Navigation trajectory prediction method and equipment based on static characteristics and dynamic characteristics of ship

Abstract

The invention discloses a navigation trajectory prediction method and device based on ship static characteristics and dynamic characteristics, and relates to the technical field of intelligent shipping, and the method comprises the steps: obtaining corresponding ship static data of a current navigation ship in a current first time interval, and corresponding ship dynamic data in a current second time interval; extracting ship dynamic characteristics corresponding to the ship dynamic data; fusing the ship static data and the ship dynamic features to obtain fused features; and inputting the fusion features into a pre-trained trajectory prediction model to obtain a ship navigation trajectory of a next second time interval output by the trajectory prediction model. The method is used for solving the problem that the prediction accuracy is poor when the ship navigation trajectory is predicted in the prior art, and the prediction accuracy of the ship navigation trajectory is improved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent shipping, and in particular to a navigation trajectory prediction method and device based on the fusion of ship static characteristics and dynamic characteristics. Background Art

[0002] With the rapid development of the maritime transportation industry, the global shipping volume has been continuously increasing, resulting in increasingly fierce competition for waterway resources and a continuous rise in the risk of ship collision accidents. In order to effectively reduce the collision risk and improve the accuracy and efficiency of maritime traffic management, ship navigation trajectory prediction has become one of the key research and application directions.

[0003] In the prior art, a trajectory prediction model is created based on the historical trajectory data of ships, and then the ship navigation trajectory is predicted based on the trajectory prediction model. However, this method has a single consideration factor and a low prediction accuracy of the model. Summary of the Invention

[0004] In view of the above problems and technical requirements, the applicant provides a navigation trajectory prediction method and device based on ship static characteristics and dynamic characteristics, so as to solve the problem of poor prediction accuracy in predicting the ship navigation trajectory in the prior art and improve the accuracy of ship navigation trajectory prediction.

[0005] An embodiment of the present application provides a navigation trajectory prediction method based on ship static characteristics and dynamic characteristics, the method comprising:

[0006] Obtaining ship static data corresponding to the currently navigating ship within a current first time interval, and ship dynamic data corresponding to the current second time interval, wherein the first time interval is greater than the second time interval, the first time interval is an integer multiple of the second time interval, the ship static data includes: ship size information, draft, displacement, and bow and stern draft difference, and the ship dynamic data includes: ship position, speed, and heading;

[0007] Extracting ship dynamic characteristics corresponding to the ship dynamic data;

[0008] Fusing the ship static data and the ship dynamic characteristics to obtain a fusion feature, wherein the ship static data and the ship dynamic characteristics have the same dimension;

[0009] Inputting the fusion feature into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval output by the trajectory prediction model, wherein the trajectory prediction model is trained based on fusion feature samples and ship navigation trajectory samples.

[0010] A navigation trajectory prediction method based on ship static features and dynamic features according to an embodiment of the present application. Inputting the fusion features into a pre-trained trajectory prediction model to obtain the ship navigation trajectory for the next second time interval output by the trajectory prediction model, including:

[0011] Input the fusion features into the trajectory prediction model, and obtain the non-linear correlation relationship between the ship static data and the ship dynamic features through the trajectory prediction model; configure different weights for the ship static data and the ship dynamic features based on the correlation relationship; predict and output the ship navigation trajectory for the next second time interval based on the ship static data and the ship dynamic features after weight assignment.

[0012] A navigation trajectory prediction method based on ship static features and dynamic features according to an embodiment of the present application. The training process of the trajectory prediction model includes:

[0013] Obtain training sample data, where the training sample data includes: the fusion sample and the ship navigation trajectory sample. The fusion sample is obtained based on the ship static feature sample and the ship dynamic feature sample. The ship static feature sample is obtained by performing kernel principal component analysis on the ship static data sample, and the ship dynamic feature sample is obtained by performing LSTM dynamic feature extraction on the ship dynamic data sample. The dimensions of the ship static feature sample and the ship dynamic feature sample are the same;

[0014] Input the training sample data into the trajectory prediction model, and automatically learn the non-linear correlation relationship between the ship static feature sample and the ship dynamic feature sample through the trajectory prediction model; configure different weights for the ship static feature sample and the ship dynamic feature sample based on the correlation relationship; predict the ship navigation trajectory based on the ship static feature sample and the ship dynamic feature sample after weight assignment; compare the consistency between the predicted ship navigation trajectory and the ship navigation trajectory sample; adjust the model parameters of the trajectory prediction model based on the consistency until the number of iterations reaches the preset number, and determine that the training of the trajectory prediction model is completed.

[0015] A navigation trajectory prediction method based on ship static features and dynamic features according to an embodiment of the present application. The ship static data sample includes: ship static data of different ship types and different sizes;

[0016] Performing kernel principal component analysis on the ship static data sample to obtain the ship static feature sample, including:

[0017] Calculate the similarity between the ship static data samples of different ships using the cosine similarity kernel function to obtain a kernel matrix;

[0018] Perform eigenvalue decomposition on the nuclear matrix, and extract the eigenvalues with contribution rates greater than the preset contribution rate;

[0019] Construct a matrix from the ship static data samples corresponding to the extracted eigenvalues to obtain the ship static feature samples.

[0020] According to a navigation trajectory prediction method based on ship static features and dynamic features according to an embodiment of the present application, performing LSTM dynamic feature extraction on ship dynamic data samples to obtain ship dynamic feature samples, including:

[0021] Input the ship dynamic data samples into a pre-trained LSTM model to obtain the ship dynamic feature samples output by the LSTM model, where the LSTM model is trained based on ship dynamic data initial samples and ship dynamic feature initial samples.

[0022] According to a navigation trajectory prediction method based on ship static features and dynamic features according to an embodiment of the present application, the training process of the LSTM model includes:

[0023] Input the ship dynamic data initial samples and the ship dynamic feature initial samples into the LSTM model to obtain ship dynamic feature prediction samples output by the LSTM model; calculate the mean square error between the ship dynamic feature prediction samples and the ship dynamic feature initial samples; optimize the model parameters of the LSTM model based on the mean square error and a preset learning decay rate strategy, and stop training the LSTM model using a preset training stop requirement based on the early stopping method.

[0024] According to a navigation trajectory prediction method based on ship static features and dynamic features according to an embodiment of the present application, after obtaining the training sample data, it further includes:

[0025] Process the training sample data based on any one or more of a data deduplication strategy, an outlier removal strategy, a missing value filling strategy, a data resampling strategy, and a data normalization processing strategy.

[0026] An embodiment of the present application further provides a navigation trajectory prediction device based on ship static features and dynamic features, including:

[0027] An acquisition module, configured to acquire ship static data corresponding to a current navigating ship within a current first time interval, and ship dynamic data corresponding to a current second time interval, where the first time interval is greater than the second time interval, the first time interval is an integer multiple of the second time interval, the ship static data includes: ship size information, draft, displacement, and bow and stern draft difference, and the ship dynamic data includes: ship position, speed, and course;

[0028] An extraction module for extracting the ship dynamic features corresponding to the ship dynamic data;

[0029] A fusion module for fusing the ship static data and the ship dynamic features to obtain fused features, wherein the ship static data and the ship dynamic features have the same dimension;

[0030] A prediction module for inputting the fused features into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval output by the trajectory prediction model, wherein the trajectory prediction model is trained based on fused feature samples and ship navigation trajectory samples.

[0031] An embodiment of the present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the navigation trajectory prediction method based on ship static features and dynamic features described in any one of the above are implemented.

[0032] An embodiment of the present application further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the navigation trajectory prediction method based on ship static features and dynamic features described in any one of the above are implemented.

[0033] The navigation trajectory prediction method and device based on ship static features and dynamic features provided by the embodiments of the present application, by obtaining the ship static data corresponding to the current navigation ship in the current first time interval and the ship dynamic data corresponding to the current second time interval, the first time interval is greater than the second time interval. The present application considers the actual situation of ship navigation, monitors ship static data with a long period and monitors ship dynamic data with a short period, providing an effective data basis for subsequent ship navigation trajectory prediction; furthermore, extracting the ship dynamic features corresponding to the ship dynamic data; fusing the ship static data and the ship dynamic features to obtain fused features; inputting the fused features into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval output by the trajectory prediction model. The present application enables the trajectory training model to simultaneously capture the synergistic effect of the ship's physical attributes and dynamic behaviors through the fusion of static features and dynamic features, which can effectively improve the prediction accuracy of the model and solve the problem of poor accuracy in predicting ship navigation trajectories in the prior art. Description of the Drawings

[0034] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0035] Figure 1 It is a schematic flowchart of a navigation trajectory prediction method based on ship static features and dynamic features provided by an embodiment of the present application;

[0036] Figure 2 It is a schematic structural diagram of a navigation trajectory prediction device based on ship static features and dynamic features provided by an embodiment of the present application;

[0037] Figure 3 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Specific embodiments

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0039] The embodiments of the present application provide a navigation trajectory prediction method based on ship static features and dynamic features. This method can be applied in intelligent terminals, can also be applied in servers, and can also be applied in the controllers of ships. The embodiments of the present application take this method applied in the controller of a ship as an example for illustration. Of course, it is only for illustration and is not used to limit the protection scope of the present application. Other examples are the same, and will not be described one by one later. The specific implementation of this method is as Figure 1 shown:

[0040] Step 101, obtain the ship static data corresponding to the currently navigating ship within the current first time interval, and the ship dynamic data corresponding to the current second time interval.

[0041] Among them, the first time interval is greater than the second time interval, the first time interval is an integer multiple of the second time interval, the ship static data includes: ship size information, draft, displacement, and bow and stern draft difference, and the ship dynamic data includes: ship position, speed, and heading.

[0042] Step 102, extract the ship dynamic features corresponding to the ship dynamic data.

[0043] Step 103: Integrate the static data of the ship and the dynamic characteristics of the ship to obtain integrated characteristics.

[0044] Among them, the dimensions of the static data of the ship and the dynamic characteristics of the ship are the same.

[0045] Specifically, splice the static data of the ship and the dynamic data of the ship to obtain integrated characteristics.

[0046] Step 104: Input the integrated characteristics into a pre-trained trajectory prediction model to obtain the ship's navigation trajectory for the next second time interval output by the trajectory prediction model.

[0047] Among them, the trajectory prediction model is trained based on integrated characteristic samples and ship navigation trajectory samples.

[0048] Among them, in the application stage, the static data of the ship is one-dimensional data, and the static data of the ship within the current first time interval is a constant; the dynamic data of the ship is a three-dimensional time series, which is the dynamic data of the ship obtained within the second time interval. The second time interval includes multiple time points, that is, the dynamic data of the ship corresponding to each time point.

[0049] Among them, the ship's navigation trajectory for the next time interval is the ship's position, speed, and heading corresponding to each time point among the multiple time points in the next time interval.

[0050] Among them, the setting of the time interval can be set by the user according to their actual needs, and this application does not make any restrictions.

[0051] Specifically, during the actual operation of the ship, the main dimension parameters may change (for example, the draft changes due to the change in the loading capacity of the cargo ship). Therefore, a dynamic monitoring mechanism is also implemented for the static data of the ship, but the monitoring period is longer than that of the dynamic data of the ship.

[0052] Among them, the dynamic monitoring of the static data of the ship mainly performs periodic monitoring and adjustment of the draft, displacement, and bow and stern draft difference.

[0053] When adjusting the static data of the ship, data can be updated when there is a change in the current first time interval compared to the previous first time interval. It is also possible to calculate the change difference of specific parameters when there is a change in the current first time interval compared to the previous first time interval, and compare the change difference with the corresponding threshold. Only when the change difference is greater than the corresponding threshold will the data be updated.

[0054] The navigation trajectory prediction method based on the static and dynamic characteristics of ships provided by the embodiments of the present application obtains the ship static data corresponding to the current ship during the current first time interval and the ship dynamic data corresponding to the current second time interval. The first time interval is greater than the second time interval. The present application considers the actual situation of ship navigation, monitors ship static data with a long period and monitors ship dynamic data with a short period, providing an effective data basis for subsequent ship navigation trajectory prediction. Furthermore, the ship dynamic characteristics corresponding to the ship dynamic data are extracted; the ship static data and the ship dynamic characteristics are fused to obtain fused characteristics; the fused characteristics are input into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval output by the trajectory prediction model. By fusing static and dynamic characteristics, the present application enables the trajectory training model to capture the synergistic effect of the physical attributes and dynamic behaviors of ships simultaneously, effectively improving the prediction accuracy of the model and solving the problem of poor accuracy in predicting ship navigation trajectories in the prior art.

[0055] In a specific embodiment, the specific implementation of inputting the fused characteristics into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval includes:

[0056] Input the fused characteristics into the trajectory prediction model, and obtain the non-linear correlation relationship between the ship static data and the ship dynamic characteristics through the trajectory prediction model; configure different weights for the ship static data and the ship dynamic characteristics based on the correlation relationship; predict and output the ship navigation trajectory of the next second time interval based on the ship static data and the ship dynamic characteristics after the weights are assigned.

[0057] Among them, the non-linear correlation relationship, for example, the combined action law of draft and speed; the combined action law of ship length and course (when a ship with a larger ship length turns, due to inertia, the response of the course change is slower); the combined action law of the difference in fore and aft draft and course (too large a difference in fore and aft draft will cause the hull to tilt, reduce the rudder effect, and thus affect the rate of change of the course), etc.

[0058] Specifically, the trajectory prediction model performs high-order abstraction on the characteristics through non-linear mapping to obtain the ship navigation trajectory of the next second time interval. Specifically, the trajectory prediction model maps the original characteristics (fused characteristics) to a high-dimensional space through complex transformations (such as activation functions like ReLU, Tanh, etc.) to capture non-linear relationships, generate new and more meaningful characteristics, and extract higher-level laws from the new characteristics, and finally predict the future trajectory.

[0059] In a specific embodiment, the training process of the trajectory prediction model includes:

[0060] Obtain training sample data; input the training sample data into the trajectory prediction model, and automatically learn the non-linear correlation relationship between the ship static feature samples and the ship dynamic feature samples through the trajectory prediction model; configure different weights for the ship static feature samples and the ship dynamic feature samples based on the correlation relationship; predict the ship navigation trajectory based on the ship static feature samples and the ship dynamic feature samples after weight assignment; compare the consistency between the predicted ship navigation trajectory and the ship navigation trajectory samples; adjust the model parameters of the trajectory prediction model based on the consistency, and return to execute the step of automatically learning the non-linear correlation relationship between the ship static feature samples and the ship dynamic feature samples through the trajectory prediction model, and execute iteratively until the number of iterations reaches the preset number, and determine that the training of the trajectory prediction model is completed.

[0061] Among them, the training sample data includes: fusion samples and ship navigation trajectory samples.

[0062] The fusion samples are obtained based on the ship static feature samples and the ship dynamic feature samples.

[0063] The ship static feature samples are obtained by performing kernel principal component analysis on the ship static data samples.

[0064] The ship dynamic feature samples are obtained by performing LSTM dynamic feature extraction on the ship dynamic data samples.

[0065] The dimensions of the ship static feature samples and the ship dynamic feature samples are the same.

[0066] Among them, the data format of the ship static feature samples is a two-dimensional array. For example, A = [n_samples, d_static]. The specific parameters of the ship static features include: ship length, ship width, draft, displacement, and bow and stern draft difference, etc.

[0067] Among them, the data format of the ship dynamic feature samples is a three-dimensional array. For example, B = [n_samples, T, d_dynamic]. The specific parameters of the ship dynamic feature samples include: position, speed, and heading, etc.

[0068] Among them, the position represents the longitude and latitude coordinates of the ship at a given time, reflecting the navigation position of the ship; the speed represents the navigation speed of the ship; the heading represents the heading angle of the ship, characterizing the heading direction of the ship.

[0069] Specifically, the fusion samples are obtained by splicing the ship static feature samples and the ship dynamic feature samples.

[0070] Among them, it is necessary to ensure that the ship static feature samples and the ship dynamic feature samples are completely aligned in the sample dimension before splicing, that is, the static features of the same ship sample and the motion trajectory data of its dynamic features correspond one by one.

[0071] Furthermore, a fully connected network is used to perform non-linear transformation on the concatenated fusion features, and its network structure is: input layer - hidden layer and output layer. The hidden layer model automatically learns the non-linear correlation between the static feature samples of the ship and the dynamic feature samples of the ship. For example, when a ship with a high draft is turning at high speed, the dynamic features and static features jointly affect the weight distribution of the fully connected layer, so as to more accurately predict the turning radius.

[0072] Among them, the structure of the fully connected layer includes multiple hidden layers, and each layer performs non-linear mapping through the ReLU activation function, and finally obtains the predicted value of the future state of the ship.

[0073] In a specific embodiment, after obtaining the training sample data, the training sample data is processed based on any one or more of the data deduplication strategy, outlier removal strategy, missing value filling strategy, data resampling strategy, and data normalization processing strategy.

[0074] Specifically, the training sample data is based on the data identified by the Automatic Identification System (AIS) of global ships. In order to ensure its quality and consistency, alignment preprocessing operations are performed.

[0075] The data deduplication strategy includes: checking whether there are duplicate records in the training sample data. Duplicate records are usually due to sensor failures or errors that occur during data transmission, so removal operations are required.

[0076] Specifically, deduplication is based on the unique identifier of the ship (for example, MMSI number) and the timestamp (for example, sampling time). For duplicate records with the same ship representation and timestamp, only keep the latest record.

[0077] The outlier removal strategy includes: using the 3-sigma rule to detect and remove outliers.

[0078] Specifically, calculate the mean and standard deviation of each feature (such as position, speed, and heading, etc.); for each data point, if its value deviates from the mean by more than 3 times the standard deviation, it is determined as an outlier. After detecting the outlier, directly delete the corresponding time step data.

[0079] The missing value filling strategy includes: to ensure the continuity and integrity of the data, the cubic spline interpolation method is used to fill the missing trajectory data. Cubic spline interpolation can smoothly fill the missing values and avoid overfitting or oversmoothing.

[0080] The interpolation operation is specifically: perform cubic spline interpolation on the key features (such as position, speed, and heading, etc.) of each time series to ensure the temporal continuity and accuracy of the data. For the interpolated data, ensure that the interpolation result conforms to the physical characteristics of ship motion (for example, the change in ship position should not be too drastic).

[0081] The data resampling strategy includes: to ensure the consistency of the sampling interval of the data, especially when the sampling interval is irregular, resampling technology is adopted to unify the sampling frequency.

[0082] Resampling frequency: According to the requirements of the ship motion trajectory data and characteristics and the application scenario, a unified sampling frequency (for example, one sample every half minute or every minute) is set for data resampling.

[0083] Difference method: Linear interpolation is used to fill in the time points with irregular intervals to ensure that the intervals of all time points are consistent.

[0084] Data alignment: Through the time alignment method, it is determined that all data are aligned on the same time axis, making the subsequent model training and prediction more stable and accurate.

[0085] The data normalization processing strategy includes: The resampled data still has different dimensions and ranges. For example, for dynamic features (such as position, speed, and heading, etc.), the dimensional differences of these features may affect the subsequent model training. Therefore, the dynamic feature data is normalized to ensure the dimensional consistency of the features.

[0086] Specifically, the minimum-maximum normalization is adopted to scale the features to the range of [0,1], as shown in formula (1):

[0087]

[0088] Among them, z' represents the normalized dynamic feature data, z represents the dynamic feature data before normalization, z min represents the minimum value of this dynamic feature data, z max represents the maximum value of this dynamic feature data.

[0089] The normalized data will help improve the training efficiency of the model, enhance the training stability of the model, and avoid the dominant influence of a certain feature on the model training process.

[0090] Specifically, the data format of the training sample data after the above preprocessing has the following characteristics:

[0091] Ship static feature samples: There are no duplicate values, missing values have been supplemented, outliers have been excluded, and the data format is a two-dimensional array. For example, A = [n_samples, d_static], where n_samples represents the number of samples and d_static represents the dimension of the ship static features.

[0092] Marine dynamic feature samples: The data intervals are consistent, missing values have been supplemented, outliers have been excluded, the sampling frequencies are unified, normalization processing has been performed, and the data format is a three-dimensional array. For example, B = [n_samples, T, d_dynamic], where n_samples represents the number of samples, T represents the time step (multiple time steps constitute the second preset time interval), and d_dynamic represents the dimension of the marine dynamic features at each time step. The length of a time series corresponds to multiple time steps.

[0093] In a specific embodiment, the marine static data samples include marine static data of different ship types and different sizes.

[0094] The specific implementation of obtaining the marine static feature samples by performing kernel principal component analysis on the marine static data samples includes:

[0095] Calculating the similarity between the marine static data samples of different ships using the cosine similarity kernel function to obtain a kernel matrix; performing eigenvalue decomposition on the kernel matrix, and extracting the eigenvalues with a contribution rate greater than the preset contribution rate; constructing a matrix with the marine static data samples corresponding to the extracted eigenvalues to obtain the marine static feature samples.

[0096] Specifically, performing an inverse calculation on the matrix of the extracted eigenvalues to obtain the marine static feature samples.

[0097] Among them, the cosine similarity kernel function is shown in formula (2):

[0098]

[0099] Among them, x represents the first eigenvector corresponding to the marine static data sample of the first ship, y represents the second eigenvector corresponding to the marine static data sample of the second ship, ‖x‖ represents the Euclidean norm of the first eigenvector, ‖y‖ represents the Euclidean norm of the second eigenvector, and K(x, y) represents the similarity between the first eigenvector and the second eigenvector.

[0100] Furthermore, calculating the similarity between any two marine static data samples of any two ships in the training sample data to obtain the kernel matrix K, which can be seen in formula (3):

[0101]

[0102] Among them, K(x i , x j ) represents the similarity between any two marine static data samples of any two ships, and n represents the number of samples of the marine static data sample of any one ship.

[0103] Finally, extracting eigenvalues using the contribution rate calculation formula, as shown in formula (4):

[0104]

[0105] Among them, a represents the contribution rate, and λ represents the eigenvalue.

[0106] Among them, the contribution rate is the cumulative contribution rate of the eigenvalues, which is obtained by dividing the sum of the eigenvalues of k samples selected from n sample quantities by the sum of the eigenvalues of n samples.

[0107] In a specific embodiment, the specific implementation of extracting the dynamic features of the ship from the ship dynamic data samples by LSTM includes:

[0108] Input the ship dynamic data samples into a pre-trained LSTM model to obtain the ship dynamic feature samples output by the LSTM model.

[0109] Among them, the LSTM model is trained based on the initial samples of ship dynamic data and the initial samples of ship dynamic features.

[0110] In a specific embodiment, the training process of the LSTM model includes:

[0111] Input the initial samples of ship dynamic data and the initial samples of ship dynamic features into the LSTM model to obtain the predicted samples of ship dynamic features output by the LSTM model; calculate the mean square error between the predicted samples of ship dynamic features and the initial samples of ship dynamic features; optimize the model parameters of the LSTM model based on the mean square error and a preset learning decay rate strategy, and stop training the LSTM model using a preset training stop requirement based on the early stopping method.

[0112] Specifically, the network architecture of the LSTM model is as follows: Adopt a 2-layer stacked LSTM structure, with each layer containing 150 hidden units, and the activation function is the Tanh function. A Dropout layer (dropout rate set to 0.2) is introduced between layers to prevent overfitting. The last layer of LSTM outputs the hidden states of all time steps, and only the hidden state of the final time step is retained as the dynamic feature representation.

[0113] For example, the second time interval is 10 minutes, including 60 time steps, and the sampling frequency is 0.1 Hz.

[0114] Specifically, the training strategy specifically includes:

[0115] Optimizer: Use the Adam optimizer, and the initial learning rate (negatively correlated with the time step, for example, the value is And adopt a learning rate decay strategy (for example, decay to 0.5 times the original value every 10 epochs) for processing.

[0116] Loss function: Use the mean square error as the loss function.

[0117] Regularization: Except for the Dropout layer, L2 weight regularization (with a coefficient of 1×10 -5 ) is introduced in the LSTM layer to further suppress overfitting.

[0118] Batch and epoch: The batch size is set to 512, the training epoch is 200 epochs, and early stopping is adopted (if the change in the training set loss for 5 consecutive epochs is less than the relative change threshold, for example, δ

[0119] = log(d_dynamic / 4)×0.01, where δ represents the change in epochs, then training is terminated early).

[0120] This application effectively solves the problem of the decline in the prediction accuracy of the model due to the change of static features during the navigation of the ship through the static feature dynamic monitoring mechanism. Moreover, the static features are obtained by using kernel principal component analysis, enabling the static features to effectively participate in the prediction of the ship's navigation trajectory and effectively improving the prediction accuracy. Finally, this application adopts the fusion of static features and dynamic features, and the model simultaneously captures the physical data and dynamic behavior of the ship, making the prediction precision reach the optimal.

[0121] The embodiment of this application also provides a navigation trajectory prediction device based on the static and dynamic features of the ship. The specific implementation of this device can refer to the description in the navigation trajectory prediction method based on the static and dynamic features of the ship, and the repeated parts will not be elaborated. As Figure 2 shown, this device includes:

[0122] An acquisition module 201, configured to acquire the ship's static data corresponding to the current navigation ship within the current first time interval and the ship's dynamic data corresponding to the current second time interval, where the first time interval is greater than the second time interval, the first time interval is an integer multiple of the second time interval, the ship's static data includes: ship size information, draft, displacement, and bow and stern draft difference, and the ship's dynamic data includes: ship position, speed, and heading;

[0123] An extraction module 202, configured to extract the ship's dynamic features corresponding to the ship's dynamic data;

[0124] A fusion module 203, configured to fuse the ship's static data and the ship's dynamic features to obtain fusion features, where the dimensions of the ship's static data and the ship's dynamic features are the same;

[0125] A prediction module 204, configured to input the fusion features into a pre-trained trajectory prediction model to obtain the ship's navigation trajectory in the next second time interval output by the trajectory prediction model, where the trajectory prediction model is trained based on the fusion feature samples and the ship's navigation trajectory samples.

[0126] In a specific embodiment, the prediction module 204 includes: inputting the fused features into a trajectory prediction model to obtain the non-linear correlation relationship between the ship static data and the ship dynamic features through the trajectory prediction model; configuring different weights for the ship static data and the ship dynamic features based on the correlation relationship; and predicting and outputting the ship navigation trajectory for the next second time interval based on the ship static data and the ship dynamic features after the weights are assigned.

[0127] In a specific embodiment, the device further includes a training module for obtaining training sample data, where the training sample data includes: fused samples and ship navigation trajectory samples. The fused samples are obtained based on ship static feature samples and ship dynamic feature samples. The ship static feature samples are obtained by performing kernel principal component analysis on ship static data samples, and the ship dynamic feature samples are obtained by performing LSTM dynamic feature extraction on ship dynamic data samples. The dimensions of the ship static feature samples and the ship dynamic feature samples are the same.

[0128] Input the training sample data into the trajectory prediction model to automatically learn the non-linear correlation relationship between the ship static feature samples and the ship dynamic feature samples through the trajectory prediction model; configure different weights for the ship static feature samples and the ship dynamic feature samples based on the correlation relationship; predict the ship navigation trajectory based on the ship static feature samples and the ship dynamic feature samples after the weights are assigned; compare the consistency between the predicted ship navigation trajectory and the ship navigation trajectory samples; and adjust the model parameters of the trajectory prediction model based on the consistency until the number of iterations reaches a preset number, and determine that the training of the trajectory prediction model is completed.

[0129] In a specific embodiment, the ship static data samples include: ship static data of different ship types and different sizes.

[0130] The training module is used to calculate the similarity between the ship static data samples of different ships by using a cosine similarity kernel function to obtain a kernel matrix; perform eigenvalue decomposition on the kernel matrix, and extract the eigenvalues with a contribution rate greater than a preset contribution rate; and construct a matrix with the ship static data samples corresponding to the extracted eigenvalues to obtain ship static feature samples.

[0131] In a specific embodiment, the training module is used to input the ship dynamic data samples into a pre-trained LSTM model to obtain the ship dynamic feature samples output by the LSTM model, where the LSTM model is trained based on ship dynamic data initial samples and ship dynamic feature initial samples.

[0132] In a specific embodiment, the training module is configured to input the initial sample of ship dynamic data and the initial sample of ship dynamic features into the LSTM model to obtain the predicted sample of ship dynamic features output by the LSTM model; calculate the mean square error between the predicted sample of ship dynamic features and the initial sample of ship dynamic features; optimize the model parameters of the LSTM model based on the mean square error and a preset learning decay rate strategy, and stop training the LSTM model using a preset training stop requirement based on the early stopping method.

[0133] In a specific embodiment, the training module is further configured to process the training sample data based on any one or more of a data deduplication strategy, an outlier removal strategy, a missing value supplementation strategy, a data resampling strategy, and a data normalization processing strategy.

[0134] Figure 3 An entity structure diagram of an electronic device is illustrated, as Figure 3 shown. The electronic device may include: a processor 301, a communication interface 302, a memory 303, and a communication bus 304. Among them, the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304. The processor 301 can call the logical instructions in the memory 303 to execute the navigation trajectory prediction method based on the static and dynamic features of the ship.

[0135] In addition, when the logical instructions in the above-mentioned memory 303 are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc that can store program codes.

[0136] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the navigation trajectory prediction method based on the static and dynamic features of the ship provided by the above-mentioned various methods.

[0137] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the navigation trajectory prediction method based on the static and dynamic characteristics of the ship provided in the above-mentioned embodiments.

[0138] The device embodiments described above are schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0139] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0140] Finally, it should be noted that the above is only the preferred implementation of the present application, and the present application is not limited to the above embodiments. Other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the scope of protection of the present application.

Claims

1. A navigation trajectory prediction method based on static and dynamic characteristics of a ship, characterized in that: The method comprises: Obtaining ship static data corresponding to a current first time interval and ship dynamic data corresponding to a current second time interval of a currently sailing ship, wherein the first time interval is greater than the second time interval, and the first time interval is an integer multiple of the second time interval, the ship static data includes: ship size information, draft, displacement, and bow-stern draft difference, and the ship dynamic data includes: ship position, speed, and heading; Extracting ship dynamic features corresponding to the ship dynamic data; fusing the static data of the ship and the dynamic features of the ship to obtain a fused feature, wherein the static data of the ship and the dynamic features of the ship have the same dimension; The fused features are input into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval output by the trajectory prediction model, wherein the trajectory prediction model is trained based on the fused feature samples and the ship navigation trajectory samples.

2. The method for predicting the navigation trajectory based on static and dynamic characteristics of a ship according to claim 1, characterized in that: Inputting the fusion feature into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval output by the trajectory prediction model, including: The fused features are input into the trajectory prediction model, and the nonlinear correlation between the static data of the ship and the dynamic features of the ship is obtained through the trajectory prediction model; different weights are configured for the static data of the ship and the dynamic features of the ship based on the correlation; and the navigation trajectory of the ship in the next second time interval is predicted and output based on the static data of the ship and the dynamic features of the ship after the weights are assigned.

3. The navigation trajectory prediction method based on static and dynamic characteristics of a ship according to claim 1 or 2, characterized in that: The training process of the trajectory prediction model includes: Acquire training sample data, wherein the training sample data includes: the fusion sample and the ship navigation track sample, the fusion sample is obtained based on the ship static feature sample and the ship dynamic feature sample, the ship static feature sample is obtained by performing kernel principal component analysis on the ship static data sample, the ship dynamic feature sample is obtained by performing LSTM dynamic feature extraction on the ship dynamic data sample, and the ship static feature sample and the ship dynamic feature sample have the same dimension; The training sample data is input into the trajectory prediction model, and the nonlinear correlation between the static feature samples of the ship and the dynamic feature samples of the ship is automatically learned through the trajectory prediction model; different weights are configured for the static feature samples of the ship and the dynamic feature samples of the ship based on the correlation; the ship's navigation trajectory is predicted based on the weighted static feature samples and the dynamic feature samples of the ship; the predicted ship's navigation trajectory is compared with the consistency of the ship's navigation trajectory samples; the model parameters of the trajectory prediction model are adjusted based on the consistency until the number of iterations reaches a preset number, and it is determined that the trajectory prediction model training is completed.

4. The method for predicting the navigation trajectory based on static and dynamic characteristics of a ship according to claim 3, characterized in that: Ship static data samples include: static data of ships of different ship types and sizes; The ship static data samples are subjected to kernel principal component analysis to obtain the ship static feature samples, including: The cosine similarity kernel function is used to calculate the similarity between the static data samples of different ships and obtain the kernel matrix; Performing eigenvalue decomposition on the kernel matrix to extract eigenvalues ​​whose contribution rates are greater than a preset contribution rate; The ship static data samples corresponding to the extracted eigenvalues ​​are matrix constructed to obtain the ship static feature samples.

5. The method for predicting the navigation trajectory based on static and dynamic characteristics of a ship according to claim 3, characterized in that: The LSTM dynamic feature extraction is performed on the ship dynamic data samples to obtain the ship dynamic feature samples, including: The ship dynamic data samples are input into a pre-trained LSTM model to obtain the ship dynamic feature samples output by the LSTM model, wherein the LSTM model is trained based on initial samples of ship dynamic data and initial samples of ship dynamic features.

6. The method for predicting the navigation trajectory based on static and dynamic characteristics of a ship according to claim 5, characterized in that: The training process of the LSTM model includes: The initial samples of the ship dynamic data and the initial samples of the ship dynamic characteristics are input into the LSTM model to obtain the predicted samples of the ship dynamic characteristics output by the LSTM model; the mean square error of the predicted samples of the ship dynamic characteristics and the initial samples of the ship dynamic characteristics is calculated; the model parameters of the LSTM model are optimized based on the mean square error and a preset learning decay rate strategy, and the training of the LSTM model is stopped using the preset training stop requirement set based on the early stopping method.

7. The method for predicting the navigation trajectory based on static and dynamic characteristics of a ship according to claim 3, characterized in that: After obtaining the training sample data, it also includes: The training sample data is processed based on any one or more of a data deduplication strategy, an outlier removal strategy, a missing value supplementation strategy, a data resampling strategy, and a data normalization processing strategy.

8. A navigation trajectory prediction device based on static and dynamic characteristics of a ship, characterized in that: The device comprises: An acquisition module is used to acquire ship static data corresponding to a current first time interval of a currently sailing ship, and ship dynamic data corresponding to a current second time interval, wherein the first time interval is greater than the second time interval, and the first time interval is an integer multiple of the second time interval, the ship static data includes: ship size information, draft, displacement, and bow and stern draft difference, and the ship dynamic data includes: ship position, speed, and heading; An extraction module, used for extracting ship dynamic features corresponding to the ship dynamic data; A fusion module, used for fusing the static data of the ship with the dynamic features of the ship to obtain fusion features, wherein the static data of the ship and the dynamic features of the ship have the same dimension; The prediction module is used to input the fusion feature into a pre-trained trajectory prediction model to obtain the ship navigation trajectory of the next second time interval output by the trajectory prediction model, wherein the trajectory prediction model is trained based on the fusion feature samples and the ship navigation trajectory samples.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the navigation trajectory prediction method based on the static characteristics and dynamic characteristics of a ship as described in any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the navigation trajectory prediction method based on static characteristics and dynamic characteristics of a ship as described in any one of claims 1 to 7 are implemented.

Images