean, calibrate and fuse multi-source data at the feature level; An artificial intelligence analysis module, including a multi-scale prediction model cluster unit, a model optimization unit, and an uncertainty analysis unit. The multi-scale prediction model cluster includes a short-term prediction model based on LSTM or TCN, a spatial diffusion model based on CNN or GNN, and a disaster identification model based on a target detection algorithm. The model optimization unit uses real-time monitoring data to update the model online and adapt it to the geographical characteristics of different regions through transfer learning. The risk assessment and warning generation module is used to combine prediction results, terrain data and regional hazard-bearing body information to assess flooding risks and generate graded warning information; The interaction and publishing module is used to display the warning results through the visual display platform and push warning information to users.

2. The AI-based marine disaster prevention monitoring and early warning system according to claim 1 is characterized in that: The IoT sensor network adopts a distributed self-organizing architecture, including: Edge computing nodes, deployed at coastal monitoring sites, are responsible for local data preprocessing and feature extraction; Regional aggregation nodes use Mesh network technology to achieve self-organizing network communication between adjacent nodes to ensure data transmission reliability; The central control node receives data from various regions and performs global fusion analysis; When the communication of a monitoring node is interrupted, the system automatically reconstructs the network topology and relays data through adjacent nodes to ensure the robustness of the monitoring network.

3. The AI-based marine disaster prevention monitoring and early warning system according to claim 1 is characterized in that: The sub-models in the multi-scale prediction model cluster unit work together, specifically including: The short-term prediction model predicts water level and wind speed parameters in the next 1-6 hours based on historical hydrological and meteorological data; The spatial diffusion model combines terrain data and coastline characteristics to simulate the spatial diffusion process of storm surges in the coastal zone; The disaster identification model identifies and segments flooded areas in remote sensing images in real time.

4. The AI-based marine disaster prevention monitoring and early warning system according to claim 1 is characterized in that: The risk assessment and early warning generation module is used to: Establish a regional risk assessment indicator system that includes inundation depth, population density, and infrastructure importance; The risk index is generated by weighted calculation: ； in, is the risk index, is the submergence depth, is the population density, is the facility importance coefficient, , , is the weight coefficient determined through training of historical disaster data; Based on the comparison between the risk index and the preset threshold, blue, yellow, orange and red graded warnings are generated.

5. The AI-based marine disaster prevention monitoring and early warning system according to claim 1 is characterized in that: The uncertainty analysis unit quantifies the uncertainty of the prediction results based on Monte Carlo simulation or Bayesian inference, generates confidence intervals, and provides a probability reference for early warning decisions.

6. The AI-based marine disaster prevention monitoring and early warning system according to claim 1 is characterized in that: The visualization display platform uses GIS maps to display the storm surge evolution process, flooding risk distribution and warning areas in real time, and supports multi-scale zooming, time slider interaction and dynamic deduction of the flooding process.

7. The AI-based marine disaster prevention monitoring and early warning method is characterized by: The following steps are involved: Real-time multi-source data acquisition: Acquisition of hydrological, meteorological, topographic and remote sensing image data through sensor networks, remote sensing equipment and meteorological interfaces; Data preprocessing and fusion: Clean the collected data, calibrate time and space, and fuse features to build a standardized data set; Artificial Intelligence Analysis and Prediction: The preprocessed data is fed into a multi-scale prediction model cluster to obtain short-term prediction results for water level and wind speed related parameters and simulate the spatial diffusion of storm surges; Use disaster identification models to conduct real-time analysis of remote sensing images and extract the characteristics of the current flooded area; Update model parameters through the model optimization unit to improve prediction accuracy; Risk assessment and early warning generation: Calculate the risk index for each region by combining prediction results, terrain data, and regional hazard-prone body information; Generate early warning information based on risk index and determine the warning level and impact scope; Early warning release and interaction: Display early warning results through a visualization platform, release early warning information through multiple channels, and support user interactive queries.

8. The AI-based marine disaster prevention monitoring and early warning method according to claim 7 is characterized in that: In the artificial intelligence analysis and prediction step, the short-term prediction sub-model adopts LSTM or TCN network structure, the spatial diffusion model adopts CNN or GNN network structure, and the disaster identification model adopts YOLO, SSD or improved UNet model.

9. The AI-based marine disaster prevention monitoring and early warning method according to claim 7 is characterized in that: In the warning release and interaction steps, warning information is released simultaneously through multiple channels such as text messages, broadcasts, and emergency platforms, and differentiated warning content is customized for different user groups.

10. The AI-based marine disaster prevention monitoring and early warning method according to claim 7, characterized in that: It also includes system self-learning and optimization steps: Collect historical warning data and actual disaster situations to build a validation dataset; Compare and analyze the predicted results with the actual situation to evaluate the model performance; Adjust model parameters and risk assessment indicator system based on the evaluation results to continuously improve the system warning accuracy.

Images