Marine wreck target drift trajectory prediction method considering physical constraint

Abstract

The invention discloses a maritime wreck target drift trajectory prediction method considering physical constraints, and belongs to the technical field of maritime search and rescue prediction. The method comprises the following steps: a, collecting related data of a maritime wreck target; b, preprocessing the collected data; c, selecting a long-short term memory neural network model, taking the normalized U component wind speed, V component wind speed, U component flow velocity and V component flow velocity in the preprocessed data as input, taking the normalized warp movement distance and weft movement distance of the wreck target drift as output, and training the long-short term memory neural network model to obtain the wreck target drift speed. Obtaining a trained long and short term memory neural network model; and d, predicting the drift trajectory of the maritime wreck target. According to the method, the complex ocean dynamic process is fully considered, and meanwhile, the long-short-term memory neural network model is effectively fused, so that the time sequence characteristics in the drift trajectory data of the marine wreck target can be effectively captured, and the prediction precision is higher.

Description

Technical Field

[0001] This invention relates to the field of maritime search and rescue prediction technology, and more specifically to a method for predicting the drift trajectory of a maritime target that has been lost due to physical constraints. Background Technology

[0002] Maritime distress incidents occur frequently, such as shipwrecks and people falling overboard. Timely and accurate prediction of the drift trajectory of a distressed target is crucial for search and rescue operations. In recent years, with the development of machine learning technology, data-driven methods have gradually gained attention. For example, Chinese invention patent CN106768837A discloses a method for predicting the drift trajectory of a person falling overboard, which uses the Lagrange particle tracking method for prediction. Chinese invention patent CN115983475A discloses a method for predicting the drift trajectory of a person falling overboard at the sea area scale, achieved through sea area partitioning and a drift trajectory prediction model. Chinese invention patent CN119397268A discloses a method and system for predicting the drift trajectory of a person falling overboard at sea, which simulates the drifting situation after a person falls overboard by training an LSTM model solely on historical trajectories. Based on existing publicly available patents, the factors considered in predicting the drift trajectory of maritime targets are relatively singular, such as relying solely on historical trajectories for LSTM learning prediction. This often results in insufficient prediction accuracy when facing complex marine environments and variable weather conditions. Summary of the Invention

[0003] Based on the above-mentioned technical problems, this invention proposes a method for predicting the drift trajectory of maritime targets that have been lost due to physical constraints.

[0004] The technical solution adopted in this invention is: A method for predicting the drift trajectory of a maritime target that has been lost due to physical constraints includes the following steps: a. Collect relevant data on maritime accidents; b. Preprocess the collected data; c. Select a model and train it; A long short-term memory neural network model was selected. The normalized U-component wind speed, V-component wind speed, U-component flow velocity, and V-component flow velocity in the preprocessed data were used as inputs, and the normalized meridional and latitudinal drift distances of the crashed target were used as outputs. The long short-term memory neural network model was trained to obtain the trained long short-term memory neural network model. d. Predict the drift trajectory of maritime targets that have been involved in accidents; The relevant data of the maritime accident target to be predicted is obtained, preprocessed, and then input into a trained long short-term memory neural network model to predict the drift trajectory of the maritime accident target.

[0005] In step c, a formula for predicting the drift trajectory of a maritime target that has been lost due to physical constraints is established: ; ; in, Let be the wind speed at time t. Let be the flow velocity at time t. Let be the drift velocity of the drifting target at time t. For time intervals, Let be the drift velocity of the drifting target at time t+1. , For learnable parameters, The drag coefficient is determined by the properties of the target itself at sea. The initial position, Let t be the position at time t.

[0006] By substituting the preprocessed existing data into the above formula for predicting the drift trajectory of a maritime target that has suffered a crash, taking into account physical constraints, the drift trajectory can be determined. , .

[0007] Compared with the prior art, the beneficial technical effects of the present invention are as follows: 1. This invention fully considers the physical constraints of complex ocean dynamic processes and effectively integrates a long short-term memory neural network (LSTM), thereby effectively capturing the time series features in the drift trajectory data of maritime targets lost in complex ocean environments. Compared with traditional single physical models and empirical formulas, the prediction accuracy is higher.

[0008] 2. The LSTM model used in this invention has powerful nonlinear modeling capabilities. Combined with the injection of physical constraints (a formula for predicting the drift trajectory of a maritime target that has been lost due to physical constraints), it can handle the drift trajectory prediction problem under complex marine environments and variable weather conditions, making it more adaptable.

[0009] 3. The method of this invention is driven by ocean dynamic field data and drift data, without the need for complex physical modeling and parameter adjustment, and is easy to implement and apply, providing fast and accurate predictive support for maritime search and rescue operations. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the method for predicting the drift trajectory of a maritime target that has suffered a loss of life, taking into account physical constraints, according to the present invention. Figure 2 This is a comparison chart showing the drift trajectories of maritime targets predicted by the method of this invention and the traditional numerical simulation method in a specific application example. Detailed Implementation

[0011] To address the shortcomings of existing methods in predicting the drift trajectory of maritime targets, such as insufficient prediction accuracy and poor adaptability, this invention proposes a method for predicting the drift trajectory of maritime targets that considers physical constraints. This method, by fully considering marine dynamic environmental factors such as wind and currents, establishes a formula for predicting the drift trajectory of maritime targets that considers physical constraints and integrates an LSTM model. This effectively captures the time-series features of drift trajectory data of maritime targets in complex marine environments, improving prediction accuracy and providing stronger support for maritime search and rescue operations.

[0012] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0013] like Figure 1 As shown, a method for predicting the drift trajectory of a maritime target considering physical constraints includes the following steps: a. Collect data; Collect relevant data on maritime accidents, including but not limited to accident location (longitude, latitude), accident time, marine dynamic environment data (ocean current speed, direction, wind speed, wind direction), and drift trajectory data over time (usually selected from observation data from marine experiments).

[0014] b. Preprocess the collected data; The collected data undergoes cleaning and normalization to eliminate noise and outliers, thereby improving data quality. Specific steps include: Data cleaning: Remove missing and outlier values, and fill in missing data (e.g., using interpolation methods).

[0015] The normalization process for time-series drift trajectory data and marine dynamic environment data is as follows: The first step is to process the drift trajectory data, converting the latitude and longitude information into UTM coordinates to form the movement distance data in meters.

[0016] ; ; Where R is the approximate distance corresponding to 1° of longitude, R = 111120.0 meters; To convert to meridional position in UTM coordinates, in meters; To convert to latitudinal position in UTM coordinates, in meters; lat represents the longitude of the drift trajectory, and lat represents the latitude of the drift trajectory. The meridional position at time t+1 Let be the latitudinal position at time t+1. The distance moved in the longitudinal direction. This represents the latitudinal movement distance.

[0017] Then, the movement distance and the corresponding wind speed, wind direction, current speed, and current direction for the time period are normalized. Before normalization, wind speed, wind direction, current speed, and current direction are decomposed into longitude and latitude, with the unit being meters per second.

[0018] ; in, For the U-component flow velocity, Let V be the component velocity. For flow rate, For flow direction; For the U-component wind speed, The V component wind speed, For wind speed, The wind direction.

[0019] The following formula is used to normalize the data for meridional distance movement, zonal distance movement, zonal wind, meridional wind, zonal current, and meridional current: ; ; Where x represents the drift trajectory data or marine dynamic environment data over a time series. For drift trajectory data or marine dynamic environment data at time i, For the normalized data, This is the average of the data. This represents the standard deviation of the data.

[0020] c. Select a model and train it; A long short-term memory neural network model was selected. The normalized U-component wind speed, V-component wind speed, U-component flow velocity, and V-component flow velocity from the preprocessed data were used as inputs, and the normalized meridional and latitudinal drift distances of the crashed target were used as outputs. The long short-term memory neural network model was trained to obtain the trained long short-term memory neural network model.

[0021] Furthermore, a formula for predicting the drift trajectory of a maritime target that has been wrecked, taking into account physical constraints, is established: ; ; in, Let be the wind speed at time t. Let be the flow velocity at time t. Let be the drift velocity of the drifting target at time t. For time intervals, Let be the drift velocity of the drifting target at time t+1. , For learnable parameters, The drag coefficient is determined by the properties of the target itself at sea. The initial position, Let t be the position at time t.

[0022] This paper combines a formula for predicting the drift trajectory of a maritime target that has encountered an accident, taking into account physical constraints, with a long short-term memory (LSTM) neural network model. By training the LSM model with preprocessed existing data, the predicted trajectory is determined. , .

[0023] More specifically: The normalized U-component wind speed, V-component wind speed, U-component current speed, and V-component current speed are used as inputs, and the normalized distance of movement of the crashed target is used as output. The long short-term memory neural network model that combines the formula for predicting the drift trajectory of a crashed target at sea considering physical constraints is trained so that the model can learn the time series features in the data, resulting in the trained long short-term memory neural network model considering physical constraints.

[0024] The underlying principles for establishing the formula for predicting the drift trajectory of a maritime target that takes into account physical constraints are as follows:

[0025] When a drifting target is moving at sea, it is affected by atmospheric drag and seawater drag in the horizontal direction. According to Newton's second law, the dynamic equation of the drifting target can be written as follows: ; in, Represents the force of wind. The force representing the flow, Represents friction. represents the drift velocity, m represents the mass of the drift target, and t represents the drift time.

[0026] Based on the calculation formulas for wind and flow forces, the dynamic equations of the drifting target can be simplified to obtain the simplified equations: ; Where α and β are learnable parameters, and λ is the drag coefficient. For wind speed, For flow rate, The drift speed of the drifting target.

[0027] Discretizing the simplified equations above, we obtain the formula for predicting the drift trajectory of a maritime target considering physical constraints, as follows: ; ; in, Let be the wind speed at time t. Let be the flow velocity at time t. Let be the drift velocity of the drifting target at time t. For time intervals, Let be the drift velocity of the drifting target at time t+1. , For learnable parameters, The drag coefficient is determined by the properties of the target itself at sea. The initial position, Let t be the position at time t.

[0028] This is then incorporated into the Long Short-Term Memory (LSTM) neural network model, and the hidden states of the LSM neural network model are... The last two dimensions are encoded as speed. The remaining dimensions retain the memory function of traditional long short-term memory neural network models.

[0029] Model training: The model is trained using 80% of the samples. The mean squared error (MSE) of the residuals is used as the physical constraint loss. The model training is guided by the physical regularization loss function. Appropriate optimization algorithms (such as the Adam algorithm) are used to adjust the model parameters until the prediction error of the model on the validation set reaches the preset threshold.

[0030] Model evaluation: The remaining 20% ​​of the samples are used as the test set to evaluate the trained model. The error between the predicted trajectory and the actual trajectory is calculated, such as mean squared error (MSE) and mean absolute error (MAE), to verify the model's predictive performance.

[0031] d. Predict the drift trajectory of maritime targets that have been involved in accidents; The relevant data of the maritime accident target to be predicted is input into the long short-term memory neural network model trained in step c to predict the drift trajectory of the maritime accident target.

[0032] When using a long short-term memory neural network model for prediction, wind speed and flow velocity can be directly input. First, conventional methods are used to decompose the wind speed into zonal wind speed, zonal flow velocity, meridional wind speed, and meridional flow velocity. Then, data preprocessing such as normalization is performed. Finally, the data is substituted into the trained long short-term memory neural network model to output the zonal and meridional movement distances of the target to be predicted. Then, conventional methods are used to merge these distances into the total movement distance of the target to be predicted.

[0033] The drift trajectory predicted by the prediction method of this invention provides a reference for the search and rescue of people who have fallen into the water, unpowered vessels, life rafts, etc.

[0034] The present invention provides a method for predicting the drift trajectory of maritime targets that have been lost due to physical constraints. The key features of this method are as follows: (1) Time series feature capture and prediction based on LSTM: This invention employs a Long Short-Term Memory (LSTM) neural network model to process drift trajectory data of maritime targets and integrates physical constraints, thereby effectively capturing the time-series features in data from complex marine environments. Compared to traditional methods (such as physical models and empirical formulas), it can predict drift trajectories more accurately, improving the success rate of search and rescue operations. This invention can process long-sequence data, fully exploring the temporal dependencies within the data, thus providing more accurate prediction results.

[0035] (2) Feature extraction and optimization considering multiple environmental factors: This invention extracts features related to drift trajectories from preprocessed data, such as time-varying characteristics of ocean currents and wind, and constructs feature vectors. By optimizing the feature extraction process, the predictive performance of the model is further improved. Compared with existing technologies, this invention is based on the equations of motion of maritime targets, comprehensively analyzes the forces acting on the target, and fully considers the combined effects of wind, waves, and currents. By establishing a prediction formula for the drift trajectory of maritime targets that have suffered accidents, considering physical constraints, and integrating the LSMT learning method, the prediction accuracy of the model is improved.

[0036] (3) Highly adaptive data-driven prediction methods: The method of this invention is data-driven, requiring no complex physical modeling or parameter adjustments, and can adapt to complex marine environments and variable weather conditions, exhibiting strong adaptability and practicality. This method is not only applicable to various maritime search and rescue scenarios, but also provides rapid response and timely, accurate prediction results in practical applications, offering strong support for search and rescue operations.

[0037] The invention will be further explained below with reference to specific application examples.

[0038] Taking a sea trial in a certain year as an example, a target tracking device was deployed at a certain sea area. A drift prediction model was needed to predict the drift trajectory of the device over the next 48 hours. At the same time, the accuracy of the prediction results was evaluated using the observation data returned by the device.

[0039] Obtain meteorological and marine dynamic environmental field data for this sea area for the next 3 days. Perform data preprocessing as follows:

[0040] First, the latitude and longitude information is converted into UTM coordinates to form the movement distance data in meters.

[0041] ; ; Where R is the approximate distance corresponding to 1° of longitude, R = 111120.0 meters; To convert to meridional position in UTM coordinates, in meters; To convert to latitudinal position in UTM coordinates, in meters; lat represents the longitude of the drift trajectory, and lat represents the latitude of the drift trajectory. The meridional position at time t+1 Let be the latitudinal position at time t+1. The distance moved in the longitudinal direction. This represents the latitudinal movement distance.

[0042] Then, the movement distance and the corresponding wind speed, wind direction, current speed, and current direction are normalized. Before normalization, wind speed, wind direction, current speed, and current direction are decomposed into longitude and latitude, with the unit being meters per second.

[0043] ; in, For the U-component flow velocity, Let V be the component velocity. For flow rate, For flow direction; For the U-component wind speed, The V component wind speed, For wind speed, The wind direction.

[0044] The specific formulas for normalizing zonal wind, meridional wind, zonal current, and meridional current data are as follows: ; ; Where x represents the time-series trajectory data and marine environmental data, This is the normalized data. This is the average of the data. The standard deviation of the data is given. A trained long short-term memory neural network model was used to predict the drift trajectory of this maritime target over the next 48 hours. The results are as follows: Figure 2 As shown.

[0045] from Figure 2 As can be seen, the corrected drift trajectory considering physical constraints in this invention matches the observed trajectory better. According to error statistics, the 24-hour error of the drift trajectory before correction was about 10 kilometers, while the drift error after correction was about 1.5 kilometers, which greatly improves the accuracy of predicting the drift trajectory of maritime targets.

Claims

1. A method for predicting the drift trajectory of a maritime target considering physical constraints, characterized in that... Includes the following steps: a. Collect relevant data on maritime accidents; b. Preprocess the collected data; c. Select a model and train it; A long short-term memory neural network model was selected. The normalized U-component wind speed, V-component wind speed, U-component flow velocity, and V-component flow velocity in the preprocessed data were used as inputs, and the normalized meridional and latitudinal drift distances of the crashed target were used as outputs. The long short-term memory neural network model was trained to obtain the trained long short-term memory neural network model. d. Predict the drift trajectory of maritime targets that have been involved in accidents; The relevant data of the maritime accident target to be predicted is obtained, preprocessed, and then input into a trained long short-term memory neural network model to predict the drift trajectory of the maritime accident target.

2. The method for predicting the drift trajectory of a maritime target considering physical constraints according to claim 1, characterized in that, In step a: Relevant data on maritime accidents include the accident location, accident time, drift trajectory data over time, and marine dynamic environment data.

3. The method for predicting the drift trajectory of a maritime target considering physical constraints according to claim 2, characterized in that, In step b: data preprocessing includes data cleaning and normalization. The normalization process for time-series drift trajectory data and marine dynamic environment data is as follows: First, the latitude and longitude information is converted into UTM coordinates to form the movement distance data in meters; ； ； Where R = 111120.0 meters; To convert to meridional position in UTM coordinates, in meters; To convert to latitudinal position in UTM coordinates, in meters; lat represents the longitude of the drift trajectory, and lat represents the latitude of the drift trajectory; The meridional position at time t+1 Let be the latitudinal position at time t+1. The distance moved in the longitudinal direction. This represents the latitudinal movement distance. Then, the longitude and latitudinal displacement distances and the corresponding marine dynamic environment data for the time period are normalized. Among them, the marine dynamic environment data includes wind speed, wind direction, current speed and current direction. Before normalization, it is decomposed into longitude and latitude, and the unit is unified as meters per second. ； in, For the U-component flow velocity, Let V be the component velocity. For flow rate, For flow direction; For the U-component wind speed, The V component wind speed, For wind speed, Wind direction; Use the following formula to , , , , and Normalize: ； ； Where x represents the drift trajectory data or marine dynamic environment data over a time series. For drift trajectory data or marine dynamic environment data at time i, For the normalized data, This is the average of the data. This represents the standard deviation of the data.

4. The method for predicting the drift trajectory of a maritime target considering physical constraints according to claim 1, characterized in that, In step c, a formula for predicting the drift trajectory of a maritime target that has been lost due to physical constraints is established: ； ； in, Let be the wind speed at time t. Let be the flow velocity at time t. Let be the drift velocity of the drifting target at time t. For time intervals, Let be the drift velocity of the drifting target at time t+1. , For learnable parameters, The drag coefficient is determined by the properties of the target itself at sea. The initial position, Let t be the position at time t.

5. The method for predicting the drift trajectory of a maritime target considering physical constraints according to claim 4, characterized in that: Hiding the state of the Long Short-Term Memory Neural Network Model The last two dimensions are encoded as speed. The remaining dimensions retain the memory function of traditional long short-term memory neural network models.

6. The method for predicting the drift trajectory of a maritime target considering physical constraints according to claim 1, characterized in that, In step d: the maritime accident targets to be predicted include people who have fallen into the water, unpowered vessels, and life rafts.

Images