Marine ice condition dynamic evaluation method and device

Abstract

The invention discloses an ocean ice condition dynamic evaluation method and device, and relates to the technical field of ocean environment monitoring, and the method comprises the steps: obtaining the historical data of the sea ice thickness of a target sea area, analyzing the temporal and spatial variation trend, and building a sea ice thickness initial digital twin; dynamically correcting the initial digital twin; a space-time self-attention mechanism model is trained, sea ice thickness probabilistic prediction is achieved, multi-climate scenes and high-intensity human activities are set as external constraints, sea ice thickness initial digital twinborn antagonism deduction after dynamic correction is driven, and sea ice thickness extreme change and critical mutation risks under different external pressures are quantitatively evaluated; based on quantitative evaluation of extreme change and critical sudden change risks of sea ice thickness under different external pressures, a sea ice thickness time-varying risk field is dynamically generated, and a self-adaptive optimal route plan with a dynamic risk avoidance capability is provided for a ship. According to the invention, the safety of polar navigation is improved.

Description

Technical Field

[0001] This invention relates to the field of marine environmental monitoring technology, specifically to a method and apparatus for dynamic assessment of marine ice conditions. Background Technology

[0002] Current methods for assessing ocean ice conditions largely rely on historical statistical data and static, deterministic forecasting models. They struggle to deeply integrate multiple key driving factors, such as temperature changes and high-intensity human activities, resulting in insufficient characterization of the dynamic changes in sea ice thickness. Furthermore, they generally lack dynamic correction capabilities based on digital twins and real-time data, causing the accuracy and timeliness of models to rapidly decline over time. Most critically, existing technologies cannot effectively simulate extreme and nonlinear abrupt changes under multiple climate scenarios. Their risk assessments are mostly static or probabilistic averages, neglecting the possibility of critical abrupt changes. This leads to insufficient risk avoidance capabilities in the face of rapidly changing extreme ice conditions, making it difficult to support safe and economical optimal adaptive navigation decisions. Summary of the Invention

[0003] To address the aforementioned problems, this application proposes a method for dynamic assessment of ocean ice conditions, and this technical solution resolves the problems.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for dynamic assessment of ocean ice conditions includes: S1. Obtain historical data on sea ice thickness in the target sea area, analyze the spatiotemporal variation trend of sea ice thickness, and establish an initial digital twin of sea ice thickness; S2. Introduce the temperature change and human activity intensity of the target sea area as key driving factors for sea ice thickness, integrate real-time sea ice thickness observation data of the target sea area, and dynamically correct the initial digital twin of sea ice thickness; S3. Based on the dynamically corrected initial digital twin of sea ice thickness, train the spatiotemporal self-attention mechanism model to achieve probabilistic prediction of sea ice thickness in the target sea area. Set multiple climate scenarios and high-intensity human activities as external constraints to drive the adversarial simulation of the dynamically corrected initial digital twin of sea ice thickness and quantitatively assess the risk of extreme changes and critical mutations in sea ice thickness under different external pressures. S4. Based on the quantitative assessment of the risks of extreme changes and critical abrupt changes in sea ice thickness under different external pressures, a time-varying risk field of sea ice thickness is dynamically generated to provide ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

[0005] Furthermore, step S1 specifically includes: Based on NASA's official website, historical data on sea ice thickness in the target area were obtained, unified to the same grid coordinates, and spatiotemporal registration and data preprocessing were performed. The historical data includes: sea ice thickness, timestamps, and latitude and longitude coordinates; Based on the UPI buoy database, real-time observation data of sea ice thickness in the target sea area were obtained, and the data format was standardized. Using historical sea ice thickness data of the target sea area as the background field and observed sea ice thickness data of the target sea area as the actual observed value, the analysis value of the background field and the actual observed value is calculated for each grid point using a weighted average formula. Calculate the Gaussian function values ​​of the background field and the actual observation at the same time, assign weights to the background field and the actual observation, and obtain the sea ice thickness field of the target sea area. Based on the sea ice thickness field of the target sea area, the sea ice thickness M of the target sea area is generated. An N-dimensional matrix, where M represents the number of time points and N represents the number of spatial grid points; Calculate the temporal average value of each spatial grid point over the entire time series; Based on the sea ice thickness M in the target sea area The N two-dimensional matrix is ​​used to calculate the sea ice thickness anomaly field matrix of the target sea area by subtracting the time series and time average of each spatial grid point. Calculate the covariance matrix of the sea ice thickness anomaly field in the target sea area; The covariance matrix of the sea ice thickness anomaly field in the target sea area is calculated by eigenvalue decomposition, and the eigenvalues ​​and eigenvectors of the covariance matrix of the sea ice thickness anomaly field in the target sea area are obtained. Based on the sea ice thickness anomaly field matrix of the target sea area, the principal component time series of sea ice thickness in the target sea area is obtained by projecting it onto the eigenvector of the covariance matrix of the sea ice thickness anomaly field of the target sea area.

[0006] Furthermore, step S1 also includes: Based on the sea ice thickness field of the target sea area, the spatial field of sea ice thickness in the target sea area is obtained by using the Kriging interpolation method. Using the historical time series of sea ice thickness data of the target sea area at each grid point as input features, the time series of the top k principal components of the sea ice thickness of the target sea area as supplementary input features, and the sea ice thickness of the target sea area at a certain future moment as the prediction target, an LSTM long short-term memory network supervised learning dataset is constructed. A neural network containing several LSTM long short-term memory network layers is constructed. The LSTM long short-term memory network layers and their number of neurons are used as hidden layers. Fully connected layers are introduced to obtain the sea ice predicted thickness at each grid point and establish a sea ice thickness prediction model. Using a sea ice thickness prediction model as a simulation engine, an initial digital twin of sea ice thickness is established.

[0007] Furthermore, step S2 specifically includes: Based on the weather station, hourly temperature change data of the target sea area is obtained and the data is preprocessed. Based on AIS data stream, ship navigation density and trajectory data are obtained, dense areas of ship activity in the target sea area are identified, ship activity density and frequency in the target sea area are calculated, and the intensity of human activities in the target sea area is quantitatively assessed. Based on the sea ice thickness prediction model, the previous sea ice thickness data of the target sea area is used as input, and the temperature change of the target sea area and the intensity of human activities are introduced as key driving factors for sea ice thickness. The current predicted value of sea ice thickness in the target sea area is obtained, and a sea ice thickness prediction model forecaster is constructed. Using real-time observation data of sea ice thickness in the target sea area as input, and coupled with the sea ice thickness prediction model forecaster, the residual sequence of sea ice thickness observation value-predicted value in the target sea area is obtained. Using the Kalman filter algorithm, the current predicted value of sea ice thickness in the target sea area is used as the prediction step, and the residual sequence of observed value-predicted value of sea ice thickness in the target sea area is used as the observation update quantity. The Kalman gain of sea ice thickness in the target sea area is calculated and iterated continuously to dynamically correct the initial digital twin of sea ice thickness.

[0008] Furthermore, step S3 specifically includes: Based on historical sea ice thickness data of the target sea area, spatiotemporal alignment and preprocessing are performed to construct a spatiotemporal dataset of sea ice thickness in the target sea area; Using the spatiotemporal dataset of sea ice thickness in the target sea area as input, a fixed sliding time window is set, and the real value of sea ice thickness in the target sea area in a certain future segment within the sliding time window is used as the prediction target to construct a training sample set for the spatiotemporal self-attention mechanism model. According to the time sequence, the training sample set of the spatiotemporal self-attention mechanism model is divided into training set, validation set and test set; Each grid point in the target sea area is mapped to a high-dimensional vector as the input embedding layer along with the spatiotemporal features of sea ice thickness. Spatial attention encoding is performed, and the spatiotemporal feature association weights of any grid point with other grid points are calculated to capture the spatial long-range association of sea ice thickness in the target sea area. The importance weight of each time step in the historical data of sea ice thickness in the target sea area for the predicted sea ice thickness is calculated. Through multi-layer attention coding, a comprehensive feature representation of the spatiotemporal evolution of sea ice thickness in the target sea area is obtained. Based on the comprehensive feature representation of the spatiotemporal evolution law of sea ice thickness in the target sea area, the decoder is used to generate the predicted probability distribution of sea ice thickness in the target sea area for each grid point at a future time step. The mean and variance of the probability distribution of sea ice thickness prediction for the target sea area at each grid point and future time step are calculated. A negative log-likelihood loss function is introduced to achieve probabilistic prediction of sea ice thickness in the target sea area.

[0009] Furthermore, step S3 also includes: Based on the target sea area, several representative greenhouse gas concentration pathways were set to simulate the long-term impact of different greenhouse gases on sea ice thickness in the target sea area. Based on shipping thermal pollution, black carbon emissions, and resource extraction disturbances, the regional impact of high-intensity human activities on sea ice thickness in the target sea area is simulated. Set a safety threshold for extreme changes in sea ice thickness. If the sea ice thickness is less than 0.5 meters, the sea area is deemed unsuitable for safe navigation. If the sea ice thickness is less than 1.0 meter, the multi-year ice in the sea area is deemed to have basically disappeared. If the sea ice area in the target sea area decreases by more than 100,000 square kilometers for three consecutive days, it is defined as a rapid collapse event; if the sea ice thickness increases by more than 0.3 meters within a week, it is defined as a rapid generation event. Multiple climate scenarios and high-intensity human activities are set as external constraints to drive the adversarial simulation of the initial digital twin of sea ice thickness after dynamic correction. Set the starting date for the simulation, use the sea ice thickness observation data of the target sea area on that date as the initial state of the dynamically corrected initial digital twin of the sea ice thickness, substitute it into the spatiotemporal self-attention mechanism model, and use the predicted probability distribution of the sea ice thickness of the target sea area in the first future time step as the output. Randomly sample the predicted probability distribution of sea ice thickness in the target sea area at the first future time step to obtain the sea ice thickness field of the target sea area determined by the corresponding predicted probability distribution, which is one of the future states to be realized for the sea ice thickness in the target sea area. Using the sea ice thickness field of the target sea area as the new initial state, and inputting the dynamically corrected initial digital twin of the sea ice thickness, the next step is to predict the probability distribution of the sea ice thickness of the target sea area. This process is iterated and looped continuously to advance the prediction.

[0010] Furthermore, step S3 also includes: Based on the output data of the adversarial simulation of the initial digital twin of sea ice thickness after dynamic correction, the number of occurrences of various critical events and abrupt changes in the target sea area is counted, the frequency of critical abrupt changes in the target sea area is calculated, and the risk of extreme changes and critical abrupt changes in sea ice thickness under different external pressures is quantitatively assessed.

[0011] Furthermore, step S4 specifically includes: Based on the dynamically corrected initial digital twin output data of sea ice thickness during adversarial simulation periods, an S... D F G represents the simulated sea ice thickness matrix of the target sea area, S represents the number of adversarial simulations of the initial digital twin of the sea ice thickness after dynamic correction, D represents the adversarial simulation time step, F represents the longitude of the adversarial simulation of the target sea area, and G represents the latitude of the adversarial simulation of the target sea area. The minimum sea ice thickness required for safe navigation is set at 0.5 m. Calculate the value of each grid point in S D F The average sea ice thickness at each grid point is obtained by taking the average sea ice thickness at all adversarial simulation times and time steps in the sea ice thickness simulation matrix for the target sea area G. Statistics and Identification S D F Calculate the sea ice condition probability risk value for all grid points in the sea ice thickness simulation matrix of target sea area G that are less than 0.5 times the safe navigation sea ice thickness. Calculate S for each grid point D F Standard deviation of all adversarial simulations and time steps in the simulation matrix of sea ice thickness in target sea area G; Based on the average value and standard deviation of sea ice thickness in adversarial simulations, the extreme risk index of sea ice conditions is calculated, and a time-varying dynamic risk field of sea ice thickness is generated.

[0012] Furthermore, step S4 also includes: Calculate the travel time from the ship's current point to the target point to obtain the ship's time cost; Calculate the fuel cost for this leg of the voyage based on the ship's engine power consumption and fuel efficiency; By integrating extreme changes and critical abrupt changes in sea ice thickness under different external pressures, multiple climate scenarios, and high-intensity human activities, the risk cost of the ship is calculated. The ship's time cost, fuel cost, and risk cost are normalized and weighted and integrated to construct a comprehensive ship cost function; With minimizing the ship's comprehensive cost function as the optimization objective, environmental information is updated in real time, and the route planning is continuously adjusted according to the time-varying dynamic risk field of sea ice thickness to dynamically avoid high-risk sea areas, providing ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

[0013] A dynamic assessment device for ocean ice conditions includes: The system acquires historical data on sea ice thickness in the target sea area, analyzes the spatiotemporal variation trend of sea ice thickness, and establishes an initial digital twin of sea ice thickness. It introduces the temperature change and human activity intensity in the target sea area as key driving factors for sea ice thickness, integrates real-time sea ice thickness observation data of the target sea area, and dynamically corrects the initial digital twin of sea ice thickness. The processing device trains a spatiotemporal self-attention mechanism model based on the dynamically corrected initial digital twin of sea ice thickness to achieve probabilistic prediction of sea ice thickness in the target sea area. It sets multiple climate scenarios and high-intensity human activities as external constraints to drive the adversarial simulation of the dynamically corrected initial digital twin of sea ice thickness and quantitatively assess the risk of extreme changes and critical mutations in sea ice thickness under different external pressures. The device provides a mechanism that dynamically generates a time-varying risk field for sea ice thickness based on a quantitative assessment of the risks of extreme changes and critical abrupt changes in sea ice thickness under different external pressures, providing ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

[0014] Beneficial effects: This invention proposes a dynamic assessment scheme for ocean ice conditions. By constructing and dynamically correcting a digital twin of sea ice thickness, integrating multi-source driving factors and real-time observation data, and utilizing spatiotemporal self-attention mechanisms and adversarial inference, this scheme achieves accurate quantitative assessment of the risks of extreme changes and critical abrupt changes in sea ice thickness under multiple external pressures. Based on this, a time-varying dynamic risk field is generated, providing ships with adaptive optimal route planning that can avoid high-risk sea areas in real time, thereby improving the safety, economy, and risk response capabilities of polar navigation.

[0015] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method for dynamic assessment of ocean ice conditions.

[0018] Figure 2 This is a flowchart of a dynamic assessment device for ocean ice conditions. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0021] Reference Figure 1 As shown, a method for dynamic assessment of ocean ice conditions includes: S1. Obtain historical data on sea ice thickness in the target sea area, analyze the spatiotemporal variation trend of sea ice thickness, and establish an initial digital twin of sea ice thickness; Step S1 specifically includes: Based on NASA's official website, historical data on sea ice thickness in the target area were obtained, unified to the same grid coordinates, and spatiotemporal registration and data preprocessing were performed. The historical data includes: sea ice thickness, timestamps, and latitude and longitude coordinates; Based on the UPI buoy database, real-time observation data of sea ice thickness in the target sea area were obtained, and the data format was standardized. Using historical sea ice thickness data of the target sea area as the background field and observed sea ice thickness data of the target sea area as the actual observed value, the analysis value of the background field and the actual observed value is calculated for each grid point using a weighted average formula. Calculate the Gaussian function values ​​of the background field and the actual observation at the same time, assign weights to the background field and the actual observation, and obtain the sea ice thickness field of the target sea area. Based on the sea ice thickness field of the target sea area, the sea ice thickness M of the target sea area is generated. An N-dimensional matrix, where M represents the number of time points and N represents the number of spatial grid points; Calculate the temporal average value of each spatial grid point over the entire time series; Based on the sea ice thickness M in the target sea area The N two-dimensional matrix is ​​used to calculate the sea ice thickness anomaly field matrix of the target sea area by subtracting the time series and time average of each spatial grid point. Calculate the covariance matrix of the sea ice thickness anomaly field in the target sea area; The covariance matrix of the sea ice thickness anomaly field in the target sea area is calculated by eigenvalue decomposition, and the eigenvalues ​​and eigenvectors of the covariance matrix of the sea ice thickness anomaly field in the target sea area are obtained. Based on the sea ice thickness anomaly field matrix of the target sea area, the principal component time series of sea ice thickness in the target sea area is obtained by projecting it onto the eigenvector of the covariance matrix of the sea ice thickness anomaly field of the target sea area.

[0022] Step S1 also includes: Based on the sea ice thickness field of the target sea area, the spatial field of sea ice thickness in the target sea area is obtained by using the Kriging interpolation method. Using the historical time series of sea ice thickness data of the target sea area at each grid point as input features, the time series of the top k principal components of the sea ice thickness of the target sea area as supplementary input features, and the sea ice thickness of the target sea area at a certain future moment as the prediction target, an LSTM long short-term memory network supervised learning dataset is constructed. A neural network containing several LSTM long short-term memory network layers is constructed. The LSTM long short-term memory network layers and their number of neurons are used as hidden layers. Fully connected layers are introduced to obtain the sea ice predicted thickness at each grid point and establish a sea ice thickness prediction model. Using a sea ice thickness prediction model as a simulation engine, an initial digital twin of sea ice thickness is established.

[0023] When using it, please refer to the steps outlined above: Existing sea ice thickness prediction studies generally suffer from shortcomings such as single historical data sources, insufficient fusion of spatiotemporal features, and weak integration of physical mechanisms and data-driven approaches. They often rely on single satellite observations or simple statistical methods, failing to effectively integrate multi-source heterogeneous data and capture the nonlinear spatiotemporal dynamics of sea ice evolution. This step, by fusing NASA historical data with real-time observations from UPI buoys and combining principal component analysis with LSTM long short-term memory neural networks, constructs a prediction framework that integrates data assimilation and machine learning. Its beneficial effects include improving the accuracy and robustness of sea ice thickness spatiotemporal trend analysis, achieving a leap from static description to dynamic prediction, providing a high-fidelity simulation engine for digital twins, and supporting subsequent sea ice disaster early warning and decision optimization.

[0024] S2. Introduce the temperature change and human activity intensity of the target sea area as key driving factors for sea ice thickness, integrate real-time sea ice thickness observation data of the target sea area, and dynamically correct the initial digital twin of sea ice thickness; Step S2 specifically includes: Based on the weather station, hourly temperature change data of the target sea area is obtained and the data is preprocessed. Based on AIS data stream, ship navigation density and trajectory data are obtained, dense areas of ship activity in the target sea area are identified, ship activity density and frequency in the target sea area are calculated, and the intensity of human activities in the target sea area is quantitatively assessed. Based on the sea ice thickness prediction model, the previous sea ice thickness data of the target sea area is used as input, and the temperature change of the target sea area and the intensity of human activities are introduced as key driving factors for sea ice thickness. The current predicted value of sea ice thickness in the target sea area is obtained, and a sea ice thickness prediction model forecaster is constructed. Using real-time observation data of sea ice thickness in the target sea area as input, and coupled with the sea ice thickness prediction model forecaster, the residual sequence of sea ice thickness observation value-predicted value in the target sea area is obtained. Using the Kalman filter algorithm, the current predicted value of sea ice thickness in the target sea area is used as the prediction step, and the residual sequence of observed value-predicted value of sea ice thickness in the target sea area is used as the observation update quantity. The Kalman gain of sea ice thickness in the target sea area is calculated and iterated continuously to dynamically correct the initial digital twin of sea ice thickness.

[0025] When using it, please refer to the steps outlined above: Existing sea ice prediction studies generally suffer from key shortcomings such as insufficient consideration of multi-source dynamic driving factors and a lack of real-time data assimilation capabilities in models. Traditional methods often rely on static climate indicators, ignoring the real-time thermodynamic impact of human activities. Observational data and prediction models are often independent of each other, resulting in limited accuracy of digital twins and significant forecast lag. This step constructs a dual-driven forecaster by fusing quantitative data on temperature changes and ship activities, and introduces Kalman filtering to achieve dynamic closed-loop correction of observation-prediction residuals. This overcomes the limitations of single driving factors and model drift problems, improves the real-time accuracy and dynamic response capability of the digital twin of sea ice thickness, and provides more reliable decision support for polar navigation safety and ecological assessment.

[0026] S3. Based on the dynamically corrected initial digital twin of sea ice thickness, train the spatiotemporal self-attention mechanism model to achieve probabilistic prediction of sea ice thickness in the target sea area. Set multiple climate scenarios and high-intensity human activities as external constraints to drive the adversarial simulation of the dynamically corrected initial digital twin of sea ice thickness and quantitatively assess the risk of extreme changes and critical mutations in sea ice thickness under different external pressures. Step S3 specifically includes: Based on historical sea ice thickness data of the target sea area, spatiotemporal alignment and preprocessing are performed to construct a spatiotemporal dataset of sea ice thickness in the target sea area; Using the spatiotemporal dataset of sea ice thickness in the target sea area as input, a fixed sliding time window is set, and the real value of sea ice thickness in the target sea area in a certain future segment within the sliding time window is used as the prediction target to construct a training sample set for the spatiotemporal self-attention mechanism model. According to the time sequence, the training sample set of the spatiotemporal self-attention mechanism model is divided into training set, validation set and test set; Each grid point in the target sea area is mapped to a high-dimensional vector as the input embedding layer along with the spatiotemporal features of sea ice thickness. Spatial attention encoding is performed, and the spatiotemporal feature association weights of any grid point with other grid points are calculated to capture the spatial long-range association of sea ice thickness in the target sea area. The importance weight of each time step in the historical data of sea ice thickness in the target sea area for the predicted sea ice thickness is calculated. Through multi-layer attention coding, a comprehensive feature representation of the spatiotemporal evolution of sea ice thickness in the target sea area is obtained. Based on the comprehensive feature representation of the spatiotemporal evolution law of sea ice thickness in the target sea area, the decoder is used to generate the predicted probability distribution of sea ice thickness in the target sea area for each grid point at a future time step. The mean and variance of the probability distribution of sea ice thickness prediction for the target sea area at each grid point and future time step are calculated. A negative log-likelihood loss function is introduced to achieve probabilistic prediction of sea ice thickness in the target sea area.

[0027] Step S3 also includes: Based on the target sea area, several representative greenhouse gas concentration pathways were set to simulate the long-term impact of different greenhouse gases on sea ice thickness in the target sea area. Based on shipping thermal pollution, black carbon emissions, and resource extraction disturbances, the regional impact of high-intensity human activities on sea ice thickness in the target sea area is simulated. Set a safety threshold for extreme changes in sea ice thickness. If the sea ice thickness is less than 0.5 meters, the sea area is deemed unsuitable for safe navigation. If the sea ice thickness is less than 1.0 meter, the multi-year ice in the sea area is deemed to have basically disappeared. If the sea ice area in the target sea area decreases by more than 100,000 square kilometers for three consecutive days, it is defined as a rapid collapse event; if the sea ice thickness increases by more than 0.3 meters within a week, it is defined as a rapid generation event. Multiple climate scenarios and high-intensity human activities are set as external constraints to drive the adversarial simulation of the initial digital twin of sea ice thickness after dynamic correction. Set the starting date for the simulation, use the sea ice thickness observation data of the target sea area on that date as the initial state of the dynamically corrected initial digital twin of the sea ice thickness, substitute it into the spatiotemporal self-attention mechanism model, and use the predicted probability distribution of the sea ice thickness of the target sea area in the first future time step as the output. Randomly sample the predicted probability distribution of sea ice thickness in the target sea area at the first future time step to obtain the sea ice thickness field of the target sea area determined by the corresponding predicted probability distribution, which is one of the future states to be realized for the sea ice thickness in the target sea area. Using the sea ice thickness field of the target sea area as the new initial state, and inputting the dynamically corrected initial digital twin of the sea ice thickness, the next step is to predict the probability distribution of the sea ice thickness of the target sea area. This process is iterated and looped continuously to advance the prediction.

[0028] Step S3 also includes: Based on the output data of the adversarial simulation of the initial digital twin of sea ice thickness after dynamic correction, the number of occurrences of various critical events and abrupt changes in the target sea area is counted, the frequency of critical abrupt changes in the target sea area is calculated, and the risk of extreme changes and critical abrupt changes in sea ice thickness under different external pressures is quantitatively assessed.

[0029] When using it, please refer to the steps outlined above: Current sea ice prediction research largely relies on traditional numerical models or static statistical learning, which struggles to effectively capture the nonlinear spatiotemporal dependence of sea ice thickness evolution and lacks the ability to quantitatively assess the risks of extreme abrupt changes under multiple external stresses, such as climate scenarios and human activities. This step, by introducing dynamically corrected digital twins and spatiotemporal self-attention mechanisms, enables probabilistic prediction and multi-scenario adversarial simulation of sea ice thickness. This improves the accuracy and interpretability of extreme change identification and critical risk quantitative assessment, providing crucial decision support for polar navigation safety and climate response strategy formulation.

[0030] S4. Based on the quantitative assessment of the risks of extreme changes and critical abrupt changes in sea ice thickness under different external pressures, a time-varying risk field of sea ice thickness is dynamically generated to provide ships with adaptive optimal route planning with dynamic risk avoidance capabilities. Step S4 specifically includes: Based on the dynamically corrected initial digital twin output data of sea ice thickness during adversarial simulation periods, an S... D F G represents the simulated sea ice thickness matrix of the target sea area, S represents the number of adversarial simulations of the initial digital twin of the sea ice thickness after dynamic correction, D represents the adversarial simulation time step, F represents the longitude of the adversarial simulation of the target sea area, and G represents the latitude of the adversarial simulation of the target sea area. The minimum sea ice thickness required for safe navigation is set at 0.5 m. Calculate the value of each grid point in S D F The average sea ice thickness at each grid point is obtained by taking the average sea ice thickness at all adversarial simulation times and time steps in the sea ice thickness simulation matrix for the target sea area G. Statistics and Identification S D F Calculate the sea ice condition probability risk value for all grid points in the sea ice thickness simulation matrix of target sea area G that are less than 0.5 times the safe navigation sea ice thickness. Calculate S for each grid point D F Standard deviation of all adversarial simulations and time steps in the simulation matrix of sea ice thickness in target sea area G; Based on the average value and standard deviation of sea ice thickness in adversarial simulations, the extreme risk index of sea ice conditions is calculated, and a time-varying dynamic risk field of sea ice thickness is generated.

[0031] Step S4 also includes: Calculate the travel time from the ship's current point to the target point to obtain the ship's time cost; Calculate the fuel cost for this leg of the voyage based on the ship's engine power consumption and fuel efficiency; By integrating extreme changes and critical abrupt changes in sea ice thickness under different external pressures, multiple climate scenarios, and high-intensity human activities, the risk cost of the ship is calculated. The ship's time cost, fuel cost, and risk cost are normalized and weighted and integrated to construct a comprehensive ship cost function; With minimizing the ship's comprehensive cost function as the optimization objective, environmental information is updated in real time, and the route planning is continuously adjusted according to the time-varying dynamic risk field of sea ice thickness to dynamically avoid high-risk sea areas, providing ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

[0032] When using it, please refer to the steps outlined above: Current research on polar route planning relies heavily on static sea ice forecasts or single-scenario simulations, lacking quantitative assessments of the risks of extreme sea ice thickness changes and critical abrupt changes under multiple external pressures. Existing risk field constructions do not fully consider the time-varying dynamic characteristics under adversarial simulations, resulting in delayed responses and insufficient avoidance capabilities of route planning to sudden sea ice risks. This step constructs a quantitative time-varying risk field for sea ice thickness by integrating dynamic digital twins and adversarial simulation data, and integrates multi-dimensional cost functions to achieve adaptive optimization. This improves the real-time perception and avoidance capabilities of route planning for dynamic risks, effectively reducing overall navigation costs and the risk of extreme sea ice accidents.

[0033] refer to Figure 2 A dynamic assessment device for ocean ice conditions, comprising: The device acquires historical data on sea ice thickness in the target sea area, analyzes the spatiotemporal variation trend of sea ice thickness, and establishes an initial digital twin of sea ice thickness. It introduces the temperature change and human activity intensity in the target sea area as key driving factors for sea ice thickness, integrates real-time sea ice thickness observation data of the target sea area, and dynamically corrects the initial digital twin of sea ice thickness. The processing device trains a spatiotemporal self-attention mechanism model based on the dynamically corrected initial digital twin of sea ice thickness to achieve probabilistic prediction of sea ice thickness in the target sea area. It sets multiple climate scenarios and high-intensity human activities as external constraints to drive the adversarial simulation of the dynamically corrected initial digital twin of sea ice thickness and quantitatively assess the risk of extreme changes and critical mutations in sea ice thickness under different external pressures. The device provides a mechanism that dynamically generates a time-varying risk field for sea ice thickness based on a quantitative assessment of the risks of extreme changes and critical abrupt changes in sea ice thickness under different external pressures, providing ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

[0034] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

[0035] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for dynamic assessment of ocean ice conditions, characterized in that, include: S1. Obtain historical data on sea ice thickness in the target sea area, analyze the spatiotemporal variation trend of sea ice thickness, and establish an initial digital twin of sea ice thickness; S2. Introduce the temperature change and human activity intensity of the target sea area as key driving factors for sea ice thickness, integrate real-time sea ice thickness observation data of the target sea area, and dynamically correct the initial digital twin of sea ice thickness; S3. Based on the dynamically corrected initial digital twin of sea ice thickness, train the spatiotemporal self-attention mechanism model to achieve probabilistic prediction of sea ice thickness in the target sea area. Set multiple climate scenarios and high-intensity human activities as external constraints to drive the adversarial simulation of the dynamically corrected initial digital twin of sea ice thickness and quantitatively assess the risk of extreme changes and critical mutations in sea ice thickness under different external pressures. S4. Based on the quantitative assessment of the risks of extreme changes and critical abrupt changes in sea ice thickness under different external pressures, a time-varying risk field of sea ice thickness is dynamically generated to provide ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

2. The method for dynamic assessment of marine ice conditions according to claim 1, characterized in that, Step S1 specifically includes: Based on NASA's official website, historical data on sea ice thickness in the target area were obtained, unified to the same grid coordinates, and spatiotemporal registration and data preprocessing were performed. The historical data includes: sea ice thickness, timestamps, and latitude and longitude coordinates; Based on the UPI buoy database, real-time observation data of sea ice thickness in the target sea area were obtained, and the data format was standardized. Using historical sea ice thickness data of the target sea area as the background field and observed sea ice thickness data of the target sea area as the actual observed value, the analysis value of the background field and the actual observed value is calculated for each grid point using a weighted average formula. Calculate the Gaussian function values ​​of the background field and the actual observation at the same time, assign weights to the background field and the actual observation, and obtain the sea ice thickness field of the target sea area. Based on the sea ice thickness field of the target sea area, the sea ice thickness M of the target sea area is generated. An N-dimensional matrix, where M represents the number of time points and N represents the number of spatial grid points; Calculate the temporal average value of each spatial grid point over the entire time series; Based on the sea ice thickness M in the target sea area The N two-dimensional matrix is ​​used to calculate the sea ice thickness anomaly field matrix of the target sea area by subtracting the time series and time average of each spatial grid point. Calculate the covariance matrix of the sea ice thickness anomaly field in the target sea area; The covariance matrix of the sea ice thickness anomaly field in the target sea area is calculated by eigenvalue decomposition, and the eigenvalues ​​and eigenvectors of the covariance matrix of the sea ice thickness anomaly field in the target sea area are obtained. Based on the sea ice thickness anomaly field matrix of the target sea area, the principal component time series of sea ice thickness in the target sea area is obtained by projecting it onto the eigenvector of the covariance matrix of the sea ice thickness anomaly field of the target sea area.

3. The method for dynamic assessment of marine ice conditions according to claim 2, characterized in that, Step S1 also includes: Based on the sea ice thickness field of the target sea area, the spatial field of sea ice thickness in the target sea area is obtained by using the Kriging interpolation method. Using the historical time series of sea ice thickness data of the target sea area at each grid point as input features, the time series of the top k principal components of the sea ice thickness of the target sea area as supplementary input features, and the sea ice thickness of the target sea area at a certain future moment as the prediction target, an LSTM long short-term memory network supervised learning dataset is constructed. A neural network containing several LSTM long short-term memory network layers is constructed. The LSTM long short-term memory network layers and their number of neurons are used as hidden layers. Fully connected layers are introduced to obtain the sea ice predicted thickness at each grid point and establish a sea ice thickness prediction model. Using a sea ice thickness prediction model as a simulation engine, an initial digital twin of sea ice thickness is established.

4. The method for dynamic assessment of marine ice conditions according to claim 1, characterized in that, Step S2 specifically includes: Based on the weather station, hourly temperature change data of the target sea area is obtained and the data is preprocessed. Based on AIS data stream, ship navigation density and trajectory data are obtained, dense areas of ship activity in the target sea area are identified, ship activity density and frequency in the target sea area are calculated, and the intensity of human activities in the target sea area is quantitatively assessed. Based on the sea ice thickness prediction model, the previous sea ice thickness data of the target sea area is used as input, and the temperature change of the target sea area and the intensity of human activities are introduced as key driving factors for sea ice thickness. The current predicted value of sea ice thickness in the target sea area is obtained, and a sea ice thickness prediction model forecaster is constructed. Using real-time observation data of sea ice thickness in the target sea area as input, and coupled with the sea ice thickness prediction model forecaster, the residual sequence of sea ice thickness observation value-predicted value in the target sea area is obtained. Using the Kalman filter algorithm, the current predicted value of sea ice thickness in the target sea area is used as the prediction step, and the residual sequence of observed value-predicted value of sea ice thickness in the target sea area is used as the observation update quantity. The Kalman gain of sea ice thickness in the target sea area is calculated and iterated continuously to dynamically correct the initial digital twin of sea ice thickness.

5. The method for dynamic assessment of marine ice conditions according to claim 4, characterized in that, Step S3 specifically includes: Based on historical sea ice thickness data of the target sea area, spatiotemporal alignment and preprocessing are performed to construct a spatiotemporal dataset of sea ice thickness in the target sea area; Using the spatiotemporal dataset of sea ice thickness in the target sea area as input, a fixed sliding time window is set, and the real value of sea ice thickness in the target sea area in a certain future segment within the sliding time window is used as the prediction target to construct a training sample set for the spatiotemporal self-attention mechanism model. According to the time sequence, the training sample set of the spatiotemporal self-attention mechanism model is divided into training set, validation set and test set; Each grid point in the target sea area is mapped to a high-dimensional vector as the input embedding layer along with the spatiotemporal features of sea ice thickness. Spatial attention encoding is performed, and the spatiotemporal feature association weights of any grid point with other grid points are calculated to capture the spatial long-range association of sea ice thickness in the target sea area. The importance weight of each time step in the historical data of sea ice thickness in the target sea area for the predicted sea ice thickness is calculated. Through multi-layer attention coding, a comprehensive feature representation of the spatiotemporal evolution of sea ice thickness in the target sea area is obtained. Based on the comprehensive feature representation of the spatiotemporal evolution law of sea ice thickness in the target sea area, the decoder is used to generate the predicted probability distribution of sea ice thickness in the target sea area for each grid point at a future time step. The mean and variance of the probability distribution of sea ice thickness prediction for the target sea area at each grid point and future time step are calculated. A negative log-likelihood loss function is introduced to achieve probabilistic prediction of sea ice thickness in the target sea area.

6. The method for dynamic assessment of marine ice conditions according to claim 5, characterized in that, Step S3 also includes: Based on the target sea area, several representative greenhouse gas concentration pathways were set to simulate the long-term impact of different greenhouse gases on sea ice thickness in the target sea area. Based on shipping thermal pollution, black carbon emissions, and resource extraction disturbances, the regional impact of high-intensity human activities on sea ice thickness in the target sea area is simulated. Set a safety threshold for extreme changes in sea ice thickness. If the sea ice thickness is less than 0.5 meters, the sea area is deemed unsuitable for safe navigation. If the sea ice thickness is less than 1.0 meter, the multi-year ice in the sea area is deemed to have basically disappeared. If the sea ice area in the target sea area decreases by more than 100,000 square kilometers for three consecutive days, it is defined as a rapid collapse event; if the sea ice thickness increases by more than 0.3 meters within a week, it is defined as a rapid generation event. Multiple climate scenarios and high-intensity human activities are set as external constraints to drive the adversarial simulation of the initial digital twin of sea ice thickness after dynamic correction. Set the starting date for the simulation, use the sea ice thickness observation data of the target sea area on that date as the initial state of the dynamically corrected initial digital twin of the sea ice thickness, substitute it into the spatiotemporal self-attention mechanism model, and use the predicted probability distribution of the sea ice thickness of the target sea area in the first future time step as the output. Randomly sample the predicted probability distribution of sea ice thickness in the target sea area at the first future time step to obtain the sea ice thickness field of the target sea area determined by the corresponding predicted probability distribution, which is one of the future states to be realized for the sea ice thickness in the target sea area. Using the sea ice thickness field of the target sea area as the new initial state, and inputting the dynamically corrected initial digital twin of the sea ice thickness, the next step is to predict the probability distribution of the sea ice thickness of the target sea area. This process is iterated and looped continuously to advance the prediction.

7. The method for dynamic assessment of marine ice conditions according to claim 6, characterized in that, Step S3 also includes: Based on the output data of the adversarial simulation of the initial digital twin of sea ice thickness after dynamic correction, the number of occurrences of various critical events and abrupt changes in the target sea area is counted, the frequency of critical abrupt changes in the target sea area is calculated, and the risk of extreme changes and critical abrupt changes in sea ice thickness under different external pressures is quantitatively assessed.

8. The method for dynamic assessment of marine ice conditions according to claim 7, characterized in that, Step S4 specifically includes: Based on the dynamically corrected initial digital twin of sea ice thickness and the output data of the adversarial simulation period, an S-shaped structure is constructed. D F G represents the simulated sea ice thickness matrix of the target sea area, S represents the number of adversarial simulations of the initial digital twin of the sea ice thickness after dynamic correction, D represents the adversarial simulation time step, F represents the longitude of the adversarial simulation of the target sea area, and G represents the latitude of the adversarial simulation of the target sea area. The minimum sea ice thickness required for safe navigation is set at 0.5 m. Calculate the value of each grid point in S D F The average sea ice thickness at each grid point is obtained by taking the average sea ice thickness at all adversarial simulation times and time steps in the sea ice thickness simulation matrix for the target sea area G. Statistics and Identification S D F Calculate the sea ice condition probability risk value for all grid points in the sea ice thickness simulation matrix of target sea area G that are less than 0.5 times the safe navigation sea ice thickness. Calculate S for each grid point D F Standard deviation of all adversarial simulations and time steps in the simulation matrix of sea ice thickness in target sea area G; Based on the average value and standard deviation of sea ice thickness in adversarial simulations, the extreme risk index of sea ice conditions is calculated, and a time-varying dynamic risk field of sea ice thickness is generated.

9. The method for dynamic assessment of marine ice conditions according to claim 8, characterized in that, Step S4 also includes: Calculate the travel time from the ship's current point to the target point to obtain the ship's time cost; Calculate the fuel cost of a voyage based on the power consumption and fuel efficiency of the ship's engine. By integrating extreme changes and critical abrupt changes in sea ice thickness under different external pressures, multiple climate scenarios, and high-intensity human activities, the risk cost of the ship is calculated. The ship's time cost, fuel cost, and risk cost are normalized and weighted and integrated to construct a comprehensive ship cost function; With minimizing the ship's comprehensive cost function as the optimization objective, environmental information is updated in real time, and the route planning is continuously adjusted according to the time-varying dynamic risk field of sea ice thickness to dynamically avoid high-risk sea areas, providing ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

10. A dynamic assessment device for ocean ice conditions, characterized in that, include: The device acquires historical data on sea ice thickness in the target sea area, analyzes the spatiotemporal variation trend of sea ice thickness, and establishes an initial digital twin of sea ice thickness. It introduces the temperature change and human activity intensity in the target sea area as key driving factors for sea ice thickness, integrates real-time sea ice thickness observation data of the target sea area, and dynamically corrects the initial digital twin of sea ice thickness. The processing device trains a spatiotemporal self-attention mechanism model based on the dynamically corrected initial digital twin of sea ice thickness to achieve probabilistic prediction of sea ice thickness in the target sea area. It sets multiple climate scenarios and high-intensity human activities as external constraints to drive the adversarial simulation of the dynamically corrected initial digital twin of sea ice thickness and quantitatively assess the risk of extreme changes and critical mutations in sea ice thickness under different external pressures. The device provides a mechanism that dynamically generates a time-varying risk field for sea ice thickness based on a quantitative assessment of the risks of extreme changes and critical abrupt changes in sea ice thickness under different external pressures, providing ships with adaptive optimal route planning with dynamic risk avoidance capabilities.

Images