Method for predicting maritime traffic flow based on multi-source AIS-meteorological data fusion

Abstract

The application discloses a marine traffic flow probability prediction method based on multi-source AIS-weather data fusion, and belongs to the field of marine traffic safety management, and comprises the following steps: S1, obtaining a standardized multi-source data set; S2, fusing AIS ship dynamic statistical characteristics and weather environment characteristics through a ship motion transfer function, and constructing a space-time patch considering short-term ship motion trend and long-term sea state trend; S3, based on the space-time patch, fusing patch Transformer time series modeling and adaptive dynamic graph space correlation learning to obtain space-time fusion characteristics; S4, outputting traffic flow state prediction values and probability distribution of traffic flow key indicators. The marine traffic flow probability prediction method based on multi-source AIS-weather data fusion breaks through the limitations of traditional methods, such as single data dimension, insufficient space-time correlation capture and lack of probability confidence, and realizes high-precision probability prediction of marine traffic flow under complex sea conditions.

Description

Technical Field

[0001] This invention relates to the field of maritime traffic safety management technology, and in particular to a method for predicting maritime traffic flow probability by fusing multi-source AIS-meteorological data. Background Technology

[0002] Maritime traffic flow forecasting is a core technological support for maritime management, route optimization, and safety control. Its accuracy directly impacts port scheduling efficiency, ship collision avoidance safety, and the sustainable operation of maritime transportation systems. With the development of intelligent shipping, maritime authorities have an increasingly urgent need for traffic flow forecasting that is "driven by multi-source data, highly robust, and quantifiable in confidence." This requires integrating vessel dynamic data (AIS) with environmental interference data (meteorology) while adapting to the uncertainties of complex sea conditions, providing dual-dimensional support for decision-making: "prediction results + reliability assessment." However, current data association methods are limited to the statistical level, often employing simple feature splicing (such as directly linking AIS speed features with meteorological wave height features), resulting in a lack of physical correlation in data fusion. This leads to significantly increased prediction errors under complex sea conditions (such as wave height > 3m). Summary of the Invention

[0003] The purpose of this invention is to provide a method for predicting the probability of maritime traffic flow by fusing multi-source AIS-meteorological data, thereby solving the aforementioned technical problems.

[0004] To achieve the above objectives, this invention provides a method for probabilistic prediction of maritime traffic flow based on multi-source AIS-meteorological data fusion, comprising the following steps: S1. Collect AIS ship dynamic data, meteorological and environmental data and nautical chart data to obtain the original multi-source dataset. Then, through format alignment, outlier removal, missing value repair and standardization, eliminate data heterogeneity and noise to obtain a standardized multi-source dataset. S2. Extract AIS ship dynamic statistical features and meteorological environment features from the standardized multi-source dataset, and fuse the AIS ship dynamic statistical features and meteorological environment features through the ship motion transfer function to construct a spatiotemporal patch that takes into account both short-term ship motion trends and long-term sea state trends. S3. Based on spatiotemporal patching, the spatiotemporal fusion features are obtained by fusing patched Transformer temporal modeling and adaptive dynamic graph space association learning, and introducing model uncertainty factors and sea state uncertainty. S4. Based on the spatiotemporal fusion features, a multi-step prediction head combined with second-order reliability probability modeling is adopted to output the traffic flow state prediction value and the probability distribution of key traffic flow indicators, and the prediction confidence is quantified by the traffic flow probability index.

[0005] Therefore, the maritime traffic flow probability prediction method based on the fusion of multi-source AIS and meteorological data described above has the following beneficial effects: 1. Multi-source data fusion enhances feature completeness: By integrating AIS ship dynamic data (position, speed, heading, etc.) with meteorological environmental data (wind speed, wave height, visibility, etc.), an input system containing 14 dimensions of features is constructed. This makes up for the deficiency that single AIS data cannot reflect the impact of sea conditions on traffic flow, enabling the model to more comprehensively capture the correlation between "ship behavior and environmental changes" and lay a data foundation for accurate prediction.

[0006] 2. Spatiotemporal joint modeling enhances dependency capture capability: The original data is mapped into high-dimensional embedded features through spatiotemporal patch projection, and long-term temporal dependencies (such as tidal cycles and weather change trends) are mined with the help of Transformer encoder. At the same time, the spatial unit association (such as the transmission of ship density in adjacent sea areas) is learned by using GNN dynamic adjacency matrix. Compared with traditional statistical models or single neural networks, this significantly improves the modeling accuracy of the "spatiotemporal coupling characteristics" of maritime traffic flow. 3. Probability quantification provides confidence support for decision-making: The SORM second-order reliability method is introduced to calculate the probability of sea state exceeding the threshold. Uncertainty weights are integrated into the dynamic graph construction and prediction process, and the traffic flow probability prediction results (such as density and speed) containing confidence intervals are output. This solves the problem that traditional deterministic prediction cannot quantify error risk and provides maritime regulatory authorities with a dual decision-making basis of "predicted value + credibility". 4. Enhanced robustness to complex scenarios through dynamic adaptability: The design of a graph convolutional layer with a dynamic similarity matrix and residual connection for temperature parameter regulation can respond in real time to sudden changes in sea state (such as typhoons and strong waves) and fluctuations in traffic flow (such as channel congestion and ship sharp turns). Compared with fixed structure models, the prediction error is reduced by more than 30% in complex scenarios such as extreme weather and peak navigation, which enhances the engineering application value of the method. 5. Practical Applications Empower Shipping Safety and Efficiency: The prediction results can directly support the optimization of ship routes (avoiding highly congested areas), port resource scheduling (pre-allocating berths and pilotage resources), and risk warning (identifying high-risk collision sections), helping to reduce the maritime accident rate and shorten the time ships spend in port, providing technical support for smart maritime and green shipping.

[0007] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0008] Figure 1 This is a flowchart of the maritime traffic flow probability prediction method based on multi-source AIS-meteorological data fusion of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0010] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0011] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0012] like Figure 1 As shown, the method for probabilistic prediction of maritime traffic flow based on multi-source AIS-meteorological data fusion includes the following steps: S1. Collect AIS ship dynamic data, meteorological and environmental data and nautical chart data to obtain the original multi-source dataset. Then, through format alignment, outlier removal, missing value repair and standardization, eliminate data heterogeneity and noise to obtain a standardized multi-source dataset. The original multi-source dataset obtained in step S1 ,in, This represents the original AIS vessel dynamic data, vessel static data, meteorological environmental data, and nautical chart data, among which... , These represent the ship identification code, latitude, latitude change rate, longitude, longitude change rate, ship speed above ground, heading above ground, and timestamp, respectively. , These represent the ship's length, beam, distance from bow to center of gravity / draft, block coefficient, and timestamp, respectively. , These represent the sea surface wave height, peak period in the wave energy spectrum, sea surface wind speed, sea surface wind direction, and timestamp, respectively. , These represent the depth of the sea area and the navigation boundary, respectively. , Indicates the latitude range. Indicates a range of longitudes. and These are the lower and upper limits of latitude, respectively. and These are the lower and upper limits of longitude, respectively. Standardized multi-source datasets ,in, These represent the standardized AIS ship dynamic data, ship static data, meteorological environment data, and nautical chart data, respectively.

[0013] S2. Extract AIS ship dynamic statistical features and meteorological environment features from the standardized multi-source dataset, and fuse the AIS ship dynamic statistical features and meteorological environment features through the ship motion transfer function to construct a spatiotemporal patch that takes into account both short-term ship motion trends and long-term sea state trends. Step S2 specifically includes the following steps: S21. Feature Extraction: Extract AIS ship dynamic statistical features respectively. and meteorological environment characteristics The basic feature set for classification is obtained. ;in, Indicates the steepness of the wave; S22. Based on ship speed-sea state matching checks and heading-wind direction rationality checks, abnormal physical logic is filtered to obtain the normal basic feature set. ; S23. Determine the core parameter set of the ship's motion transfer function. , These represent the wave number, Froude number, ship-wave direction angle, and response coefficient, respectively. Among them, wave number The calculation formula is as follows: ; In the formula, Indicates wavelength; Represents gravitational acceleration; Fu Rude Number The calculation formula is as follows: ; Ship-wave direction angle The formula is as follows: ; Response coefficient The calculation formula is as follows: ; In the formula, Represents the ship-wave interaction coefficient; This represents an approximate value for hydrodynamic damping, and ; S24, Core parameter set based on ship motion transfer function The physical characteristics of ship motion are calculated, and the influence of sea state on ship motion is quantified into characteristics to enhance the physical correlation of multi-source fusion. Among them, the physical characteristics of ship motion include heave transfer function (reflecting the response amplitude of ship's vertical up and down motion), pitch transfer function (reflecting the response amplitude of ship's pitching motion), and bow vertical acceleration transfer function (reflecting the vertical acceleration response of the bow position, which is related to the influence of ship turbulence on traffic flow speed). hysteresis transfer function The expression is as follows: ; In the formula, Represents the heave excitation function; Pitch transfer function The expression is as follows: ; In the formula, Represents the pitch excitation function; Vertical acceleration transfer function of bow The expression is as follows: ; In the formula, Indicates the frequency of encounters; S25. Use Z-score normalized heave transfer function Pitch transfer function and the bow vertical acceleration transfer function The standardized heave transfer function is obtained. Pitch transfer function and the bow vertical acceleration transfer function Generate derived feature sets ; S26. Transform the normal basic feature set With derived feature set By concatenating the data along the sample-time-feature dimensions, a multi-source fusion feature matrix is ​​obtained. , Represents the set of real numbers. This indicates the number of sampling points within a historical time window. Indicates the number of ships. This represents the sum of the dimensions of AIS ship dynamic data and meteorological environment data; S27. Based on spatial grid partitioning, the multi-source fusion feature matrix is ​​transformed into a structured token of spatial unit-temporal patch. This not only retains the "long dependency capture efficiency" of the PTSTG patch idea, but also adapts to the ship motion and sea state change patterns through sea-specific parameters (such as grid size and patch length), providing efficient input for subsequent Transformer temporal modeling.

[0014] In step S22, the ship speed-sea state matching check rules are as follows: when ,and If it is, then it is marked as a physical logic exception; among which, Indicates the maximum safe speed of the vessel; The rules for checking the reasonableness of heading and wind direction are as follows: when And the sea wind direction Comparison with heading If the angle between them is less than 30°, it is marked as a physical logic anomaly.

[0015] Step S27 specifically includes the following steps: S271. Divide the predicted sea area into uniform grids according to latitude and longitude, and define spatial cells as the basic units for traffic flow statistics to obtain spatial grid parameters. , These represent the grid side lengths respectively. Corresponding latitudinal intervals and grid side lengths The corresponding latitudinal interval, number of grids in the latitudinal direction, number of grids in the longitude direction, and total number of spatial units; Among them, the grid side length Corresponding latitudinal interval The calculation formula is as follows: ; In the formula, This represents the distance corresponding to 1° latitude at the equator; Grid side length Corresponding latitudinal interval The calculation formula is as follows: ; In the formula, This represents the latitude value of the center point of the target space grid; Latitude grid number The calculation formula is as follows: ; Longitude grid number The calculation formula is as follows: ; Total number of spatial units The calculation formula is as follows: ; And the grid numbering rule is from Numbering them in order of latitude first, then longitude, yields the... The coordinates of each spatial unit are ,in , , Indicates latitude direction index. Indicates longitude direction index; and marks the spatial units where no-navigation zones and wind farms are located as impassable; S272. Map AIS ship dynamic data to spatial units, statistically analyze the basic traffic flow indicators of each spatial unit, aggregate ship-level features into spatial unit-level features, and obtain a spatial unit-level feature matrix to provide a unified spatial dimension input for time-series patch segmentation. S2721, when When the space belongs to a passage space unit, establish the AIS ship dynamic data-space unit mapping relationship: ; In the formula, Indicates time Ships The spatial unit code to which it belongs; and Indicates time Ships The latitude and longitude of the location; S2722, Basic Indicators for Statistical Traffic Flow: Spatial Unit Density and the average velocity of space units ,in, , , Indicates time , No. Number of ships within each spatial unit Indicates time Ships The speed of navigation; S2723, Statistics , With meteorological and environmental data, derived feature sets Integration to form a spatial unit-level feature matrix ; S273. Determine the length of the time-series patch based on the temporal variation characteristics of maritime traffic flow and sea state. Step length and number of patches : ; ; ; In the formula, Indicates the maximum time lag; Indicates the sampling interval; The time autocorrelation coefficient representing sea state, and , Represents the mathematical expectation operator. express High wave height at any moment This represents the historical average wave height. Indicates the historical wave height variance. express High wave height at any moment Indicates a time lag; Indicates the correlation threshold; Indicates the overlap rate between patches; S274, according to the determined timing patch length Step length and number of patches The temporal features of each spatial unit are slidably segmented to generate a two-dimensional token structure of spatial unit-temporal patch. And the first The spatial feature is the first Patch features , Indicates the first The spatial feature is the first The starting index of the time window for each patch. ; S275. Using linear projection, the low-dimensional features (below 14 dimensions) of each patch are linearly projected to a high-dimensional space of 128 dimensions to enhance feature representation and provide high-dimensional input for subsequent Transformer attention modeling. Normalization is then performed to obtain the normalized patch projection features. ; S276. Add sinusoidal position encoding to the time steps within the patch, enabling the Transformer to recognize the temporal order within the patch and avoiding the loss of temporal information due to the disorder of the attention mechanism, thus obtaining the final patch feature containing sinusoidal position encoding. : ; In the formula, Indicates the first The spatial feature is the first The final encoded characteristics of each patch; Indicates the first The spatial feature is the first Projection features of each patch; Indicates the first Position encoding of each time step; S277, Finalize the features of all patches within the spatial unit. The resulting spatiotemporal patch matrix is ​​then integrated into a unified spatiotemporal patch matrix. .

[0016] S3. Based on spatiotemporal patching, the spatiotemporal fusion features are obtained by fusing patched Transformer temporal modeling and adaptive dynamic graph space association learning, and introducing model uncertainty factors and sea state uncertainty. Step S3 specifically includes the following steps: S31. Construct time-series coding features that incorporate model uncertainty factors; S311. Determine the model uncertainty factors : ; In the formula, , , and All represent regression coefficients; S312, Model uncertainty factors Added to the feedforward network of the Transformer encoder to correct model errors: ; In the formula, This represents the output feature vector of the feedforward network in the Transformer encoder; and These represent the weight matrices of the first and second layers of the feedforward network, respectively. and These represent the bias vectors of the first and second layers of the feedforward network, respectively. Represents the activation function of the Gaussian error linear unit; S313, Input the final spacetime patch matrix Using the two-layer Transformer encoder described in step S32, temporal coding features are generated. ; S32. Construct an adaptive dynamic graph that incorporates sea state uncertainties; S321, Node Embedding Learning: Average pooling of temporally encoded features into spatial unit embeddings: ; In the formula, Indicates the first Embedding vectors of spatial units; Indicates the first The spatial unit of the first The temporal coding characteristics of each patch; S322, Determine the weights for sea state uncertainty : ; In the formula, Indicates the first The probability of sea state exceeding the threshold in each spatial unit, and , The cumulative distribution function represents the standard normal distribution. Indicates the first The second-order reliability index of each spatial unit, Indicates the first The principal curvature of the limit state surface of a spatial unit at the design point; and These represent the indices of the principal curvatures of the limit state surface and the total dimension of the random variables involved in the calculation, respectively. S323, Incorporating Sea State Uncertainty Weights Construct a similarity matrix: ; In the formula, Indicates the first The spatial unit and the first The similarity of spatial units; Represents a rectifier linear unit; Indicates the first The spatial unit and the first Feature similarity of spatial units Indicates transpose; Indicates the first Sea state uncertainty weights for each spatial unit; S324. Construct a dynamic adjacency matrix. : ; in, ; In the formula, Indicates the first The spatial unit for the first The influence weight of each spatial unit; This indicates the total number of spatial units for the predicted sea area division; Indicates the first The spatial unit and the first Similarity score of each spatial unit; Indicates temperature parameter; S33, Based on Embedded Vectors and dynamic adjacency matrix The graph neural network propagates the correlation features between spatial units; Convolutional layers: ; Residuals and Layer Normalization Layer: In the formula, This represents the feature vector output by the graph convolutional layer; Indicates node embedding, ; Indicates the adjacency weight; Indicates the graph convolution parameters; Indicates the final spatiotemporal fusion characteristics; This represents the feature vector output by the graph convolutional layer after layer normalization; Represents a multilayer perceptron; Representation layer normalization.

[0017] S4. Based on the spatiotemporal fusion features, a multi-step prediction head combined with second-order reliability probability modeling is adopted to output the traffic flow state prediction value and the probability distribution of key traffic flow indicators, and the prediction confidence is quantified by the traffic flow probability index.

[0018] Step S4 specifically includes the following steps: S41, Based on the final spatiotemporal fusion characteristics A multi-step prediction result matrix is ​​generated through a prediction head that uses a single-layer mapping and multi-step parallel output. : ; In the formula, Indicates the first The output of the intermediate layer; Indicates the first The intermediate layer dimension to the first The weights of each predictive indicator; Indicates the first The first predictive indicator The offset of each time step; S42. Convert the multi-step prediction result matrix It is restructured into an intuitive dimension of time step-spatial unit-indicator, and the predicted values ​​are verified to conform to the physical laws of maritime traffic flow, generating traffic flow state prediction values. : ; In the formula, ,when hour, Indicates the first Time step The predicted density of each spatial unit; when hour, Indicates the first Time step Prediction speed of each spatial unit; S43. Construct the density limit state function and the velocity limit state function respectively: ; ; In the formula, and They represent the first Time step The density limit state function value and velocity limit state function value of each spatial element; Indicates the maximum safe density in the sea area; Indicates the minimum safe speed; S44. Calculate the probability of exceeding the safety threshold. : ; In the formula, Indicates the second-order reliability index; Represents the principal curvature of the limit state surface at the design point; Obtain density hyperprobability and speed overprobability ; S45, Based on the probability of exceeding the safety threshold Fit the normal distribution parameters of density and velocity to determine the probability of exceeding the safety threshold. Transform into a probability distribution; S46. Quantifying prediction confidence using traffic flow probability index. : ; in, ; ; In the formula, Indicates the average confidence level; Indicates the quantile of the t-distribution; Indicates the sample standard deviation; Indicates the first Traffic flow probability index at each time step; Indicates the total time steps; Represents the density safety probability, and ; This represents the speed safety probability, and .

[0019] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting the probability of maritime traffic flow by fusing multi-source AIS-meteorological data, characterized by: Includes the following steps: S1. Collect AIS ship dynamic data, meteorological and environmental data and nautical chart data to obtain the original multi-source dataset. Then, through format alignment, outlier removal, missing value repair and standardization, eliminate data heterogeneity and noise to obtain a standardized multi-source dataset. S2. Extract AIS ship dynamic statistical features and meteorological environment features from the standardized multi-source dataset, and fuse the AIS ship dynamic statistical features and meteorological environment features through the ship motion transfer function to construct a spatiotemporal patch that takes into account both short-term ship motion trends and long-term sea state trends. S3. Based on spatiotemporal patching, the spatiotemporal fusion features are obtained by fusing patched Transformer temporal modeling and adaptive dynamic graph space association learning, and introducing model uncertainty factors and sea state uncertainty. S4. Based on the spatiotemporal fusion features, a multi-step prediction head combined with second-order reliability probability modeling is adopted to output traffic flow state prediction values ​​and probability distributions of key traffic flow indicators, and the prediction confidence is quantified by using the traffic flow probability index. Step S3 specifically includes the following steps: S31. Construct time-series coding features that incorporate model uncertainty factors; S311. Determine the model uncertainty factors ; S312, Model uncertainty factors Added to the feedforward network of the Transformer encoder to correct model errors; S313, Input the final spacetime patch matrix Using a two-layer Transformer encoder, temporal encoded features are generated. ; S32. Construct an adaptive dynamic graph that incorporates sea state uncertainties; S321, Node Embedding Learning: Average pooling of temporal encoded features into spatial unit embeddings; S322, Determine the weights for sea state uncertainty ; S323, Incorporating Sea State Uncertainty Weights Construct a similarity matrix; S324. Construct a dynamic adjacency matrix based on the similarity matrix. ; S33, Based on Embedded Vectors and dynamic adjacency matrix The graph neural network propagates the correlation features between spatial units.

2. The method for predicting maritime traffic flow probability by fusing multi-source AIS-meteorological data according to claim 1, characterized in that: The original multi-source dataset obtained in step S1 ,in, This represents the original AIS ship dynamic data, ship static data, meteorological environmental data, and nautical chart data. , These represent the ship identification code, latitude, latitude change rate, longitude, longitude change rate, ship speed above ground, heading above ground, and timestamp, respectively. , These represent the ship's length, beam, distance from bow to center of gravity / draft, block coefficient, and timestamp, respectively. , These represent the sea surface wave height, peak period in the wave energy spectrum, sea surface wind speed, sea surface wind direction, and timestamp, respectively. , These represent the depth of the sea area and the navigation boundary, respectively. , Indicates the latitude range. Indicates a range of longitudes. and These are the lower and upper limits of latitude, respectively. and These are the lower and upper limits of longitude, respectively. Standardized multi-source datasets ,in, These represent the standardized AIS ship dynamic data, ship static data, meteorological environment data, and nautical chart data, respectively.

3. The method for predicting maritime traffic flow probability by fusing multi-source AIS-meteorological data according to claim 2, characterized in that: Step S2 specifically includes the following steps: S21. Feature Extraction: Extract AIS ship dynamic statistical features respectively. and meteorological environment characteristics The basic feature set for classification is obtained. ;in, Indicates the steepness of the wave; S22. Based on ship speed-sea state matching checks and heading-wind direction rationality checks, abnormal physical logic is filtered to obtain the normal basic feature set. ; S23. Determine the core parameter set of the ship's motion transfer function. , These represent the wave number, Froude number, ship-wave direction angle, and response coefficient, respectively. Among them, wave number The calculation formula is as follows: ； In the formula, Indicates wavelength; Represents gravitational acceleration; Fu Rude Number The calculation formula is as follows: ； Ship-wave direction angle The calculation formula is as follows: ； Response coefficient The calculation formula is as follows: ； In the formula, Represents the ship-wave interaction coefficient; This represents an approximate value for hydrodynamic damping, and ; S24, Core parameter set based on ship motion transfer function Calculate the physical characteristics of ship motion, including the heave transfer function, pitch transfer function, and bow vertical acceleration transfer function; hysteresis transfer function The expression is as follows: ； In the formula, Represents the heave excitation function; Pitch transfer function The expression is as follows: ； In the formula, Represents the pitch excitation function; Vertical acceleration transfer function of bow The expression is as follows: ； In the formula, Indicates the frequency of encounters; S25. Use Z-score normalized heave transfer function Pitch transfer function and the bow vertical acceleration transfer function The standardized heave transfer function is obtained. Pitch transfer function and the bow vertical acceleration transfer function Generate derived feature sets ; S26. Transform the normal basic feature set With derived feature set By concatenating the data along the sample-time-feature dimensions, a multi-source fusion feature matrix is ​​obtained. , Represents the set of real numbers. This indicates the number of sampling points within a historical time window. Indicates the number of ships. This represents the sum of the dimensions of AIS ship dynamic data and meteorological environment data; S27. Based on spatial grid partitioning, the multi-source fusion feature matrix is ​​transformed into a structured token of spatial unit-temporal patch.

4. The method for predicting maritime traffic flow probability by fusing multi-source AIS-meteorological data according to claim 3, characterized in that: In step S22, the ship speed-sea state matching check rules are as follows: when ,and If it is, then it is marked as a physical logic exception; among which, Indicates the maximum safe speed of the vessel; The rules for checking the reasonableness of heading and wind direction are as follows: when And the sea wind direction Comparison with heading If the angle between them is less than 30°, it is marked as a physical logic anomaly.

5. The method for predicting maritime traffic flow probability by fusing multi-source AIS-meteorological data according to claim 3, characterized in that: Step S27 specifically includes the following steps: S271. Divide the predicted sea area into uniform grids according to latitude and longitude, and define spatial cells as the basic units for traffic flow statistics to obtain spatial grid parameters. , These represent the grid side lengths respectively. Corresponding latitudinal intervals and grid side lengths The corresponding longitude interval, number of grids in the latitudinal direction, number of grids in the longitude direction, and total number of spatial units; Among them, the grid side length Corresponding latitudinal interval The calculation formula is as follows: ； In the formula, This represents the distance corresponding to 1° latitude at the equator; Grid side length Corresponding longitude interval The calculation formula is as follows: ； In the formula, This represents the latitude value of the center point of the target space grid; Latitude grid number The calculation formula is as follows: ； Longitude grid number The calculation formula is as follows: ； Total number of spatial units The calculation formula is as follows: ； And the grid numbering rule is from Numbered in order of latitude first, then longitude, to obtain the number. The coordinates of each spatial unit are ,in , , Indicates latitude direction index. Indicates longitude direction index; and marks the spatial units where no-navigation zones and wind farms are located as impassable; S272. Map AIS ship dynamic data to spatial units, statistically analyze the basic traffic flow indicators of each spatial unit, aggregate ship-level features into spatial unit-level features, and obtain a spatial unit-level feature matrix. S2721, when When the space belongs to a passage space unit, establish the AIS ship dynamic data-space unit mapping relationship: ； In the formula, Indicates time Ships The spatial unit code to which it belongs; and Indicates time Ships The latitude and longitude of the location; S2722, Basic Indicators for Statistical Traffic Flow: Spatial Unit Density and the average velocity of space units ,in, , , Indicates time , No. Number of ships within each spatial unit Indicates time Ships The speed of navigation; S2723, Statistics , With meteorological and environmental data, derived feature sets Integration to form a spatial unit-level feature matrix ; S273. Determine the length of the time-series patch based on the temporal variation characteristics of maritime traffic flow and sea state. Step length and number of patches : ； ； ； In the formula, Indicates the maximum time lag; Indicates the sampling interval; The time autocorrelation coefficient representing sea state, and , Represents the mathematical expectation operator. express High wave height at any moment This represents the historical average wave height. Indicates the historical wave height variance. express High wave height at any moment Indicates a time lag; Indicates the correlation threshold; Indicates the overlap rate between patches; S274, according to the determined timing patch length Step length and number of patches The temporal features of each spatial unit are slidably segmented to generate a two-dimensional token structure of spatial unit-temporal patch. And the first The spatial feature is the first Patch features , Indicates the first The spatial feature is the first The starting index of the time window for each patch. ; S275. Using linear projection, the low-dimensional features (below 14 dimensions) of each patch are linearly projected to a high-dimensional space of 128 dimensions, and then normalized to obtain the normalized patch projection features. ; S276. Add sine position encoding to the time steps within the patch to obtain the final patch feature containing sine position encoding. : ； In the formula, Indicates the first The spatial feature is the first The final encoded characteristics of each patch; Indicates the first The spatial feature is the first Projection features of each patch; Indicates the first Position encoding of each time step; S277, Finalize the features of all patches within the spatial unit. The resulting spatiotemporal patch matrix is ​​then integrated into a unified spatiotemporal patch matrix. .

6. The method for predicting maritime traffic flow probability by fusing multi-source AIS-meteorological data according to claim 5, characterized in that: The model uncertainty factor mentioned in step S311 The calculation formula is as follows: ； In the formula, , , and All represent regression coefficients; In S312, the output feature vector expression of the feedforward network in the Transformer encoder is as follows: ； In the formula, This represents the output feature vector of the feedforward network in the Transformer encoder; and These represent the weight matrices of the first and second layers of the feedforward network, respectively. and These represent the bias vectors of the first and second layers of the feedforward network, respectively. Represents the activation function of the Gaussian error linear unit; The node embedding expression for S321 is as follows: ； In the formula, Indicates the first Embedding vectors of spatial units; Indicates the first The spatial unit of the first The temporal coding characteristics of each patch; Sea state uncertainty weights as described in S322 The expression is as follows: ； In the formula, Indicates the first The probability of sea state exceeding the threshold in each spatial unit, and , The cumulative distribution function represents the standard normal distribution. Indicates the first The second-order reliability index of each spatial unit, Indicates the first The principal curvature of the limit state surface of a spatial unit at the design point; and These represent the indices of the principal curvatures of the limit state surface and the total dimension of the random variables involved in the calculation, respectively. The similarity matrix expression described in S323 is as follows: ； In the formula, Indicates the first The spatial unit and the first The similarity of spatial units; Represents a rectifier linear unit; Indicates the first The spatial unit and the first Feature similarity of spatial units Indicates transpose; Indicates the first Sea state uncertainty weights for each spatial unit; The dynamic adjacency matrix described in step S324 The expression is as follows: ； in, ； In the formula, Indicates the first The spatial unit for the first The influence weight of each spatial unit; This indicates the total number of spatial units for the predicted sea area division; Indicates the first The spatial unit and the first Similarity score of each spatial unit; Indicates temperature parameter; In step S33, the graph convolutional layer: ; Residuals and Layer Normalization Layer: In the formula, This represents the feature vector output by the graph convolutional layer; Indicates node embedding, ; Indicates the adjacency weight; Indicates the graph convolution parameters; Indicates the final spatiotemporal fusion characteristics; This represents the feature vector output by the graph convolutional layer after layer normalization; Represents a multilayer perceptron; Representation layer normalization.

7. The method for predicting maritime traffic flow probability by fusing multi-source AIS-meteorological data according to claim 6, characterized in that: Step S4 Specifically, the following steps are included: S41, Based on the final spatiotemporal fusion characteristics A multi-step prediction result matrix is ​​generated through a prediction head that uses a single-layer mapping and multi-step parallel output. : ； In the formula, Indicates the first The output of the intermediate layer; Indicates the first The intermediate layer dimension to the first The weights of each predictive indicator; Indicates the first The first predictive indicator The offset of each time step; S42. Convert the multi-step prediction result matrix It is restructured into an intuitive dimension of time step-spatial unit-indicator, and the predicted values ​​are verified to conform to the physical laws of maritime traffic flow, generating traffic flow state prediction values. : ； In the formula, ,when hour, Indicates the first Time step The predicted density of each spatial unit; when hour, Indicates the first Time step Prediction speed of each spatial unit; S43. Construct the density limit state function and the velocity limit state function respectively: ； ； In the formula, and They represent the first Time step The density limit state function value and velocity limit state function value of each spatial element; Indicates the maximum safe density in the sea area; Indicates the minimum safe speed; S44. Calculate the probability of exceeding the safety threshold. : ； In the formula, Indicates the second-order reliability index; Represents the principal curvature of the limit state surface at the design point; Obtain density hyperprobability and speed overprobability ; S45, Based on the probability of exceeding the safety threshold Fit the normal distribution parameters of density and velocity to determine the probability of exceeding the safety threshold. Transform into a probability distribution; S46. Quantifying prediction confidence using traffic flow probability index. : ； in, ； ； In the formula, Indicates the average confidence level; Indicates the quantile of the t-distribution; Indicates the sample standard deviation; Indicates the first Traffic flow probability index at each time step; Indicates the total time steps; Represents the density safety probability, and ; This represents the speed safety probability, and .

Images