A method for predicting meteorological factors for ship navigation based on marine meteorological information

Abstract

This invention relates to the fields of marine meteorology and navigation, and discloses a method for predicting meteorological factors for ship navigation based on marine meteorological information. This method is used to predict meteorological factors for ship navigation, assess navigation risks, and plan alternative routes. The method includes processing meteorological data and ship data for a target sea area to obtain historical meteorological time series and ship motion response sequences; constructing state vector trajectories of meteorological factors using historical meteorological data, establishing a dynamic model library, and obtaining the ship motion response function to meteorology based on meteorological and motion response data; generating deterministic meteorological factor prediction sequences by embedding the current meteorological state into the dynamic model; and constructing stochastic differential equations to generate probabilistic prediction distributions of meteorological factors. This invention integrates data-driven and physical constraints, providing more comprehensive navigation decision support.

Description

Technical Field

[0001] This invention relates to the fields of marine meteorology and navigation, and in particular to a method for predicting meteorological factors for ship navigation based on marine meteorological information. Background Technology

[0002] Ship navigation safety and efficiency are highly dependent on accurate understanding and forecasting of the marine meteorological environment. Changes in meteorological factors such as wind, waves, and currents directly affect ship speed, fuel consumption, sailing time, and maneuverability, thus impacting navigation safety and economy. With the rapid development of the global shipping industry and the increasing frequency of extreme weather events, accurately predicting ship navigation meteorological factors based on marine meteorological information to provide a scientific basis for route optimization and navigation decisions has become an important research direction in the field of maritime technology. Currently, the prediction of ship navigation meteorological factors has the following shortcomings in practical applications:

[0003] Marine meteorological systems are inherently complex, nonlinear, and nonstationary dynamic systems. Extreme events such as abrupt typhoon path changes and rapid wave height growth are often accompanied by significant nonlinear dynamic characteristics. Traditional numerical weather prediction models, based on simplified physical equations, struggle to accurately describe these nonlinear processes.

[0004] Weather forecasts are inherently uncertain, and accurately quantifying this uncertainty is crucial for navigation risk assessment. Current mainstream methods primarily rely on ensemble forecasts, which generate probabilistic information through multiple integral numerical models. However, this method is computationally expensive and difficult to widely apply in real-time navigation decision-making. Simple statistical interval estimation, on the other hand, is often oversimplified and fails to accurately describe the stochastic characteristics of weather evolution.

[0005] Current technologies mostly focus on meteorological factor prediction, lacking a complete chain for directly applying prediction results to optimize navigation decisions. When high-risk sea states are predicted, there are currently no mature technical solutions for automatically generating alternative routes or for weighing the risks and benefits of different routes, thus limiting the practical application value of prediction information.

[0006] Therefore, we propose a method for predicting meteorological factors for ship navigation based on marine meteorological information to solve the above problems. Summary of the Invention

[0007] This invention provides a method for predicting meteorological factors for ship navigation based on marine meteorological information, which is used to predict meteorological factors for ship navigation, assess navigation risks, and plan alternative routes.

[0008] The first aspect of this invention provides a method for predicting meteorological factors for ship navigation based on marine meteorological information. The method includes: acquiring and processing meteorological data and ship data from a target sea area to obtain historical meteorological time series and historical ship motion response sequences; constructing state vector trajectories of meteorological factors based on the historical meteorological time series and establishing a dynamic model library; constructing and solving variational functionals based on the historical meteorological time series and the historical ship motion response sequences to obtain a response function; embedding the meteorological state into the dynamic model library to generate a deterministic meteorological factor prediction sequence; constructing a system of stochastic differential equations, determining the parameters of the system of stochastic differential equations, solving the equations, and generating a probability prediction distribution of meteorological factors; and inputting the probability prediction distribution of meteorological factors into the response function to obtain a predicted probability distribution of the ship's navigation state.

[0009] Optionally, in a first implementation of the first aspect of the present invention, historical marine meteorological reanalysis data of the target sea area is obtained, and ship navigation log data of a specified ship type within the target sea area is obtained; spatial interpolation processing is performed on the historical marine meteorological reanalysis data, and temporal interpolation processing is performed on the interpolated historical marine meteorological reanalysis data to obtain a matched meteorological dataset; outlier detection is performed on the matched meteorological dataset and the ship navigation log data, outlier data points are removed, and linear interpolation is performed to fill in the missing data after removal to obtain a meteorological and motion synchronization dataset; the meteorological data sequence and ship motion response data sequence in the meteorological and motion synchronization dataset are processed to obtain a historical meteorological time series and a historical ship motion response sequence.

[0010] Optionally, in a second implementation of the first aspect of the present invention, for the meteorological factors in the historical meteorological time series, the self-mutual information value of the meteorological factors is calculated, and the delay time corresponding to the decrease of the self-mutual information value to a minimum value is determined as the optimal delay time; based on the optimal delay time, the embedding dimension is gradually increased and the proportion of pseudo-neighborhood points under each dimension is calculated, and the minimum embedding dimension corresponding to the decrease of the proportion of pseudo-neighborhood points to a preset threshold is determined as the optimal embedding dimension; based on the optimal delay time and the optimal embedding dimension corresponding to the meteorological factors, the time points in the historical meteorological time series are expanded into state vectors to obtain a set of state vector trajectories; for the state vectors in the set of state vector trajectories, neighboring state vectors are searched, the evolution direction is recorded, and a local neighborhood evolution information set is formed; all state vectors and their corresponding local neighborhood evolution information sets are compiled to construct a dynamic model library.

[0011] Optionally, in a third implementation of the first aspect of the present invention, based on the multi-factor combination in the historical meteorological time series and the ship motion state parameters in the historical ship motion response series, an undetermined parameterized response function expression is constructed; the momentum conservation, energy conservation, and wave breaking critical conditions of the air-sea interface are extracted, and these physical principles are transformed into mathematical constraints on the undetermined parameterized response function expression; a variational functional is constructed, consisting of data fitting terms, physical constraint terms, and smoothness regularization terms; the variational functional is iteratively solved, and the value of the variational functional is minimized by adjusting the undetermined parameters in the undetermined parameterized response function expression, which is the optimal parameter; the optimal parameter is substituted into the undetermined parameterized response function expression to obtain a deterministic ship motion state response function to meteorological factors.

[0012] Optionally, in the fourth implementation of the first aspect of the present invention, meteorological observation data and output data of numerical weather prediction models are acquired to form a multi-factor meteorological state vector; the multi-factor meteorological state vector is expanded according to the delay time and embedding dimension to form an embedded state vector; the embedded state vector is input into the dynamic model library, neighboring state vectors are searched, local neighborhood evolution information sets are extracted, evolution increments are generated, and deterministic meteorological factor prediction sequences are generated; the prediction error distribution characteristics of the deterministic meteorological factor prediction sequences are statistically analyzed to obtain the statistical characteristics of the errors; based on the statistical characteristics of the errors, a set of stochastic differential equations is constructed, the set of stochastic differential equations is numerically solved to obtain probability density functions, and the probability density functions are combined to form a probabilistic prediction distribution of meteorological factors.

[0013] Optionally, in a fifth implementation of the first aspect of the present invention, a stochastic differential equation describing the evolution of the wave height factor is constructed based on the statistical characteristics of the error:

[0014] ;

[0015] in, To standardize meteorological factors at discrete time steps Increment within; Represents the deterministic drift increment; For random perturbations that are normally distributed, For time step The corresponding autocorrelation coefficient.

[0016] Optionally, in the sixth implementation of the first aspect of the present invention, the probability prediction distribution of the meteorological factors is sampled to generate a meteorological factor probability equivalent sample sequence. The meteorological factor probability equivalent sample sequence is input into the ship motion state response function to calculate the ship speed time series, roll amplitude time series, and pitch amplitude time series. Based on the ship speed time series, the expected sailing time is calculated to obtain the sailing time sample value. The ship speed sample value, roll amplitude sample value, and sailing time sample value are sorted, and quantile values ​​are extracted. The quantile values ​​are combined to generate the prediction probability interval bands for ship speed, roll amplitude, and expected sailing time.

[0017] Optionally, in the seventh implementation of the first aspect of the present invention, after the predicted probability distribution of the ship's navigation state, an upper bound value is extracted, and the upper bound value is compared with a ship's safe roll threshold; when the upper bound value exceeds the ship's safe roll threshold, a cumulative probability is calculated, and the cumulative probability is compared with a warning trigger probability threshold; when the cumulative probability exceeds the warning trigger probability threshold, a ship navigation risk warning signal is generated; based on the generated speed prediction probability interval and navigation duration prediction probability interval, combined with the overshoot period corresponding to the ship navigation risk warning signal, an alternative route is generated; the path point of the alternative route is used as the new target sea area location, and the process is repeated to generate a ship navigation state predicted probability distribution; the original route is compared with the ship navigation state predicted probability distribution.

[0018] Optionally, in the eighth implementation of the first aspect of the present invention, the waypoint sequence of the original route is extracted, risk segments are identified, and the time window of the risk segments is determined; marine geographic information data of the risk segments are obtained, and multiple candidate detour paths are generated based on the start and end positions of the risk segments, wherein the start and end points of the multiple candidate detour paths are the same as the start and end points of the risk segments; for each candidate detour path, a detour duration probability interval is estimated based on the path length and the speed prediction probability interval; the detour duration probability interval is compared with the duration of the overrun period to select candidate detour paths as a set of feasible detour paths; for the paths in the set of feasible detour paths, a safety gain coefficient is calculated based on the shortest distance, and a range loss coefficient is calculated; based on the safety gain coefficient and the range loss coefficient, a comprehensive evaluation index is constructed, and the feasible detour path with the highest comprehensive evaluation index is selected as the final alternative route.

[0019] The mechanism of this invention is as follows: the meteorological probability prediction distribution is input into the response function of physical constraints to generate the full probability distribution of the ship's speed, roll amplitude and other navigation states, and the closed-loop iteration of route dynamic optimization and alternative solutions is realized based on the risk warning triggering mechanism.

[0020] Beneficial effects: Extracting the nonlinear evolution trajectory of meteorological systems from historical data can effectively capture the dynamic characteristics of extreme events such as abrupt typhoon path changes and rapid wave height growth, achieving higher prediction accuracy compared to traditional linear statistical methods. By transforming fundamental principles of ocean dynamics such as momentum conservation and energy conservation into constraint terms of variational functionals, the prediction results of multiple factors such as wind, waves, and currents satisfy the inherent physical laws, avoiding the physical contradictions that may occur with purely data-driven methods.

[0021] Constructing a set of stochastic differential equations and solving its probability density evolution equations can efficiently obtain the complete probability distribution of meteorological factor evolution without relying on time-consuming ensemble forecasts, providing a scientific basis for navigation risk assessment.

[0022] By directly estimating the ship's response function to weather factors from actual ship navigation data using variational inversion methods, errors associated with traditional ship model tests or empirical formulas are avoided. This approach enables more accurate estimations of speed loss and roll amplitude for specific ship types. It generates a full-range probability distribution of the ship's navigation status, including probability intervals for speed, time, and roll amplitude. Based on a risk warning trigger mechanism, it achieves dynamic route optimization, allowing crew members to choose the optimal route with a full understanding of the risks, significantly improving navigation safety. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of an embodiment of the method for predicting ship navigation meteorological factors based on marine meteorological information in this invention.

[0024] Figure 2 A schematic diagram illustrating the deterministic wind speed forecast for the next 6 hours;

[0025] Figure 3 A schematic diagram illustrating the probability prediction of ship speed and roll amplitude in the 6th hour ahead (14:00 UTC);

[0026] Figure 4 A schematic diagram illustrating the spatiotemporal matching and preprocessing of historical marine meteorological reanalysis data obtained from public meteorological databases and historical ship navigation data obtained from AIS and ship navigation logs.

[0027] Figure 5 This is a schematic diagram of another embodiment of the method for predicting ship navigation meteorological factors based on marine meteorological information in this invention;

[0028] Figure 6 This is a schematic diagram illustrating the expected output values ​​of the optimal response function under different weather combinations.

[0029] Figure 7 A schematic diagram showing the predicted distribution of significant waves in the target sea area over the next 6 hours (based on measured values).

[0030] Figure 8 A schematic diagram illustrating the probability distribution of predicted vessel navigation status for the next 6 hours and this voyage segment (65 nautical miles);

[0031] Figure 9 This is a schematic diagram comparing the probability distribution of navigation status between the original route and the alternative route. Detailed Implementation

[0032] This invention provides a method for predicting ship navigation weather factors based on marine meteorological information, used to predict ship navigation weather factors, assess navigation risks, and plan alternative routes. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the method for predicting ship navigation meteorological factors based on marine meteorological information in this invention includes:

[0034] 101. Obtain historical marine meteorological reanalysis data and corresponding historical ship navigation log data for the target sea area. Perform spatiotemporal matching and preprocessing on the historical marine meteorological reanalysis data and historical ship navigation log data to obtain standardized historical meteorological time series and standardized historical ship motion response series.

[0035] It is understood that the executing entity of this invention can be a device for predicting meteorological factors for ship navigation based on marine meteorological information, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0036] It should be noted that, taking a certain roll-on / roll-off cargo ship route in the Bohai Strait (Yantai to Dalian) as an example, annual data from June 2023 to May 2024 is collected and matched to complete the data preparation in step 101.

[0037] Historical data for the target sea area (latitude and longitude range: 120.5°E-121.5°E, 37.5°N-38.5°N) was obtained from the publicly available European Centre for Medium-Range Weather Forecasts (ERA5) dataset. The data includes hourly sea surface wind speed (m / s), wind direction (degrees), significant wave height (m), and wave period (seconds), with a temporal resolution of 1 hour. The data is in grid point format with a spatial resolution of 0.25 degrees. Navigation logs for the same time period were extracted from the Automatic Identification System (AIS) data and ship motion sensor records of the typical roll-on / roll-off cargo ship "Bohai Pearl" on this route. The logs include: ship position (longitude, latitude), speed over ground (knots), heading (degrees), and roll angle (degrees) recorded by the ship's motion measurement unit. The timestamps are aligned with the meteorological data, and the data is recorded hourly.

[0038] Meteorological data and ship log data were unified to UTC time, ensuring complete correspondence of timestamps (both hourly). For each hourly ship position, bilinear interpolation was used to extract wind speed, wind direction, wave height, and wave period values ​​from four surrounding ERA5 meteorological grid points. This yielded a meteorological time series that precisely matched the ship's position.

[0039] Exact outliers were checked and removed (data points with a speed of 0 knots (during berthing in port) and extreme outliers with a roll angle greater than 15 degrees). The matched sequences were standardized to eliminate the influence of dimensions. For the meteorological factor of "wind speed," its mean and standard deviation were calculated over the entire time series. Then, the mean was subtracted from each original value, and the result was divided by the standard deviation to obtain a standardized historical meteorological time series with a mean of 0 and a standard deviation of 1. The ship motion response data underwent similar processing to obtain a standardized historical ship motion response series, resulting in two sets of standardized time series of equal length.

[0040] 102. Based on standardized historical meteorological time series, the phase space reconstruction method is used to determine the delay time and embedding dimension of each meteorological factor, construct the state vector trajectory of each meteorological factor in the phase space, and establish a dynamic model library to characterize the local evolution law of meteorological factors based on the proximity relationship between the state vector trajectories.

[0041] It should be noted that, based on the standardized historical meteorological time series obtained above, the "standardized wind speed" factor is used as an example.

[0042] In the obtained standardized historical meteorological time series, the "wind speed" series was selected as the processing object. This series was defined as discrete data points with an hourly time interval and a length of 8760 (corresponding to one year of data), denoted as {v1, v2, ..., v8760}. To reconstruct the one-dimensional time series into a high-dimensional phase space that reflects the system's dynamic characteristics, two key parameters need to be determined.

[0043] Delay time ( The mutual information (MIA) value is typically determined using the autocorrelation function method or the mutual information method. Taking the mutual information method as an example, the MIA value of the sequence at different delays is calculated, and the delay at which the MIA first reaches a local minimum is taken as the MIA value. After calculation, the delay time The time frame was set to 4 hours; the embedding dimension (m) was determined using the spurious nearest neighbor method. As the embedding dimension m increases, the proportion of "spurious" nearest neighbors of the state vector in the high-dimensional space decreases and tends to stabilize. When m increases from 1 to 6, the proportion of spurious nearest neighbors decreases from 90% to about 5%, and stabilizes at below 2% when m ≥ 7. Therefore, the embedding dimension m was determined to be 7.

[0044] Using the defined parameters ( =4, m=7), reconstruct the original one-dimensional time series. Starting from the beginning of the sequence, the state at each time t is composed of m observations arranged in order of time delay, forming an m-dimensional vector. For time t=100, the corresponding 7-dimensional state vector is: According to this rule, a state vector is generated for all constructible moments in the sequence (t=1 to t=8737). The set of these vectors constitutes the trajectory of the wind speed factor in the 7-dimensional phase space.

[0045] For each state vector (V100) on the phase space trajectory, find its k nearest historical state vectors (k=20, i.e., the 20 points with the closest Euclidean distance) throughout the entire trajectory. For the current state V100, analyze the evolution of its k nearest neighbors in the next step (i.e., after 1 hour). These neighbors, Vneighbor, evolve into Vneighbor+1 in the next step. Based on this pairing data of "current state - next state", a simple local linear regression model can be used to characterize the local evolution law starting from V100, i.e., the approximate mapping relationship f from V100 to V101. local .

[0046] For each (or a representative) state vector obtained from dense sampling on the entire phase space trajectory, the above process of "nearest neighbor identification - local model construction" is repeated. All state vectors and their corresponding local evolution laws f... localTogether, they form a "dynamic model library" covering the main regions of phase space. This model library records the most likely short-term evolution of wind speed under different wind speed states (and their historical evolution patterns).

[0047] 103. Based on standardized historical meteorological time series and standardized historical ship motion response series, a variational functional containing physical constraint terms is constructed. By solving the variational functional, the response function of ship motion state to meteorological factors is obtained by inversion.

[0048] It should be noted that this step is based on standardized historical data to establish a physical response function that links meteorological inputs with ship motion outputs.

[0049] The input (meteorological factors) is selected from standardized historical meteorological time series, choosing three key factors: standardized wind speed (W), standardized significant wave height (H), and standardized wave period (T). The output (motion response) is selected from standardized historical ship "ground speed" (S). It contains 8760 spatiotemporally strictly matched input-output data pairs.

[0050] The ship's speed can be expressed as a function of meteorological factors (W, H, T), but this relationship is not arbitrary and needs to incorporate known physical laws as constraints. The initial design of the response function is a polynomial form containing cross terms: , where a0, a1, ... are the coefficients to be determined.

[0051] During navigation, a ship's speed typically decreases (within a safe threshold) as wind speed and wave height increase. Therefore, the partial derivatives of the function S with respect to W and H (within a reasonable range) should be negative. The work done by waves on the ship's hull is proportional to the square of the wave height. Therefore, the energy loss term in the response function caused by the wave factor (H,T) should be proportional to H. 2 There is a positive correlation trend. Since the data has been standardized, the constraint is reflected in the relative magnitude of the coefficients. The influence of wave height H on speed should generally be significantly greater than the influence of wave period T.

[0052] The solution process is transformed into a problem of finding the optimal function under physical constraints. An objective functional J[F] is constructed, consisting of two parts: a data fitting term, which measures the ability of the candidate function F(W,H,T) to predict the speed S at all 8760 data points. pred With actual speed S trueThe sum of squared errors between the two. This part ensures that the function can reproduce historical data. The physical constraint term transforms the above-mentioned monotonicity, energy and other physical laws into a penalty term added to the functional on the function F and its derivative. At multiple discrete (W,H) sample points, if the partial derivative of F with respect to W is positive (meaning that the speed increases with the wind speed, which violates common sense), then a large penalty value is applied.

[0053] The overall functional J[F] is minimized using an optimization algorithm. A specific function F* is sought that can fit the historical data as well as possible while maximally satisfying the pre-defined physical laws. After solving, a specific mathematical function expression with coefficients is obtained. This function is the response function of the ship's motion state (speed) to meteorological factors obtained through inversion.

[0054] In the inversion, the constant term (0.05) at the end of the response function expression is used to compensate for the nonlinear square term ( ) after variable standardization. , The baseline offset caused by the expectation that the overall output of the ship's motion response is not zero is used to ensure that the expected zero-mean standardization property is still met.

[0055] 104. Acquire meteorological observation data and numerical forecast data of the target sea area at the current moment, embed the current meteorological state into the dynamic model library, identify its position in phase space, and call the nearby local evolution laws to generate a deterministic meteorological factor prediction sequence for a future period of time; at the same time, construct a set of stochastic differential equations describing the stochastic evolution process of meteorological factors, determine the parameters of the set of stochastic differential equations based on the error statistical characteristics of the deterministic meteorological factor prediction sequence, solve its probability density evolution equation, and generate the probability prediction distribution of meteorological factors for a future period of time.

[0056] It should be noted that, based on the dynamic model library constructed in step 102, and combined with real-time observation and forecast data, deterministic and probabilistic predictions of future meteorological factors are made.

[0057] The current time is 08:00 UTC on June 1, 2024. After assimilating meteorological observation data and initial field data from the numerical weather prediction model for the target sea area (near the Yantai-Dalian shipping route), the current meteorological conditions are obtained: standardized wind speed is 0.8, and standardized wave height is 0.5. Based on the parameters determined in step 102 (delay time...),... =4 hours, embedding dimension m=7), construct the state vector of the current moment in the 7-dimensional phase space using the latest normalized sequence (containing the current moment and sufficient historical data before it). This vector is V now =[0.8,0.75,0.6,0.5,0.3,0.2,0.1].

[0058] In the phase space trajectory constructed in step 102, find the current state vector V. now The k nearest neighbors (k=20) of historical state vectors. The local evolutionary rules (i.e., the local linear mapping function f) corresponding to these neighbor vectors are then invoked. local ), for V now Perform a one-step evolution (predicting the next hour) to obtain the next predicted state vector, and extract the first component (i.e., the one-step predicted value of the normalized wind speed, 0.72). Use this newly predicted state vector as the new current state, and repeat the "nearest neighbor search-local evolution" steps to iteratively generate a deterministic meteorological factor prediction sequence for the next T hours (T=24 hours). Figure 2 An example of a deterministic wind speed forecast for the next 6 hours is shown:

[0059] Based on a comparison of numerous historical forecasts (backtested using a model library) and corresponding actual conditions, the statistical characteristics of the error in the deterministic forecast sequence are analyzed. It is found that the forecast error (predicted value - actual value) for the 6th hour approximately follows a normal distribution with a mean of 0 and a standard deviation of 0.15. Based on this statistical regularity, a stochastic differential equation describing the random evolution of meteorological factors is constructed. Its drift term is determined by the trend of the deterministic forecast sequence, while the diffusion term (intensity of random fluctuations) is determined by the standard deviation of the error at each forecast step ([0.05, 0.08, 0.10, 0.12, 0.14, 0.15]).

[0060] By solving the probability density evolution equation corresponding to this stochastic differential equation, the complete probability distribution of meteorological factors at each predicted time can be obtained. For the 6th hour in the future (14:00 UTC), the predicted wind speed probability distribution can be expressed as a normal distribution with a mean of 0.80 and a standard deviation of 0.15. This yields the probability prediction distribution of meteorological factors over a future period.

[0061] 105. Input the predicted probability distribution of meteorological factors over a future period into the ship's motion state response function to calculate the predicted probability distribution of the ship's navigation state over a future period. The predicted probability distribution includes the probability intervals of speed, time, and roll amplitude.

[0062] It should be noted that the probability interval of the future ship navigation state is estimated by using the ship response function obtained in step 103 and the probability prediction distribution of meteorological factors generated in step 104.

[0063] Following the conclusion of step 104, it is known that the probability distribution of the standardized wind speed (W) in the 6th hour ahead (i.e., 14:00 UTC on June 1, 2024) follows a normal distribution with a mean of 0.80 and a standard deviation of 0.15. For simplicity, it is assumed that the probability distribution of the standardized significant wave height (H) is also known, with a mean of 1.0 and a standard deviation of 0.20; the standardized wave period (T) is set to a fixed value of 0.2.

[0064] The standardized speed response function obtained from step 103 is used: .

[0065] Since the input meteorological factors (W and H) are random variables, after substituting them into the response function, the output predicted airspeed value S pred It is also a random variable. The Monte Carlo simulation method is used to calculate its distribution: a large number of samples (10,000) are independently drawn from the probability distributions of meteorological factors W and H. Each sample is a specific set of (W,H,T) values, and one set of samples is (W=0.82,H=1.05,T=0.2).

[0066] Substitute each sample into the response function to calculate a corresponding standardized airspeed prediction. Repeat this process 10,000 times to obtain a sample of 10,000 standardized airspeed predictions. Perform statistical analysis on these 10,000 standardized airspeed predictions. Calculate their mean, standard deviation, and determine their quantiles.

[0067] Statistical analysis results of standardized airspeed predictions (expressed as quantiles) are as follows: Figure 3 As shown. Then, using the inverse operation of speed standardization in step 101 (mean 18 knots, standard deviation 4 knots), it is restored to the actual speed. Taking a 50% probability interval as an example, it means that there is a 50% probability that the speed in the next 6 hours will fall within this interval. Figure 3 Example of a probability prediction for ship speed and roll amplitude in the 6th hour ahead (14:00 UTC);

[0068] The prediction process for roll amplitude is similar to that for air speed. It requires inputting the wave height / period output from step 104 into its specific response function with equal probability, and then obtaining the result through Monte Carlo simulation. Airtime prediction, on the other hand, needs to be based on the air speed probability distribution and estimated using a fixed-distance integral.

[0069] The probability interval prediction of the ship's key navigation states (speed, roll) at a specified future time was obtained, providing a quantitative risk assessment basis for navigation decisions.

[0070] In this embodiment of the invention, spatiotemporal matching and standardization of multi-source data ensures a strict correspondence between meteorological and ship motion sequences. Phase space reconstruction technology is employed to capture the nonlinear evolution of meteorological factors, and a dynamic model library is constructed. Physical constraints are introduced to invert the ship motion response function to meteorological factors, guaranteeing the physical interpretability of the model. Combining stochastic differential equations and response functions, probabilistic interval predictions of key navigation indicators such as future speed and roll amplitude are achieved. This method can quantify prediction uncertainties, providing a scientific basis for ship route planning and risk warning, and significantly improving the safety and economy of navigation in complex sea conditions.

[0071] Please see Figure 4 and Figure 5 Another embodiment of the method for predicting ship navigation meteorological factors based on marine meteorological information in this invention includes:

[0072] 201. Obtain historical marine meteorological reanalysis data and corresponding historical ship navigation log data for the target sea area. Perform spatiotemporal matching and preprocessing on the historical marine meteorological reanalysis data and historical ship navigation log data to obtain standardized historical meteorological time series and standardized historical ship motion response series.

[0073] Specifically, historical marine meteorological reanalysis data for the target sea area is obtained from publicly available meteorological databases. This data includes wind field data, wave data, ocean current data, and sea surface temperature data at different spatiotemporal resolutions. Ship log data for a specified vessel type within the target sea area is obtained from the historical database of the Automatic Identification System (AIS). This log data includes timestamps, vessel position (latitude and longitude), speed, heading, roll angle, and pitch angle. Spatial interpolation is performed on the historical marine meteorological reanalysis data, uniformly interpolating it to a grid of vessel position (latitude and longitude) that matches the ship log data. Temporal interpolation is then performed on the interpolated historical marine meteorological reanalysis data, uniformly interpolating it to... At the same timestamps as the ship's navigation log data, a matching meteorological dataset corresponding one-to-one with the ship's motion state in time and space is obtained. Outlier detection is performed on the matching meteorological dataset and the ship's navigation log data to remove abnormal data points caused by sensor failure or data transmission errors. The missing data after removal is filled by linear interpolation based on observations at adjacent times to obtain a complete meteorological and motion synchronization dataset. The meteorological data sequences and ship motion response data sequences in the complete meteorological and motion synchronization dataset are standardized to eliminate the influence of different physical dimensions, resulting in dimensionless standardized historical meteorological time series and standardized historical ship motion response sequences.

[0074] It should be noted that the historical autumn navigation data of a 50,000-ton bulk carrier in a specific area of ​​the North Pacific is used as an example. Historical marine meteorological reanalysis data for the target sea area (30°N to 35°N, 140°E to 145°E) was obtained from a public meteorological database (European Centre for Medium-Range Weather Forecasts, ERA5). This data is gridded, with a spatial resolution of 0.25° x 0.25° and a temporal resolution of 1 hour. The meteorological factors obtained include: 10m wind speed at sea surface, significant wave height, surface current velocity, and sea surface temperature. Simultaneously, data for a specified time period was extracted from the AIS historical database and the ship's logbook. A typical original record extracted is as follows: timestamp "October 5, 2023, 10:15 AM", ship position "30.12°N, 140.35°E", speed 12.5 knots, heading 045 degrees, roll angle 5.2 degrees, pitch angle 2.1 degrees.

[0075] Since the meteorological data is fixed at the intersection of latitude and longitude (30.00°N and 30.25°N), while the ship's actual position lies between the grids, the system first performs spatial interpolation. Using bilinear interpolation, the data from four adjacent meteorological grid points around the ship are mapped to the ship's actual coordinates. The calculated effective wave height after spatial interpolation at these coordinates is 3.42 meters. Temporal interpolation is then performed. The meteorological data only records at the hourly times of 10:00 and 11:00. After setting the spatial interpolation, the effective wave height at 10:00 is 3.42 meters, and at 11:00 is 3.60 meters. To match the ship's position at "10:15," the system performs linear interpolation on the time axis, obtaining a matching effective wave height of 3.46 meters for the ship's position at 10:15. Wind field, ocean current, and other data are processed similarly to form a one-to-one matching meteorological dataset.

[0076] A smoothing scan was performed on the initially matched dataset. In the record for "October 5th, 10:30 AM," the roll sensor, due to occasional vibration interference from severe sea conditions, recorded a sudden change in the ship's roll angle to 85 degrees. Based on the physical characteristics and stability features of a 50,000-ton bulk carrier, this value far exceeds the normal extreme value under non-capsulating conditions. The system identified it as an outlier and removed it directly based on threshold rules. Next, the system extracted normal observations from adjacent times: the roll value at "10:15 AM" (5.2 degrees) and the roll value at "10:45 AM" (5.6 degrees). Linear interpolation was used to calculate a reasonable filler value of 5.4 degrees for 10:30 AM. This resulted in a continuous and uninterrupted dataset.

[0077] To eliminate the bias in calculation weights caused by different physical dimensions, the sequences were dimensionless. A maximum-minimum standardization method was used to map each indicator to a range of zero to one. The maximum historical significant wave height for this sea area was set to 10 meters, and the minimum to 0 meters; thus, the current wave height of 3.46 meters was converted to 0.346. Similarly, the roll angle of 5.2 degrees was linearly scaled in conjunction with the historical maximum roll angle (with a safe limit of 20 degrees) to obtain a standardized value of 0.26. After this step, all physical quantities were converted into pure proportional values, successfully outputting standardized historical meteorological time series and standardized historical ship motion response sequences.

[0078] 202. Based on standardized historical meteorological time series, the phase space reconstruction method is used to determine the delay time and embedding dimension of each meteorological factor, construct the state vector trajectory of each meteorological factor in the phase space, and establish a dynamic model library to characterize the local evolution law of meteorological factors based on the proximity relationship between the state vector trajectories.

[0079] Specifically, for each meteorological factor in the standardized historical meteorological time series, the average mutual information function method is used to calculate the self-mutual information value of the meteorological factor at different delay times, and the delay time corresponding to the first decrease of the self-mutual information value to a minimum value is determined as the optimal delay time of the meteorological factor. Based on the optimal delay time, the pseudo-neighborhood point analysis method is used to gradually increase the embedding dimension and calculate the proportion of pseudo-neighborhood points under each dimension. The minimum embedding dimension corresponding to the decrease of the proportion of pseudo-neighborhood points to below a preset threshold is determined as the optimal embedding dimension of the meteorological factor. Based on the optimal delay time and optimal embedding dimension corresponding to each meteorological factor, the standardized historical meteorological time series is further analyzed. Each time point in the meteorological time series is expanded into a multi-dimensional state vector. The entire time series is traversed to obtain a set of state vector trajectories for each meteorological factor. For each state vector in the set of state vector trajectories, a predetermined number of neighboring state vectors with the smallest Euclidean distance are searched in the phase space. The evolution direction of the neighboring state vectors and their subsequent time steps is recorded to form a local neighborhood evolution information set for that state vector. All state vectors and their corresponding local neighborhood evolution information sets are compiled to construct a dynamic model library covering the evolution characteristics of the entire historical meteorological time series. The dynamic model library is used to describe the local evolution law of meteorological states at any location in the phase space.

[0080] It should be noted that the standardized "significant wave height" time series output in step 201 is extracted. To find the inherent temporal correlation in wave height evolution, the system uses the average mutual information function method for calculation. Testing with different delay times (1 hour, 2 hours, 3 hours, etc.), the system found that: when the delay time is 1 hour and 2 hours, the mutual information value of the wave height data is high (too strong correlation, information redundancy); when the delay time increases to 3 hours, the mutual information value drops to a local minimum for the first time. This indicates that data at 3-hour intervals are both correlated and provide sufficient new information. Therefore, the optimal delay time for the significant wave height in autumn in this sea area is determined to be 3 hours.

[0081] With a delay of 3 hours, the system employs pseudo-neighborhood point analysis to determine the required number of dimensions to fully unfold the wave height evolution pattern. The system sets a safe threshold of 5% for the proportion of pseudo-neighborhood points and tests are conducted by gradually increasing the dimension, starting from one dimension: in one dimension, the proportion of pseudo-neighborhood points reaches as high as 85%; in two dimensions, it drops to 42%; in three dimensions, it is 15%; and when expanded to four dimensions, the proportion drops sharply to 3.5%, below the preset threshold of 5%. This means that in four-dimensional space, the wave height evolution trajectory is fully unfolded, and false intersections of trajectories no longer occur. Therefore, the optimal embedding dimension for determining the effective wave height is 4 dimensions.

[0082] Based on the optimal delay time (3 hours) and optimal embedding dimension (4 dimensions), the system folds the original one-dimensional time series into a multi-dimensional state vector. Taking "10:00 AM on October 5th" as an example, the system extracts the standardized wave height values ​​(0.35, 0.32, 0.28, and 0.25 respectively) from four times: this moment, 3 hours ago (7:00 AM), 6 hours ago (4:00 AM), and 9 hours ago (1:00 AM), and combines them into a four-dimensional state vector representing "10:00 AM on October 5th". The system traverses all autumn historical data within five years, transforming each time point into such a four-dimensional vector, forming a massive set of state vector trajectories.

[0083] For the four-dimensional state vector constructed earlier, "10:00 AM on October 5th," the system calculates the Euclidean distance in the four-dimensional phase space to find its five closest historical "neighbors." Five extremely similar meteorological state vectors from the autumns of 2021 and 2022 were found. Next, the system examines the actual changes of these five historical "neighbors" at the next time step (i.e., one hour later). Assuming that the wave height normalization values ​​of these five neighbors increased by 0.01, 0.02, 0.015, 0.02, and 0.01 respectively within the next hour, the system records these subsequent evolutionary directions and packages them as a "local neighborhood evolution information set" for this state vector.

[0084] The system repeats the above steps for other meteorological factors such as wind field and ocean current, compiling all multidimensional state vectors and their corresponding evolution information sets to generate a dynamic model library.

[0085] 203. Based on standardized historical meteorological time series and standardized historical ship motion response series, a variational functional containing physical constraint terms is constructed. By solving the variational functional, the response function of ship motion state to meteorological factors is obtained by inversion.

[0086] Specifically, based on multi-factor combinations in standardized historical meteorological time series and ship motion state parameters in standardized historical ship motion response series, an undetermined parameterized response function expression describing the mapping relationship between the two is constructed. Based on the fundamental principles of ocean dynamics and ship hydrodynamics, momentum conservation, energy conservation, and wave breaking critical conditions at the air-sea interface are extracted, and these physical principles are transformed into mathematical constraints on the undetermined parameterized response function expression. A variational functional is constructed consisting of a data fitting term, a physical constraint term, and a smoothness regularization term, where the data fitting term is used to measure the responsiveness of the undetermined parameterized response function expression to historical data. The fitting error is determined by physical constraints, which are used to impose mathematical constraints, and smoothness regularization terms are used to control the sensitivity of the response function to changes in meteorological factors. The adjoint method is used to iteratively solve the variational functional. By continuously adjusting the undetermined parameters in the expression of the undetermined parameterized response function, the value of the variational functional is minimized. The corresponding undetermined parameters are then the optimal parameters. Substituting the optimal parameters into the expression of the undetermined parameterized response function, a deterministic response function of the ship's motion state to meteorological factors is obtained. The response function can output the expected values ​​of the corresponding ship speed, roll amplitude, and pitch amplitude under any combination of meteorological factors.

[0087] It should be noted that the standardized historical meteorological sequence (wind speed, significant wave height) and ship motion sequence (speed, roll amplitude) output in step 201 are extracted. To establish the mapping, the system constructs a multi-layered network structure of undetermined response function, with meteorological factors as inputs and the expected ship speed and roll amplitude as outputs. At this point, the function contains a large number of unknown weight parameters.

[0088] To prevent purely data-driven predictions from violating common-sense ocean physics, the system extracts key principles from fluid mechanics and transforms them into mathematical constraints: Energy conservation constraint: the kinetic energy transferred from the waves to the ship can never exceed the total wave energy contained in the water under the current sea state. Wave breaking critical constraint: when the wave steepness (the ratio of wave height to wavelength) reaches a specific threshold, the wave will break. At this point, the excitation force of the wave on the ship's roll no longer increases linearly and infinitely with wave height, but rather the increase in excitation force slows down due to the dissipation of water energy.

[0089] The above elements are assembled into a variational functional, which acts as a comprehensive scoring criterion, consisting of three parts: a data fitting term, which measures the error between the model's predicted roll and speed and historical log data; a physical constraint term, which immediately incurs a large penalty if the model attempts to output extreme values ​​that violate energy conservation or wave breaking laws; and a smoothness regularization term, which ensures that when meteorological factors are slightly disturbed (wind speed increases slightly from 15.0 knots to 15.1 knots), the predicted ship response does not exhibit irrational and drastic changes.

[0090] An iterative solution using the adjoint method is employed. The system first randomly initializes a set of undetermined parameters, resulting in an extremely high total functional error at the initial calculation. After approximately 500 iterations, the system continuously fine-tunes its internal parameters, causing the total functional error to steadily decrease from 18% until it converges and reaches a minimum at 3.2%. This set of parameters is the optimal set. Substituting the optimal parameters, the system formally establishes the deterministic response function for the bulk carrier. Given any weather combination, this function can output the expected ship state value that conforms to physical laws. Figure 6 Examples of expected output values ​​for the optimal response function under different weather combinations.

[0091] 204. Acquire meteorological observation data and numerical forecast data of the target sea area at the current moment, embed the current meteorological state into the dynamic model library, identify its position in phase space, and call the nearby local evolution laws to generate a deterministic meteorological factor prediction sequence for a future period of time; at the same time, construct a set of stochastic differential equations describing the stochastic evolution process of meteorological factors, determine the parameters of the set of stochastic differential equations based on the error statistical characteristics of the deterministic meteorological factor prediction sequence, solve its probability density evolution equation, and generate the probability prediction distribution of meteorological factors for a future period of time.

[0092] Specifically, the system acquires measured meteorological observation data and numerical weather prediction model output data of the target sea area at the current moment, and fuses them to form a multi-factor meteorological state vector at the current moment. This multi-factor meteorological state vector is then expanded according to delay time and embedding dimension to form an embedded state vector in phase space at the current moment. The embedded state vector is input into a dynamic model library, and a predetermined number of neighboring state vectors closest to the embedded state vector at the current moment are searched in phase space. Local neighborhood evolution information sets corresponding to these neighboring state vectors are extracted. Multiple evolution directions in the local neighborhood evolution information set are weighted and synthesized to generate the evolution increment from the current moment to the previous time step. Through iterative recursion, the system generates the future evolution increment for a predetermined time period step by step. A qualitative meteorological factor prediction sequence is generated. Backtracking analysis is performed on multiple cases from historical periods similar to the current meteorological conditions to statistically analyze the prediction error distribution characteristics of the deterministic meteorological factor prediction sequence under different forecast periods, obtaining the mean, variance, and autocorrelation statistics of the error. Based on the statistical characteristics of the error, a set of stochastic differential equations describing the stochastic evolution process of meteorological factors is constructed, where the deterministic drift term is provided by the deterministic meteorological factor prediction sequence, and the stochastic diffusion term is determined by the variance and autocorrelation statistics of the error. The probability density evolution equations corresponding to the stochastic differential equations are numerically solved to obtain the probability density function of meteorological factors at each moment within a preset future time period. The probability density functions are combined in chronological order to form the probability prediction distribution of meteorological factors over a future period.

[0093] It should be noted that the measured significant wave height of the ship's location at the current time (12:00 on October 6th) is 3.0 meters, while the wave height of the same grid point given by numerical weather prediction is 3.2 meters. The system integrates the two data sets, finely anchors the current standardized wave height state to 0.31, calls the optimal delay time (3 hours) and embedding dimension (4 dimensions) determined in step 202, captures the sequence of the past 9 hours, and constructs the embedding state vector of the current time in phase space as follows. .

[0094] This four-dimensional vector is input into a dynamic model library, and the three closest historical similar states with Euclidean distance are retrieved from massive historical data. In the subsequent time step (1 hour), the wave height evolution increments of these three historical states are +0.015, +0.020, and +0.010, respectively. Weighted summation of these three increments yields an evolution increment of +0.015 for the previous hour. Through this iterative process, a deterministic wave height prediction sequence for the next 6 hours is generated (the standardized prediction value at 13:00 is 0.325, which is approximately 3.25 meters after restoration).

[0095] Because marine meteorology is a chaotic system, deterministic predictions alone are insufficient. The system retrospectively analyzes historical cases similar to the current weather pattern in its database, statistically revealing that prediction errors propagate with increasing lead times. Based on the statistical characteristics of these errors, the system constructs a stochastic differential equation describing the evolution of the wave height factor:

[0096]

[0097] In the formula To standardize meteorological factors at discrete time steps Increment within; Represents the deterministic drift increment; It is the discrete time step; It is the scaling factor of the diffusion coefficient; It is the variance factor of the standardized disturbance; For random perturbations that are normally distributed, For time step The corresponding autocorrelation coefficient.

[0098] By numerically solving the evolution equation corresponding to this equation, the system generates the probability density function of meteorological factors at each moment in the future.

[0099] like Figure 7 As shown, with the increase of the forecast period, the interval (difference between upper and lower bounds) of the probability distribution widens significantly due to the cumulative effect of the random diffusion term.

[0100] 205. Input the predicted probability distribution of meteorological factors over a future period into the ship's motion state response function to calculate the predicted probability distribution of the ship's navigation state over a future period. The predicted probability distribution includes the probability intervals of speed, time, and roll amplitude.

[0101] Specifically, the probability prediction distribution of meteorological factors over a future period is sampled using stratified sampling with equal probability to generate a preset number of meteorological factor probability equivalent sample sequences. Each sample sequence represents a possible spatiotemporal evolution path of a meteorological factor. These preset number of meteorological factor probability equivalent sample sequences are then input into the obtained ship motion state response function. The response function calculates the ship speed time series, roll amplitude time series, and pitch amplitude time series corresponding to each sample sequence. Based on the ship speed time series and a preset route distance, the expected sailing time corresponding to each sample sequence is calculated, resulting in a preset number of sailing time sample values. For each preset future time point, the preset number of ship speed sample values, roll amplitude sample values, and sailing time sample values ​​are statistically sorted, and quantile values ​​at a specified confidence level are extracted for each time point. The quantile values ​​at the specified confidence level at each time point are combined in chronological order to generate prediction probability intervals for ship speed, roll amplitude, and expected sailing time over a future period, which serve as the final output of the ship navigation state prediction probability distribution.

[0102] It should be noted that the system obtains the probability distribution of wave height and wind speed for the next 6 hours from the output of step 204. Since directly processing the continuous probability density function involves extremely high computational cost, the system employs an equal-probability stratified sampling technique, extracting 1000 meteorological sample sequences representing different evolutionary possibilities: Sample sequence A (approximately 5% probability, representing an optimistic path with relatively good sea conditions): the significant wave height fluctuates steadily between 3.1 meters and 3.3 meters for the next 6 hours. Sample sequence B (approximately 50% probability, representing the most likely median path): the significant wave height gradually increases from 3.2 meters to 4.1 meters. Sample sequence C (approximately 5% probability, representing a pessimistic path with deteriorating sea conditions): the significant wave height rises sharply to above 4.8 meters.

[0103] These 1000 meteorological sample sequences are input one by one into the ship motion state response function established in step 203. For each sample sequence, the response function outputs the corresponding changes in speed and roll amplitude. Taking the sample sequence C with deteriorating sea state as an example, due to the surge in wave height, the response function calculates that the ship speed corresponding to this sequence will gradually decrease from 12.5 knots to 9.5 knots, while the roll amplitude will increase from 5.0 degrees to 13.5 degrees.

[0104] The remaining distance of the current planned voyage segment is set to 65 nautical miles. The system combines the speed time series corresponding to these 1000 samples, integrating each to calculate the time required to travel these 65 nautical miles, thus obtaining 1000 sample values ​​for the estimated voyage duration. For each future time point (3rd hour, 6th hour) and the overall voyage duration, the system sorts these 1000 sample results from smallest to largest, and extracts the 5th percentile (lower bound) and 95th percentile (upper bound) values ​​to form a prediction probability interval with a 90% confidence level. Figure 8 Predict the probability distribution of the ship's navigation status for the next 6 hours and for this segment (65 nautical miles);

[0105] This step moves beyond simply providing a single, absolute but easily invalidated deterministic value like "expected speed of 11 knots," and instead clearly defines the dynamic boundaries of the future ship's motion.

[0106] 206. Generate alternative routes based on risk warning trigger results, and compare and display the probability distribution of the navigation status of the original route and the alternative route.

[0107] Specifically, after obtaining the predicted probability distribution of the ship's navigation status, the upper bound of the predicted probability interval for roll amplitude is extracted and compared with a preset safe roll threshold. When the upper bound of the predicted probability interval for roll amplitude exceeds the preset safe roll threshold, the corresponding overshoot period is marked within a future time period, and the cumulative probability of roll amplitude exceeding the safe threshold within the overshoot period is calculated. The cumulative probability is compared with a preset warning trigger probability threshold. When the cumulative probability exceeds the warning trigger probability threshold, a ship navigation risk warning signal is generated for the overshoot period. Based on the generated predicted probability intervals for speed and navigation duration, combined with the overshoot period corresponding to the ship navigation risk warning signal, a multi-objective trade-off analysis is performed on the original route to generate at least one alternative route to avoid the overshoot period. Returning to step 204, the waypoint of the alternative route is used as the new target sea area location, and the subsequent steps are executed again to generate the predicted probability distribution of the ship's navigation status corresponding to the alternative route. The predicted probability distributions of the ship's navigation status of the original route and at least one alternative route are compared and displayed. The comparison and display content includes the predicted navigation duration probability interval, the maximum roll amplitude probability interval, and the risk warning period distribution for each route.

[0108] Furthermore, the waypoint sequence of the original route is extracted, and the risk segments on the original route corresponding to the exceeded time period are identified. The start and end points of the risk segments and their corresponding time windows are determined. Marine geographic information data within a preset range around the risk segments are acquired. The marine geographic information data includes the distribution of isobaths, the location of navigational obstructions, the usual route direction, and the lane separation system boundary. Based on the start and end points of the risk segments, combined with the marine geographic information data, a dynamic programming method is used to generate multiple candidate detour routes. The start and end points of the multiple candidate detour routes are the same as the start and end points of the risk segments, and all are located within navigable waters. For each candidate detour route, the route length and the speed are considered. The probability interval is predicted, and the corresponding detour time probability interval is estimated. The detour time probability interval of each candidate detour path is compared with the length of the overshoot period that the path needs to avoid. Candidate detour paths whose upper limit of the detour time probability interval is less than the length of the overshoot period are selected as the set of feasible detour paths. For each path in the set of feasible detour paths, its safety gain coefficient is calculated based on the shortest distance between the path and the center of the risk area, and its range loss coefficient is calculated based on the path deviation between the path and the original route. Based on the safety gain coefficient and the range loss coefficient, a comprehensive evaluation index of the detour path is constructed, and the feasible detour path with the highest comprehensive evaluation index is selected as the final alternative route.

[0109] It should be noted that the safe roll threshold for this 50,000-ton bulk carrier was set at 12.0 degrees. System scanning revealed that the upper limit of the predicted roll amplitude range reached 14.2 degrees during the 4th to 6th hour (marked as the exceeding period). Further calculations showed that the cumulative probability of roll exceeding 12 degrees during this period was 18%, far exceeding the 10% warning trigger probability threshold. Therefore, a red navigation risk warning was immediately generated, and the corresponding risk segment on the original route was locked: starting point A, ending point B.

[0110] Marine geographic information of the area surrounding waypoints A and B was retrieved. The data showed that shallow water isobaths and islands / reefs obstructed navigation in the southern part of the original route, while the northern part was open, deep, navigable water. Based on the starting point A and the ending point B, a dynamic programming method was used to generate two alternative routes within the navigable water: "Alternative Route 1," which detours significantly northward, and "Alternative Route 2," which runs close to the edge of the shallow water in the south.

[0111] Based on the previous speed forecast range, the estimated travel time for these two routes is as follows: Directly traversing the original high-risk segment would take approximately 3 hours. Alternative Route 1: The estimated detour time is 4.5 to 5.2 hours. Its upper limit (5.2 hours) covers and avoids the 3-hour period of extreme sea conditions. Alternative Route 2: The estimated detour time is 3.5 to 4.0 hours. Due to its shorter travel time and proximity to the storm's core area, it cannot completely avoid the extended sea conditions. Therefore, Alternative Route 2 is eliminated, and only Alternative Route 1 is included in the set of feasible detour routes.

[0112] For alternative route 1, a multi-objective trade-off was conducted. Because this route significantly moves northward away from the high-wave area and increases the distance from the original risk area center, its "safety gain coefficient" is extremely high (system score of 90). However, this route increases the total voyage distance by approximately 15 nautical miles, resulting in a "range loss coefficient" at a moderate level (system score of 70). A weighted calculation of both factors yielded the highest comprehensive evaluation index, officially establishing it as the final alternative route.

[0113] Using the waypoints of the alternative route as the new target locations, the previous prediction steps are repeated to generate the navigation state probability distribution corresponding to the new route. This distribution is then compared and displayed side-by-side with the original route, providing an intuitive basis for the captain's final decision. Figure 9 A comparison of the probability distributions of the original route and the alternative route;

[0114] This step completes the entire closed loop from passively receiving weather forecasts to actively issuing risk avoidance decisions, maximizing the safety of ships navigating under complex marine weather conditions.

[0115] In this embodiment of the invention, a risk warning triggering mechanism is constructed, which can dynamically identify periods of excessive roll and calculate the cumulative risk probability. For high-risk voyages, by combining marine geographic information and multi-objective dynamic programming, an optimal alternative route that balances safety gains and range losses is automatically generated, and the probability distribution comparison with the original route is visualized. This complete closed loop not only provides crew members with a scientific basis for avoidance decisions, but also significantly improves the ship's survivability and navigation economy in adverse sea conditions.

[0116] The present invention also provides an apparatus for predicting meteorological factors for ship navigation based on marine meteorological information. The apparatus for predicting meteorological factors for ship navigation based on marine meteorological information includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the method for predicting meteorological factors for ship navigation based on marine meteorological information in the above embodiments.

[0117] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the method for predicting ship navigation meteorological factors based on marine meteorological information.

[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0119] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0120] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention 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. Such 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 the present invention.

Claims

1. A method of predicting a ship navigation meteorological factor from marine meteorological information, characterized by, include: Meteorological data and ship data of the target sea area are acquired and processed to obtain historical meteorological time series and historical ship motion response series; Based on the historical meteorological time series, the state vector trajectory of meteorological factors is constructed, and a dynamic model library is established; Based on the historical meteorological time series and the historical ship motion response series, a variational functional is constructed and solved to obtain the response function, including: Based on the multi-factor combination in the historical meteorological time series and the ship motion state parameters in the historical ship motion response series, an undetermined parameterized response function expression is constructed; the momentum conservation, energy conservation, and wave breaking critical conditions of the air-sea interface are extracted, and these physical principles are transformed into mathematical constraints on the undetermined parameterized response function expression; a variational functional composed of data fitting terms, physical constraint terms, and smoothness regularization terms is constructed; the optimal parameters are obtained by iteratively solving the variational functional, and then substituted to obtain the deterministic ship motion state response function to meteorological factors; The meteorological conditions are embedded into the dynamic model library to generate a deterministic meteorological factor prediction sequence. A system of stochastic differential equations is constructed, the parameters of the system of stochastic differential equations are determined, and the probability prediction distribution of the meteorological factors is solved, including: The prediction error distribution characteristics of the deterministic meteorological factor prediction sequence are statistically analyzed to obtain the statistical characteristics of the error; based on the statistical characteristics of the error, a system of stochastic differential equations is constructed, and the system of stochastic differential equations is numerically solved to obtain the probability density function; the probability density functions are combined to form the probability prediction distribution of meteorological factors. The predicted probability distribution of the meteorological factors is input into the response function to obtain the predicted probability distribution of the ship's navigation status.

2. The method of predicting a marine weather factor for a ship's voyage from marine weather information according to claim 1, characterized in that, include: For the meteorological factors in the historical meteorological time series, calculate the self-mutual information value of the meteorological factors, and determine the delay time corresponding to the self-mutual information value dropping to a minimum value as the optimal delay time; Based on the optimal delay time, the embedding dimension is gradually increased and the proportion of pseudo-neighborhood points under each dimension is calculated. The minimum embedding dimension corresponding to the decrease of the proportion of pseudo-neighborhood points to below a preset threshold is determined as the optimal embedding dimension. Based on the optimal delay time and the optimal embedding dimension corresponding to the meteorological factors, the time points in the historical meteorological time series are expanded into state vectors to obtain a set of state vector trajectories. For the state vectors in the set of state vector trajectories, search for neighboring state vectors, record the evolution direction, and form a local neighborhood evolution information set; All state vectors and their corresponding local neighborhood evolution information sets are compiled to construct a dynamic model library.

3. The method of predicting a marine weather factor for a marine vessel based on marine weather information according to claim 1, wherein, include: The probability prediction distribution of the meteorological factors is sampled to generate a probability equivalent sample sequence of meteorological factors. The probability equivalent sample sequence of meteorological factors is then input into the ship motion state response function to calculate the ship speed time series, roll amplitude time series, and pitch amplitude time series. Based on the ship speed time series, the estimated sailing time is calculated, and sample values ​​of sailing time are obtained; The sample values ​​of ship speed, roll amplitude, and sailing time are sorted, and the quantile values ​​are extracted. The quantile values ​​are combined to generate the predicted probability intervals for ship speed, roll amplitude, and estimated sailing time.

4. The method of predicting a ship navigation meteorological factor from marine meteorological information according to claim 3, characterized in that, Also includes: After predicting the probability distribution of the ship's navigation status, an upper bound value is extracted and compared with the ship's safe roll threshold. When the upper limit value exceeds the ship safety roll threshold, the cumulative probability is calculated and compared with the warning trigger probability threshold. When the cumulative probability exceeds the warning trigger probability threshold, a ship navigation risk warning signal is generated. Based on the generated speed prediction probability interval and sailing duration prediction probability interval, and combined with the overrun period corresponding to the ship navigation risk warning signal, an alternative route is generated. The alternative route path points are used as the new target sea area locations, and the process is repeated to generate a ship navigation status prediction probability distribution. The original route is then compared with the ship navigation status prediction probability distribution.

5. The method of predicting a ship navigation meteorological factor from marine meteorological information according to claim 4, characterized in that, Extract the waypoint sequence of the original route, identify risky segments, and determine the time window for the risky segments; Obtain marine geographic information data of the risky waterway segment, and generate multiple alternative detour routes based on the start and end locations of the risky waterway segment. The start and end points of the multiple alternative detour routes are the same as the start and end points of the risky waterway segment. For each candidate detour route, the detour time probability range is estimated based on the route length and the predicted speed probability range. The probability interval of the detour time is compared with the length of the overrun period to select candidate detour routes as a set of feasible detour routes; For the paths in the set of feasible detour paths, calculate their safety gain coefficient and their range loss coefficient based on the shortest distance; Based on the safety gain coefficient and the range loss coefficient, a comprehensive evaluation index is constructed, and the feasible detour route with the highest comprehensive evaluation index is selected as the alternative route.

Images