Ship navigation sea wave dynamic space-time forecasting method and system based on deep learning

Abstract

The invention belongs to the technical field of marine environment prediction, and discloses a ship navigation sea wave dynamic space-time prediction method and system based on deep learning. The method comprises the following steps: carrying out space-time alignment, missing value repair and standardization processing on acquired ship AIS data and an ERA5 reanalysis data set, and generating node feature vectors containing latitudes and longitudes, timestamps, wind speeds and significant wave heights; through node feature coding, centrality coding, space coding and time coding, ship trajectory node importance and time-space interaction relation are quantified. Constructing a SeaGraph model, and outputting an effective wave height prediction value of a target waypoint; and performing verification. According to the method, the space-time constraint of a traditional static modeling framework is broken through, the advantages of a self-attention mechanism and a graph network are integrated, the predictive modeling capability of dynamic evolution of a wave field in front of a ship navigation track is enhanced, and high-precision sea wave forecasting support is provided for intelligent ship navigation under complex sea conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine environment prediction, and particularly relates to a method and system for dynamically predicting ship navigation sea waves in space and time based on deep learning. Background Art

[0002] With the accelerating progress of globalization, maritime shipping, as the core carrier of trade, is experiencing unprecedented development. However, the expansion of shipping scale and climate change have led to frequent extreme meteorological events, and complex sea conditions pose a severe challenge to the safety of ship navigation. Therefore, the development of high-precision and real-time dynamic ship route-level sea wave prediction technology has become a key requirement for improving navigation safety and intelligent decision-making efficiency.

[0003] Currently, traditional ship marine environment perception mainly relies on three types of technical means: (1) a point data acquisition system based on observation buoys and coastal marine stations, which has a low coverage rate in the open ocean and is difficult to reflect the characteristics of the dynamic wave field; (2) satellite remote sensing technology can obtain wide-area wave field information, but due to data inversion delay, long revisit period, and weather interference limitations, the timeliness is significantly insufficient; (3) numerical prediction models rely on solving physical equations, and the high-resolution requirements lead to an exponential increase in the amount of calculation, and they are sensitive to parameterization errors of sudden events such as typhoons, making it difficult to meet the real-time prediction requirements. The imbalance among the data update frequency, spatial resolution, and computational efficiency of the above methods severely restricts the ship's ability to capture the dynamic evolution of the wave field around the route.

[0004] In recent years, deep learning technology has provided new ideas for breaking through the limitations of traditional physical modeling. Existing research mainly focuses on two types of paths: (1) a fixed-point wave prediction method based on a time series model. Although such methods are superior to numerical models in short-term accuracy, the problem of spatial and temporal misalignment between the static modeling framework and the dynamic navigation trajectory of the ship is prominent; (2) a wave inversion technology based on ship motion response, which inversely deduces wave parameters through six-degree-of-freedom sensor data. However, such methods can only sense the local wave field around the ship and cannot predict the dynamic evolution trend of the unnavigated sea area ahead. The lack of dynamic perception and predictability of existing systems leads to limited efficiency in intelligent obstacle avoidance and route optimization decision-making under complex sea conditions.

[0005] In the prior art, "Feasibility of a Wave Measurement Method for Self-Propelled Buoys Based on Artificial Neural Networks" proposed an inversion model based on artificial neural networks. Using the ship's sway motion history as the input and the wave history as the output, the artificial neural network is used to extract the ship's motion characteristics, and the input linear function is used to invert the wave surface history. However, this method has only been verified by pool tests and can only invert the wave characteristics currently encountered by the ship and cannot predict the wave information at future route points.

[0006] Through the above analysis, the problems and defects existing in the prior art are as follows: Problem 1: Limited coverage. The point data acquisition system based on observation buoys and coastal ocean stations has a low coverage rate in the open ocean and is difficult to reflect the global characteristics of the dynamic wave field. For example, buoys can only provide real-time data for local sea areas and cannot capture the spatial propagation law of the wave field over time, resulting in insufficient global perception ability for complex sea conditions.

[0007] Problem 2: Insufficient timeliness. Although satellite remote sensing can obtain wide-area wave field information, it is limited by data inversion delay, long revisit period, and weather interference, and cannot meet the needs of ships for real-time dynamic sea wave forecasting.

[0008] Problem 3: Low computational efficiency. Traditional numerical models rely on solving physical equations, and the high-resolution requirement leads to an exponential increase in the amount of calculation. In addition, the model initialization depends on high-quality initial field data, which is difficult to obtain in real time in practical applications, further limiting its forecasting accuracy and response speed.

[0009] Problem 4: Temporal and spatial misalignment between the static modeling framework and the dynamic navigation trajectory. Existing fixed-point wave prediction methods based on time series models usually take historical data at fixed positions as input and are difficult to adapt to the temporal and spatial change requirements of ships' dynamic navigation trajectories.

[0010] Problem 5: Insufficient local perception and global predictability. The wave inversion technology based on ship motion response can only perceive the local wave field around the ship and cannot predict the dynamic evolution trend of the unnavigated sea area ahead. Summary of the Invention

[0011] To overcome the problems existing in the related technologies, the disclosed embodiments of the present invention provide a method and system for dynamic spatio-temporal forecasting of sea waves for ship navigation based on deep learning, specifically related to a method and system for dynamic spatio-temporal forecasting of sea waves for ship navigation based on SeaGraph.

[0012] The technical solution is as follows: A method for dynamic spatio-temporal forecasting of sea waves for ship navigation based on deep learning, the method comprising the following steps: S1, perform spatio-temporal alignment, missing value repair, and standardization processing on the collected ship AIS data and ERA5 reanalysis dataset to generate a node feature vector containing longitude and latitude, timestamp, wind speed, and significant wave height; the node feature vector quantifies the importance of ship trajectory nodes and spatio-temporal interaction relationships through node feature encoding, centrality encoding, spatial encoding, and time encoding; S2, based on the quantification results of the importance of ship trajectory nodes and spatio-temporal interaction relationships, construct a SeaGraph model using the Transformer architecture, dynamically capture the spatio-temporal propagation law of the wave field through the fusion of the multi-head self-attention mechanism and the graph network, and output the predicted value of the significant wave height of the target waypoint through the dynamic output layer; S3. Validate the SeaGraph model and analyze its performance through SeaGraph model training, comparative experiments, and extreme event verification.

[0013] In step S1, align the collected ship AIS data with the ERA5 reanalysis dataset in terms of time and space, including: Extract AIS data from the ship trajectory data in the global route, including timestamp UTC, longitude and latitude, speed, and heading, with a sampling interval of 0.5 - 3 hours; For ocean environmental data, use the ERA5 reanalysis dataset to extract the 10 - meter - height wind speed and significant wave height in the target sea area; For time - space alignment, match the ship trajectory points with the ERA5 grid points in terms of time and space. For non - integer grid points, use bilinear interpolation to complete the parameters. The expression is: ; In the formula, is the ship position, is the weight of the th grid point, is the value at a certain position, are the 4 nearest ERA5 grid points, and the weights are determined by the inverse square of the distance.

[0014] In step S1, repair missing values and perform standardization processing on the collected ship AIS data and the ERA5 reanalysis dataset, including: Repair missing trajectories. For trajectory segments with continuous missing values exceeding 3 hours, use linear interpolation to complete them. The expression is: ; In the formula, is the longitude and latitude position at time, and are the longitude and latitude positions of adjacent valid track points, are both time intervals, ; For standardization processing, perform Z - score standardization on the wind speed and significant wave height. The expression is: ; In the formula, is the value after normalization, is the value before normalization, is the mean of the training set, is the standard deviation.

[0015] In step S1, encode node features, including: taking each ship trajectory point as a node, and its feature vector includes longitude and latitude, time, wind field, and wave information; the ship trajectory has A node, and the feature vector of each node is represented as , where is the feature dimension; is the real number space of dimension d; The centrality encoding is used to capture the importance of key nodes in the ship trajectory. Nodes close to the target waypoint or in complex sea conditions are important. A learnable embedding vector is assigned to each node according to its importance and added to the node features. The expression is: ; In the formula, is the initial embedding vector of node 𝑖, is the original feature vector of node 𝑖, is the learnable embedding vector assigned according to the importance of node .

[0016] In step S1, the spatial encoding is used to capture the spatial relationship between nodes in the ship trajectory. The distance or direction between two adjacent trajectory points is used as a measure of the spatial relationship. A learnable bias term is assigned to each pair of nodes and added to the softmax attention module of the Transformer: ; In the formula, is the attention score, are the input representations of nodes and respectively, and are the weight matrices of the query Query and the key Key respectively, is the matrix transpose, is the scaling factor, is the learnable scalar indexed according to the distance, is the distance between nodes .

[0017] In step S1, the temporal encoding is used to capture the temporal relationship between nodes in the ship trajectory. The time interval between two trajectory points is used as a measure of the temporal relationship. A learnable bias term is assigned to each pair of nodes and added to the softmax attention module of the Transformer. The expression is: ; In the formula, is the learnable scalar indexed according to the time interval, is the time between nodes .

[0018] In step S2, the SeaGraph model includes multiple cascaded SeaGraph layers, and each layer integrates layer normalization, multi-head self-attention module and feed-forward network; The SeaGraph layer is implemented based on the Transformer encoder, and layer normalization LN is applied before the multi-head self-attention MHA and the feed-forward network FFN. The specific calculation is as follows: ; ; In the formula, is the node representation of the th layer; is the multi-head self-attention, is the feed-forward neural network, is the normalization, is the node representation of the th layer, is the th layer of the intermediate node representation obtained through multi-head attention.

[0019] In step S2, the spatio-temporal propagation law of the wave field is dynamically captured, and the predicted value of the significant wave height of the target waypoint is output through the dynamic output layer, including: Mapping the global representation to the predicted value of the significant wave height: ; In the formula, is the predicted wave feature, is the linear mapping.

[0020] Another object of the present invention is to provide a ship navigation sea wave dynamic spatio-temporal forecasting system based on deep learning. The system implements the ship navigation sea wave dynamic spatio-temporal forecasting method based on deep learning. The system includes: A multi-source data preprocessing and feature encoding module, which is used to perform spatio-temporal alignment, missing value repair and standardization processing on the collected ship AIS data and ERA5 reanalysis dataset, and generate a node feature vector including longitude and latitude, timestamp, wind speed and significant wave height; the node feature vector quantifies the importance of ship trajectory nodes and spatio-temporal interaction relationships through node feature encoding, centrality encoding, spatial encoding and time encoding; A SeaGraph model construction module, which is used to construct a SeaGraph model based on the quantification results of the importance of ship trajectory nodes and spatio-temporal interaction relationships, use the Transformer architecture to construct the SeaGraph model, dynamically capture the spatio-temporal propagation law of the wave field through the multi-head self-attention mechanism and graph network fusion, and output the predicted value of the significant wave height of the target waypoint through the dynamic output layer; ​​​​​The model verification module is used to verify the SeaGraph model and analyze its performance through SeaGraph model training, comparative experiments, and extreme event verification.

[0021] Furthermore, the deep learning-based dynamic spatio-temporal prediction system for ship navigation is equipped with a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it can implement the functions in the deep learning-based dynamic spatio-temporal prediction system for ship navigation.

[0022] Combining all the above technical solutions, the beneficial effects of the present invention are as follows: First, by fusing multi-source time-series data of ship historical trajectories and spatio-temporal features of target waypoints, the present invention designs an adaptive spatio-temporal encoding mechanism and constructs a dynamic graph network architecture to capture the spatio-temporal propagation law of the wave field. This method breaks through the spatio-temporal constraints of traditional static modeling frameworks, integrates the advantages of self-attention mechanisms and graph networks, and strengthens the predictive modeling ability for the dynamic evolution of the wave field in front of the ship's navigation trajectory, providing high-precision wave forecasting support for intelligent ship navigation in complex sea conditions.

[0023] Second, through the multi-head self-attention mechanism and spatio-temporal encoding technology, the present invention captures the complex spatio-temporal dependence relationships in the ship's navigation trajectory and the dynamic propagation of the wave field. The model can adaptively quantify the importance of trajectory nodes, dynamically allocate attention weights to historical waypoints, spatially adjacent regions, and temporal evolution trends, break through the spatio-temporal misalignment limitations of traditional static modeling, and achieve high-precision modeling of the evolution law of the wave field in front of the target waypoint.

[0024] Third, compared with traditional numerical models, satellite remote sensing, and fixed-point prediction methods based on RNN / LSTM, the present invention designs node feature encoding, centrality encoding, and spatio-temporal bias terms. The mean absolute error within a 6-hour prediction range is much lower than that of the LSTM model, and the peak error change is stable in extreme events (such as wave mutations). Through the high-precision and low-latency dynamic wave forecasting ability, the present invention can significantly reduce the operating costs of ships caused by bad sea conditions, improve the route optimization efficiency and navigation safety at the same time, provide high-timeliness data support for marine environmental monitoring, disaster warning, and resource exploration, and promote the transformation and upgrading of the shipping industry towards intelligence and greenness. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure; Figure 1 It is a schematic diagram of the principle of the deep learning-based dynamic spatio-temporal prediction method for ship navigation provided by an embodiment of the present invention; Figure 2It is a flowchart of a method for dynamically predicting ship navigation sea waves based on deep learning provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of SeaGraph model construction provided by an embodiment of the present invention; Figure 4 It is a graph showing the change of model error with the same prediction time duration provided by an embodiment of the present invention; Figure 5 It is a comparison graph of the MAE change and extreme stage error of different models in wave mutation events provided by an embodiment of the present invention. Detailed implementation manners

[0026] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the accompanying drawings. Many specific details are set forth in the following description to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein. Those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific implementations disclosed below.

[0027] The innovation of the present invention lies in: the present invention first proposes a dynamic modeling framework that integrates multi-source spatio-temporal feature encoding and an adaptive graph network, breaking through the constraints of the traditional static modeling framework on the spatio-temporal relationship between ship trajectories and wave fields, and realizing high-precision dynamic prediction of the sea wave field in front of the navigation path. Specifically, the present invention integrates ship AIS trajectory data and ERA5 reanalysis environmental data through a multi-source data preprocessing and feature encoding module to construct a multi-dimensional spatio-temporal feature system including node features, centrality encoding, spatial encoding, and time encoding; designs the SeaGraph model based on the Transformer architecture, and dynamically captures the spatio-temporal propagation law of the wave field by fusing the multi-head self-attention mechanism and graph network characteristics through multi-level cascaded SeaGraph layers; further combines comparative experiments and extreme event verification technologies to significantly improve the robustness of the model in long-term prediction and sudden sea conditions. The error increase rate of this method is reduced by 29.6% compared with the traditional LSTM model within the 6-hour prediction range, and the peak error of MAE is reduced by 50% in the wave mutation scenario, providing high-precision and low-latency dynamic environment perception support for intelligent ship navigation and obstacle avoidance decision-making in complex sea conditions.

[0028] The present invention proposes multi-source spatio-temporal feature encoding: integrating the spatio-temporal data of ship trajectories and ocean environmental parameters to construct node, centrality, spatial, and time encoding; Adaptive graph network architecture: integrating the self-attention mechanism and graph network based on SeaGraph to capture the dynamic propagation law of the wave field; Dynamic prediction model optimization: achieving high precision and robustness through training with spatio-temporal alignment datasets and multi-dimensional verification.

[0029] Example 1, as Figure 1 , the method for dynamically forecasting ship navigation sea waves in space-time based on deep learning provided by the embodiment of the present invention breaks through the limitations of the traditional static modeling framework by integrating multi-source time-series data of ship historical trajectories and the spatio-temporal characteristics of target waypoints, designing an adaptive spatio-temporal encoding mechanism, and constructing a dynamic graph network architecture. Its core innovation lies in integrating the advantages of self-attention mechanism and graph network, dynamically capturing the spatio-temporal propagation law of the wave field, and realizing high-precision predictive modeling of the dynamic evolution of sea waves in the unexplored sea area ahead of the ship navigation trajectory, thereby improving the intelligent navigation safety and route optimization decision-making efficiency in complex sea conditions.

[0030] Multi-source spatio-temporal feature encoding technology, adaptive graph network modeling technology, dynamic prediction model training and verification technology. Among them, the multi-source spatio-temporal feature encoding technology is used to capture the spatio-temporal correlation between ship trajectories and the marine environment, the adaptive graph network modeling technology realizes the learning of the dynamic propagation law of the wave field, and the dynamic prediction model training and verification technology ensures the high precision and robustness of the model.

[0031] First, based on the multi-source time-series data (latitude and longitude, time, wind field, sea waves) of ship historical trajectories and the spatio-temporal characteristics of target waypoints, the multi-source spatio-temporal feature encoding technology constructs node feature encoding, centrality encoding, spatial encoding and time encoding. Specifically, it includes: Node feature encoding: Each ship trajectory point is used as a graph node, and its feature vector contains latitude and longitude, time, wind field and sea wave information; Centrality encoding: According to the spatial proximity or sea condition complexity between the node and the target waypoint, a learnable embedding vector is assigned to the key node to enhance the model's attention to important nodes; Spatial encoding: By the distance and direction relationship between adjacent trajectory points, a learnable bias term is assigned to the node pair to quantify the spatial interaction intensity; Time encoding: Based on the time interval between nodes, a time relationship bias term is constructed to capture the dynamic evolution law of the wave field.

[0032] Secondly, design the SeaGraph model based on the Transformer architecture, integrating the multi-head self-attention mechanism and graph network characteristics. The specific implementation is as follows: Construction of SeaGraph layer: Layer normalization (LN), multi-head self-attention (MHA) and feed-forward network (FFN) are introduced into the classical Transformer encoder to capture the long-range dependencies and complex interactions between nodes through adaptive attention weights; Dynamic graph output layer: Use a fully connected network to map the global representation of the graph to the forward wave characteristics (such as significant wave height) to achieve end-to-end dynamic prediction.

[0033] Finally, the dynamic prediction model training and verification technology improves the model performance through multi-source dataset construction and experimental optimization. Specifically, it includes: Dataset construction: Integrate the ship AIS data (time, longitude and latitude) of 830 shipping lanes around the world and the ERA5 reanalysis dataset (wind speed, significant wave height). After spatio-temporal alignment, linear interpolation (repair of missing trajectories) and Z-score standardization, it is divided into an 80% training set and a 20% test set; Model verification: Verify the superiority of the model in terms of indicators such as mean absolute error (MAE) and the increase in long-term prediction error through comparative experiments; Adaptability verification for extreme events: For the wave mutation scenario, through time series error analysis and box plot comparison, prove the prediction stability of the model under extreme sea conditions.

[0034] Through the above technical means, the present invention realizes the high-precision dynamic prediction of the wave field in front of the ship's navigation path, providing reliable environmental perception support for navigation safety and intelligent decision-making in complex sea conditions.

[0035] Embodiment 2, as another possible implementation manner of the present invention, as Figure 2 shown, the method for dynamically predicting the sea waves in the space-time domain of ship navigation provided by the embodiment of the present invention includes: S1. Perform spatio-temporal alignment, missing value repair and standardization processing on the collected ship AIS data and ERA5 reanalysis dataset to generate a node feature vector containing longitude and latitude, timestamp, wind speed and significant wave height; the node feature vector quantifies the importance of ship trajectory nodes and spatio-temporal interaction relationships through node feature encoding, centrality encoding, spatial encoding and time encoding; S2. Based on the quantification results of the importance of ship trajectory nodes and spatio-temporal interaction relationships, use the Transformer architecture to construct a SeaGraph model, dynamically capture the spatio-temporal propagation law of the wave field through the multi-head self-attention mechanism and graph network fusion, and output the predicted value of the significant wave height of the target waypoint through the dynamic output layer; S3. Perform SeaGraph model verification and performance analysis through SeaGraph model training, comparative experiments and extreme event verification.

[0036] Verify the prediction accuracy and stability of the model in normal sea conditions and extreme events through systematic experiments, and comparatively analyze its advantages over the prior art.

[0037] Exemplarily, in step S1, by integrating ship trajectory data and ocean environmental parameters, a spatio-temporal feature encoding system is constructed to provide standardized data for the input of the SeaGraph model. First, collect the ship AIS data of 830 global routes and the ERA5 reanalysis dataset. Through spatio-temporal alignment, missing value repair, and standardization processing, generate node feature vectors containing longitude, latitude, timestamp, wind speed, and significant wave height. Further, through centrality encoding, spatial encoding, and temporal encoding, quantify the node importance and spatio-temporal interaction relationship, and enhance the SeaGraph model's ability to capture the dynamic propagation law of the wave field.

[0038] Specifically, step S1 specifically includes: S101, data collection and integration; Ship trajectory data: Extract AIS data from global routes, including timestamp (UTC), longitude, latitude, speed, and heading, with a sampling interval of 0.5 - 3 hours.

[0039] Ocean environmental data: Use the ERA5 reanalysis dataset to extract the 10-meter height wind speed and significant wave height of the target sea area, with spatio-temporal resolutions of 1 hour and 0.25°×0.25°, respectively.

[0040] Spatio-temporal alignment: Match ship trajectory points with ERA5 grid points in terms of time and space. The present invention innovatively proposes that non-integer grid points use bilinear interpolation to complete the parameters: ; In the formula, is the ship position, is the weight of the th grid point, is the value at a certain position, are the 4 nearest ERA5 grid points, and the weights are determined by the inverse square of the distance.

[0041] S102, data cleaning and standardization.

[0042] Missing trajectory repair: The present invention innovatively proposes that for trajectory segments with continuous missing for more than 3 hours, linear interpolation is used to complete: ; In the formula, is the longitude and latitude position at time and are both the longitude and latitude positions of adjacent valid track points, are both time intervals, ; Standardization processing: The present invention innovatively proposes to perform Z-score standardization on wind speed and significant wave height: ; In the formula, is the value after normalization, is the value before normalization, is the mean of the training set, is the standard deviation.

[0043] S103, Spatiotemporal Feature Encoding.

[0044] Node Feature Encoding: Each ship trajectory point is regarded as a node, and its feature vector includes longitude, latitude, time, wind field, and wave information. Assume that the ship trajectory has nodes, and the feature vector of each node is expressed as , where is the feature dimension; is the real number space with dimension . Centrality Encoding: To capture the importance of key nodes in the ship trajectory, centrality encoding is adopted. Nodes close to the target waypoint or in complex sea conditions may be more important. Specifically, a learnable embedding vector is assigned to each node according to its importance and added to the node features: ; In the formula, is the initial embedding vector of node 𝑖, is the original feature vector of node 𝑖, is the learnable embedding vector assigned according to the importance of node .

[0045] Spatial Encoding: To capture the spatial relationship between nodes in the ship trajectory, spatial encoding is adopted. The distance or direction between two adjacent trajectory points is used as a measure of the spatial relationship. Specifically, a learnable bias term is assigned to each pair of nodes and added to the softmax attention module of the Transformer: ; In the formula, is the attention score, are the input representations of nodes and respectively, and are the weight matrices of the query Query and the key Key respectively, is the matrix transpose, is the scaling factor, is the learnable scalar indexed by the distance, is the distance between nodes .

[0046] Temporal Encoding: To capture the temporal relationships between nodes in the ship trajectory, temporal encoding is adopted. The time interval between two trajectory points is used as a measure of the temporal relationship. Specifically, a learnable bias term is assigned to each pair of nodes and added to the softmax attention module of the Transformer: ; where, is a learnable scalar indexed by the time interval, is the time between nodes .

[0047] Exemplarily, as shown in Figure 3 , the step S2 specifically includes: S201, Model architecture design.

[0048] SeaGraph layer: The SeaGraph layer is implemented based on the Transformer encoder, but layer normalization (LN) is applied before the multi-head self-attention (MHA) and the feed-forward network (FFN). The specific calculation is as follows: ; ; where, is the node representation of the th layer; is the multi-head self-attention, is the feed-forward neural network, is the normalization, is the node representation of the th layer, is the th intermediate node representation obtained through multi-head attention.

[0049] S202, Dynamic graph output layer: Map the global representation to the significant wave height prediction value: ; where, is the predicted wave feature, is the linear mapping.

[0050] Exemplarily, the step S3 specifically includes: S301, Model training.

[0051] Dataset division: Divide the training set and the test set in the ratio of 8:2 to ensure balanced route distribution.

[0052] Loss function: Adopt the smooth L1 loss: ; Optimization strategy: Use the Adam optimizer with an initial learning rate , and weight decay , and train for 100 epochs.

[0053] S302, Comparative experiment: Compare with the prediction accuracies of the ANN and LSTM models, and calculate the error increase rates for 1 - 6 hours of different models ( Figure 4 ). SeaGraph shows the lowest MAE in all prediction time ranges, and its error growth rate is relatively small. This verifies the advantage of SeaGraph in processing multivariate time - series data of ship historical trajectories, especially in long - term prediction tasks. In contrast, the error growth of the ANN model is the most significant, mainly because it cannot effectively capture complex patterns and long - range dependencies in time - series data. Although LSTM can utilize the sequence memory mechanism to a certain extent, its performance is still inferior to the SeaGraph model based on self - attention and graph network.

[0054] S303, Extreme event verification: Select a dynamic process where the wave height rapidly rises and then gradually decays for prediction accuracy verification ( Figure 5 ). In the figure, the red area marks the key stage where the wave height significantly increases. SeaGraph (blue) shows the most stable MAE fluctuations throughout the process. Even in the extreme stage (red area) where the wave height rapidly rises, its error increase rate is relatively small, with a peak MAE of about 0.2, lower than that of LSTM (orange, peak about 0.3) and ANN (green, peak about 0.4). This indicates that SeaGraph can maintain the prediction accuracy more effectively when dealing with wave mutation events.

[0055] Compared with the prior art, the present invention significantly improves the accuracy and dynamic adaptability of on - ship sea - wave forecasting by integrating the adaptive graph network architecture of SeaGraph and the multi - source spatio - temporal feature encoding mechanism. Experiments show that its mean absolute error and mean square error are lower than those of the traditional LSTM model, and the error increase rate within the 6 - hour prediction range is only 29.6%, lower than 40.8% of the LSTM model, effectively solving the problems of high computational resource consumption in traditional numerical simulations and insufficient spatio - temporal dynamic modeling of existing deep - learning methods. At the same time, through spatio - temporal alignment of multi - source data integration and lightweight model design, the complexity of data acquisition and processing is reduced, avoiding the timeliness bottleneck of satellite remote sensing and buoy observations. In addition, the model still maintains stable prediction ability in wave mutation scenarios, providing high - precision and low - latency dynamic environment perception support for intelligent ship navigation, and significantly improving the navigation safety and decision - making efficiency in complex sea conditions.

[0056] Example 3, The ship navigation sea - wave dynamic spatio - temporal forecasting system based on deep learning provided by the embodiment of the present invention includes: The multi-source data preprocessing and feature encoding module is used to perform spatio-temporal alignment, missing value repair, and standardization processing on the collected ship AIS data and ERA5 reanalysis dataset, and generate a node feature vector containing longitude and latitude, timestamp, wind speed, and significant wave height; the node feature vector quantifies the importance of ship trajectory nodes and spatio-temporal interaction relationships through node feature encoding, centrality encoding, spatial encoding, and temporal encoding; The SeaGraph model construction module is used to construct the SeaGraph model based on the quantification results of the importance of ship trajectory nodes and spatio-temporal interaction relationships, and utilize the Transformer architecture. Through the fusion of the multi-head self-attention mechanism and the graph network, it dynamically captures the spatio-temporal propagation law of the wave field, and outputs the predicted value of the significant wave height of the target waypoint through the dynamic output layer; The model verification module is used to verify the SeaGraph model and analyze its performance through SeaGraph model training, comparative experiments, and extreme event verification.

[0057] As mentioned above, only the relatively optimal specific implementation manner of the present invention is described, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention by those skilled in the art within the technical scope disclosed by the present invention shall be covered by the protection scope of the present invention.

Claims

1. A method for dynamically predicting the temporal and spatial conditions of ocean waves during ship navigation based on deep learning, characterized in that, The method includes the following steps: S1. Spatially and temporally align the collected ship AIS data with the ERA5 reanalysis dataset, repair missing values, and perform standardization processing to generate a node feature vector containing longitude, latitude, timestamp, wind speed, and significant wave height. The node feature vector quantifies the importance of ship trajectory nodes and spatio-temporal interaction relationships through node feature encoding, centrality encoding, spatial encoding, and temporal encoding. S2. Based on the quantification results of the importance of ship trajectory nodes and spatio-temporal interaction relationships, use the Transformer architecture to construct a SeaGraph model. Through the fusion of the multi-head self-attention mechanism and the graph network, dynamically capture the spatio-temporal propagation law of the wave field, and output the predicted value of the significant wave height of the target waypoint through the dynamic output layer. S3. Conduct SeaGraph model verification and performance analysis through SeaGraph model training, comparative experiments, and extreme event verification.

2. The method for dynamically forecasting the sea waves during ship navigation based on deep learning according to claim 1, wherein In step S1, the spatio-temporal alignment of the collected ship AIS data with the ERA5 reanalysis dataset includes: The ship trajectory data extracts AIS data from the global shipping routes, including timestamp UTC, longitude, latitude, speed, and heading, with a sampling interval of 0.5 - 3 hours. For the marine environmental data, use the ERA5 reanalysis dataset to extract the 10-meter height wind speed and significant wave height of the target sea area. For spatio-temporal alignment, match the ship trajectory points with the ERA5 grid points in terms of time and space. For non-integer grid points, use bilinear interpolation to complete the parameters. The expression is: ； Where, is the ship's position, For the The weight of the grid points, is the value at a certain position, The nearest four ERA5 grid points, with weights determined by the inverse square of the distance.

3. The method for dynamically forecasting the ship navigation sea waves in space and time based on deep learning according to claim 1, wherein In step S1, the repair of missing values and standardization processing of the collected ship AIS data with the ERA5 reanalysis dataset include: For missing trajectory repair, for trajectory segments with continuous missing values exceeding 3 hours, use linear interpolation to complete. The expression is: ； Wherein, is the latitude and longitude position at the moment and are both the latitude and longitude positions of adjacent valid track points, are both time intervals, ; For standardization processing, perform Z-score standardization on the wind speed and significant wave height. The expression is: ； In the formula, is the value after normalization, is the value before normalization, is the mean value of the training set, is the standard deviation.

4. The method for dynamically forecasting the sea waves during ship navigation based on deep learning according to claim 1, wherein, In step S1, node feature encoding includes: regarding each ship trajectory point as a node, whose feature vector contains longitude, latitude, time, wind field and wave information; the ship trajectory has nodes, and the feature vector of each node is expressed as , where is the feature dimension; is the real number space with the dimension of ; Centrality encoding is used to capture the importance of key nodes in the ship trajectory. Nodes close to the target waypoint or in complex sea conditions are important. Assign a learnable embedding vector to each node according to its importance and add it to the node features. The expression is: ； In the formula, is the initial embedding vector of node 𝑖, is the original feature vector of node 𝑖, is the learnable embedding vector assigned according to the importance of node ​ 5. The method for dynamically predicting the sea waves during ship navigation based on deep learning according to claim 1, wherein In step S1, spatial encoding is used to capture the spatial relationship between nodes in the ship trajectory, and the distance or direction between two adjacent trajectory points is used as a measure of the spatial relationship. A learnable bias term is assigned to each pair of nodes and added to the softmax attention module of the Transformer: ； In the formula, is the attention score, are the input representations of nodes and respectively, and are the weight matrices of query Query and key Key respectively, is the matrix transpose, is the scaling factor, is the learnable scalar indexed according to the distance, is the distance between nodes respectively.

6. The method for dynamically forecasting the sea waves during ship navigation based on deep learning according to claim 5, characterized in that, In step S1, temporal encoding is used to capture the temporal relationship between nodes in the ship trajectory. The time interval between two trajectory points is used as a measure of the temporal relationship. Assign a learnable bias term to each pair of nodes and add it to the softmax attention module of the Transformer. The expression is: ； wherein, is a learnable scalar indexed according to the time interval, is the node is the time between.

7. The method for dynamically predicting the sea waves during ship navigation based on deep learning according to claim 1, characterized in that, In step S2, the SeaGraph model includes multiple cascaded SeaGraph layers, and each layer integrates layer normalization, multi-head self-attention module, and feed-forward network. The SeaGraph layer is implemented based on the Transformer encoder. Layer normalization LN is applied before the multi-head self-attention MHA and the feed-forward network FFN. The specific calculation is as follows: ； ； In the formula, represents the node of the layer; is the multi-head self-attention, is the feed-forward neural network, is the normalization, is the node representation of the layer, is the intermediate node representation obtained after the layer passes through the multi-head attention.

8. The method for dynamically predicting the sea waves during ship navigation based on deep learning according to claim 1, wherein In step S2, dynamically capture the spatio-temporal propagation law of the wave field, and output the predicted value of the significant wave height of the target waypoint through the dynamic output layer, including: Map the global representation to the significant wave height prediction value: ； In the formula, is the predicted wave feature, is the linear mapping.

9. A ship navigation sea wave dynamic spatio-temporal prediction system based on deep learning, characterized in that, Implement the deep learning-based dynamic spatio-temporal prediction method for ship navigation waves according to any one of claims 1 - 8. The system includes: The multi-source data preprocessing and feature encoding module is used to perform spatio-temporal alignment, missing value repair, and normalization processing on the collected ship AIS data and ERA5 reanalysis dataset, and generate a node feature vector containing longitude and latitude, timestamp, wind speed, and significant wave height; the node feature vector quantifies the importance of ship trajectory nodes and spatio-temporal interaction relationships through node feature encoding, centrality encoding, spatial encoding, and temporal encoding; The SeaGraph model construction module is used to construct the SeaGraph model based on the quantification results of the importance of ship trajectory nodes and spatio-temporal interaction relationships, using the Transformer architecture. Through the fusion of the multi-head self-attention mechanism and the graph network, it dynamically captures the spatio-temporal propagation law of the wave field, and outputs the predicted value of the significant wave height of the target waypoint through the dynamic output layer; The model verification module is used to verify and analyze the performance of the SeaGraph model through SeaGraph model training, comparative experiments, and extreme event verification.

10. The ship navigation wave dynamic spatio-temporal prediction system based on deep learning according to claim 9, characterized in that, The deep learning-based dynamic spatio-temporal prediction system for ship navigation is equipped with a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it can implement the functions in the deep learning-based dynamic spatio-temporal prediction system for ship navigation.

Images