Ship navigation trajectory prediction method and system based on AIS

Abstract

The present invention provides a ship navigation track prediction method and system based on AIS, which relates to the technical field of ship navigation track prediction, including: taking the parameters to be optimized by the vSVR model as genes and randomly generating an initial population, obtaining the ship AIS data and inputting it into each vSVR model individual of the initial population for training; dividing the initial population into multiple evolutionary groups according to the similarities and differences of the individual output prediction values, and determining the individual fitness value according to the deviation between the prediction value and the actual value; using the fitness value as the basis for selecting operations in the genetic algorithm, performing independent evolution and overall evolution alternately on the evolutionary group, and iterating multiple times to determine the navigation track prediction model; inputting the AIS data of the ship to be tested into the navigation track prediction model to perform navigation track prediction. The present invention finds the optimal model parameters by dividing the evolutionary groups, performing independent evolution alternately on the evolutionary groups, and transmitting the settings of excellent individuals and overall evolution, thereby improving the prediction accuracy of the ship's navigation track.

Description

Technical Field

[0001] The present invention relates to the technical field of ship navigation track prediction, and in particular to a ship navigation track prediction method and system based on AIS. Background Art

[0002] Ship transportation has the advantages of large cargo capacity, low freight, and high economic benefits. It is one of the main modes of cargo transportation in my country and currently accounts for about two-thirds of my country's total international trade volume. The large load of ships leads to large inertia of movement. It is impossible to stop or turn sharply during navigation. It is easy to collide when sailing in waters with high traffic density, thus endangering life, property and even ecological environment safety. Therefore, it is particularly important to reduce ship collisions by analyzing ship AIS data to carry out ship navigation trajectory prediction and dynamic forecast of ship navigation behavior.

[0003] In the prior art, a ship trajectory prediction method based on machine learning and AIS data is disclosed with publication number "CN113283653B", which includes: preprocessing a historical AIS data set of a ship; dividing the preprocessed AIS data set according to the trajectory to obtain multiple trajectory feature data; then detecting and deleting erroneous data in the trajectory feature data; gridding each trajectory feature data, and dividing the prediction target of the trajectory point in each trajectory feature data into eight neighborhood grid directions; establishing a trajectory prediction model based on the xgboost algorithm, and predicting the trajectory points through the trajectory prediction model, so as to achieve the effect of predicting the ship's navigation trajectory through the established machine learning model. However, when establishing the trajectory prediction model, it is necessary to determine some parameters of the model in advance, and the determination of the model parameters is often affected by subjective factors of the operator, and there is a problem that the determined model parameters are inappropriate, which limits the prediction ability and prediction accuracy of the model.

[0004] Based on the above defects, in the prior art, a ship trajectory prediction method based on the GA-vSVR algorithm (main classification number G06F30 / 27) with the publication number "CN116011335A" can output the trajectory of the ship in the future based on the historical AIS data of the ship. The model first applies vSVR to the field of ship trajectory prediction, and uses a genetic algorithm to optimize its parameters, so as to determine the appropriate model parameters and ensure the model's ability to predict the ship's navigation trajectory. However, this prior art still has major defects, such as: the evolution process of the genetic algorithm in the prior art is carried out as a whole, which makes the population easy to fall into a stagnant state, and the optimization of the model parameters is easy to fall into a local optimal solution, that is, the above method has the problem of poor quality of model parameters and limited model prediction ability.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The object of the present invention is to provide a method and system for predicting the navigation trajectory of a ship based on AIS, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A ship navigation track prediction method based on AIS includes the following steps:

[0009] S1, obtain AIS data arranged in a continuous time series and environmental data aligned with the AIS data time series, preprocess the AIS data and environmental data, and randomly divide the AIS data into a training set and a test set. The preprocessing includes removing outliers, supplementing missing values, and normalizing.

[0010] S2, establish a vSVR model with parameters to be optimized, use the parameters to be optimized by the vSVR model as genes and randomly generate an initial population, the initial population consists of multiple vSVR model individuals with different genes, use the AIS data in the training set to train the vSVR model individuals, and divide the individuals in the initial population into multiple evolutionary groups according to the size of similarity;

[0011] S3, based on environmental data, the deviation between the longitude coordinate prediction value and the longitude coordinate, and the deviation between the latitude coordinate prediction value and the latitude coordinate, the fitness value of the vSVR model individual in each evolution group is calculated, and the fitness value is used to evaluate the quality of the vSVR model individual;

[0012] S4, each group of evolutionary groups is subjected to a round of independent evolution according to the standard operation of the genetic algorithm, and the excellent vSVR model individuals are determined based on the fitness values ​​of the vSVR model individuals after evolution, and the excellent vSVR model individuals are transferred to other evolutionary groups;

[0013] S5, using the standard operation of the genetic algorithm to carry out a round of overall evolution on all the evolutionary groups after independent evolution, and placing the newly generated vSVR model individuals into the evolutionary group where their parent vSVR model individuals are located;

[0014] S6, repeating steps S3-S5 until the optimal vSVR model individual is found, or after reaching a predetermined number of evolutions, the vSVR model individual with the highest fitness value in all evolution groups is taken as the optimal vSVR model individual;

[0015] S7, inputting the AIS data in the test set into the optimal vSVR model individual for testing, and taking the optimal vSVR model individual as the navigation trajectory prediction model after the test is qualified, and returning to step S2 if the test is unqualified, and re-determining the value range of the parameters to be optimized for the vSVR model;

[0016] S8, after pre-processing the AIS data of the ship to be tested, input it into the navigation track prediction model to predict the navigation track of the ship to be tested.

[0017] Furthermore, the AIS data includes longitude coordinates, latitude coordinates, speed and heading, and the environmental data includes wind speed, current speed and wave height at the ship.

[0018] Furthermore, a vSVR model with parameters to be optimized is established. The parameters to be optimized by the vSVR model are used as genes and the initial population is randomly generated. The initial population consists of multiple vSVR model individuals with different genes. The vSVR model individuals are trained using the AIS data in the training set, and the individuals in the initial population are divided into multiple evolutionary groups according to their similarity. The specific logic is as follows:

[0019] S21, establishing a vSVR model with parameters to be optimized, wherein the input of the vSVR model is the longitude coordinates, latitude coordinates, speed, and heading of the previous multiple consecutive acquisition moments, and the output is the longitude coordinates and latitude coordinates of the subsequent multiple consecutive moments, taking the parameters to be optimized by the vSVR model as genes, determining the value range of the parameters, and determining the binary code length of the genes according to the value range of the parameters, and randomly generating an initial population according to the determined code length, wherein the initial population consists of multiple vSVR model individuals with different genes;

[0020] S22, input the AIS data in the training set into each vSVR model individual for training, and calculate the similarity coefficients of different vSVR model individuals according to the similarities and differences of the longitude coordinate prediction values ​​and latitude coordinate prediction values ​​output by each vSVR model individual after training. The calculation formula is as follows:

[0021]

[0022] Among them, the actual values ​​corresponding to the predicted values ​​of the vSVR model individuals in the training set are numbered according to the numbering method of 1, 2, 3, ..., n, x'(j, k1) represents the predicted longitude coordinate output by the k1th vSVR model individual for the jth longitude coordinate, x'(j, k2) represents the predicted longitude coordinate output by the k2th vSVR model individual for the jth longitude coordinate, j∈(1,n), n represents the total number of actual values ​​corresponding to the predicted values ​​of the vSVR model individuals in the training set, and n∈N +, k represents the number of different vSVR model individuals, and k = 1, 2, 3, ..., Z, Z is the total number of vSVR model individuals, and Z∈N + , and k1∈(1,2,…,k,…,Z), k2∈(1,2,…,k,…,Z), k1≠k2, Sx(k1,k2) represents the longitude prediction similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual;

[0023] Wherein, y'(j,k1) represents the latitude coordinate prediction value output by the k1th vSVR model individual for the jth latitude coordinate, y'(j,k2) represents the latitude coordinate prediction value output by the k2th vSVR model individual for the jth latitude coordinate, Sy(k1,k2) represents the latitude prediction similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual, and S(k1,k2) represents the similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual;

[0024] S23, based on the similarity coefficient, the K-means clustering algorithm is used to group the Z vSVR model individuals in the initial population to divide them into multiple evolutionary groups.

[0025] Furthermore, based on the environmental data, the deviation between the predicted longitude coordinate and the longitude coordinate, and the deviation between the predicted latitude coordinate and the latitude coordinate, the fitness value of the vSVR model individual in each evolution group is calculated. The fitness value is used to evaluate the specific logic of the quality of the vSVR model individual:

[0026] S31, based on the deviation between the longitude coordinate prediction value and the longitude coordinate, and the deviation between the latitude coordinate prediction value and the latitude coordinate, a distance deviation coefficient and a deviation extreme value coefficient are generated. The distance deviation coefficient is used to evaluate the accuracy of the vSVR model individual, and the deviation extreme value coefficient is used to evaluate the stability of the vSVR model individual. The calculation formula is as follows:

[0027]

[0028] Among them, x(j) represents the longitude coordinate corresponding to the longitude coordinate prediction value x'(j,k), y(j) represents the latitude coordinate corresponding to the latitude coordinate prediction value x'(j,k), JLpc(j,k) represents the distance deviation value between the predicted point and the actual coordinate point output by the kth vSVR model individual for the jth actual coordinate point, represents the distance deviation coefficient of the kth vSVR model individual, JLpc(j,k) max represents the maximum value of n distance deviation values ​​in the kth vSVR model individual, JLpc(j,k) minrepresents the minimum value of n distance deviation values ​​in the k-th vSVR model individual, PCjz(j,k) represents the deviation extreme value coefficient of the k-th vSVR model individual;

[0029] S32, combining the environmental data and the distance deviation value in step S31, generating a precision ring ratio coefficient, which is used to evaluate the robustness of the vSVR model individual, and the calculation formula is as follows:

[0030]

[0031] Among them, v f (j), v l (j) H l (j) represents the wind speed, current speed and wave height at the same time as the longitude coordinate x(j), HJ(j) represents the environmental coefficient at the same time as the longitude coordinate x(j), λ1, λ2 and λ3 are all preset proportional coefficients, and on the basis of λ1+λ2+λ3=1, and 0<λ1<λ3<λ2<1, HJ(j1) represents the maximum value of the environmental coefficient among the n environmental coefficients, HJ(j2) represents the minimum value of the environmental coefficient among the n environmental coefficients, JLpc(j1,k) represents the distance deviation value between the predicted point and the actual coordinate point output by the kth vSVR model individual for the actual coordinate point at the moment of the maximum value of the environmental coefficient HJ(j1), JLpc(j2,k) represents the distance deviation value between the predicted point and the actual coordinate point output by the kth vSVR model individual for the actual coordinate point at the moment of the minimum value of the environmental coefficient HJ(j2), and JHB(k) represents the precision ring ratio coefficient of the kth vSVR model individual;

[0032] S33, scoring the distance deviation coefficient, the deviation extreme value coefficient and the precision ring ratio coefficient according to the expert scoring method, generating an accuracy evaluation coefficient, a stability evaluation coefficient and a robustness evaluation coefficient between 0 and 1, and decreasing as the distance deviation coefficient, the deviation extreme value coefficient and the precision ring ratio coefficient increase, and determining the weights of the accuracy evaluation coefficient, the stability evaluation coefficient and the robustness evaluation coefficient in the fitness value calculation according to the expert scoring method;

[0033] S34, combining the accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient, calculates the fitness value of the vSVR model individuals in each evolution group. The calculation formula is as follows:

[0034] SYD(k)=ω1*JD(k)+ω2*WDX(k)+ω3*LBX(k)

[0035] Among them, JD(k), WDX(k), and LBX(k) represent the accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient of the kth vSVR model individual, SYD(k) represents the fitness value of the kth vSVR model individual, ω1, ω2, and ω3 are all weights determined by experts, and ω1+ω2+ω3=1.

[0036] Furthermore, the criteria for determining the excellent vSVR model individuals are as follows: the fitness values ​​of the vSVR model individuals in all the evolutionary groups are sorted from large to small, and the vSVR model individuals with the highest fitness values ​​are regarded as excellent vSVR model individuals.

[0037] Furthermore, the specific logic of placing the newly generated vSVR model individual into the evolutionary group where its parent vSVR model individual is located is as follows: if a newly generated vSVR model individual is generated by two vSVR model individuals in the same evolutionary group, then the newly generated vSVR model individual is placed in the evolutionary group where the two vSVR model individuals coexist; if a newly generated vSVR model individual is generated by two vSVR model individuals in different evolutionary groups, then the newly generated vSVR model individual is copied and pasted and placed in the respective evolutionary groups of the two vSVR model individuals.

[0038] A ship navigation track prediction system based on AIS, used to execute the above-mentioned ship navigation track prediction method based on AIS, comprising:

[0039] A data processing module is used to obtain AIS data arranged in a continuous time series and environmental data aligned with the AIS data time series. After preprocessing the AIS data and environmental data, the AIS data is randomly divided into a training set and a test set. The preprocessing includes removing outliers, supplementing missing values, and normalizing.

[0040] The grouping module is used to establish a vSVR model with parameters to be optimized. The parameters to be optimized by the vSVR model are used as genes and the initial population is randomly generated. The initial population consists of multiple vSVR model individuals with different genes. The vSVR model individuals are trained using the AIS data in the training set, and the individuals in the initial population are divided into multiple evolutionary groups according to the size of similarity.

[0041] The fitness evaluation module is used to calculate the fitness value of the vSVR model individuals in each evolution group based on environmental data, the longitude coordinate prediction value and the deviation of the longitude coordinate, and the latitude coordinate prediction value and the deviation of the latitude coordinate. The fitness value is used to evaluate the quality of the vSVR model individuals;

[0042] The independent evolution module is used to perform a round of independent evolution on each evolution group according to the standard operation of the genetic algorithm, and determine the excellent vSVR model individuals based on the fitness value of the vSVR model individuals after evolution, and transfer the excellent vSVR model individuals to other evolution groups;

[0043] The overall evolution module is used to use the standard operation of the genetic algorithm to carry out a round of overall evolution on all the evolutionary groups after independent evolution, and put the newly generated vSVR model individuals into the evolutionary group where their parent vSVR model individuals are located;

[0044] An iterative judgment module is used to repeat steps S3-S5 until the optimal vSVR model individual is found, or after reaching a predetermined number of evolution times, the vSVR model individual with the highest fitness value in all evolution groups is taken as the optimal vSVR model individual;

[0045] A testing module is used to input the AIS data in the test set into the optimal vSVR model individual for testing, and use the optimal vSVR model individual as the navigation trajectory prediction model after the test is qualified, and return to step S2 when the test is unqualified to re-determine the value range of the parameters that need to be optimized in the vSVR model;

[0046] The navigation track prediction module is used to pre-process the AIS data of the ship to be tested and input it into the navigation track prediction model to predict the navigation track of the ship to be tested.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The ship navigation trajectory prediction method and system based on AIS of the present invention divides the vSVR model individuals in the initial population into multiple evolutionary groups according to similarity, and then uses each evolutionary group to independently evolve and transfer excellent individuals, and all evolutionary groups to evolve as a whole in an alternating manner for data processing, so as to avoid the problem of the population falling into a stagnant state, and help to better explore the global optimal solution while maintaining the diversity of the population, so as to search for the optimal model parameters, improve the prediction accuracy and prediction efficiency of the navigation trajectory prediction model for the ship navigation trajectory, and thus improve the ship navigation safety level. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of the method for predicting the ship's navigation trajectory based on AIS in the present invention;

[0050] Figure 2 This is a module unit diagram of the ship navigation trajectory prediction system based on AIS in the present invention;

[0051] Figure 3 The figure is a logical schematic diagram of the ship navigation trajectory prediction method based on AIS in the present invention. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0053] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0054] Embodiment 1:

[0055] See also Figure 1 and Figure 3 The present invention provides a method for predicting a ship's navigation trajectory based on AIS, comprising the following steps:

[0056] S1, obtaining AIS data arranged in a continuous time series and environmental data aligned with the AIS data time series, and preprocessing the AIS data and the environmental data, and then randomly dividing the AIS data into a training set and a test set. The preprocessing includes conventional outlier removal, missing value supplementation, and normalization processing, etc. Specifically, the existing data preprocessing method can be used, which will not be elaborated here;

[0057] Among them, AIS data includes longitude coordinates, latitude coordinates, speed and heading; environmental data includes wind speed, current speed and wave height at the ship. Wind speed can be obtained by a cup wheel anemometer, current speed can be obtained by a mechanical current meter or an electronic current meter, and wave height can be obtained by a radar wave height meter or a pressure wave height meter. Specifically, the wind speed, current speed and wave height at multiple locations of the ship can be collected and the average value can be calculated, and the average value can be used as the final wind speed, current speed and wave height data.

[0058] S2, establish a vSVR model with parameters to be optimized, use the parameters to be optimized by the vSVR model as genes and randomly generate an initial population, the initial population consists of multiple vSVR model individuals with different genes, use the AIS data in the training set to train the vSVR model individuals, and divide the individuals in the initial population into multiple evolutionary groups according to the similarity, including the following steps:

[0059] S21, establishing a vSVR model with parameters to be optimized, and the input of the vSVR model is the longitude coordinates, latitude coordinates, speed, and heading of the previous multiple consecutive acquisition moments, and the output is the longitude coordinates and latitude coordinates of the subsequent multiple consecutive moments, taking the parameters to be optimized by the vSVR model as genes, determining the value range of the parameters, and determining the binary code length of the genes according to the value range of the parameters, and randomly generating an initial population according to the determined code length, the initial population is composed of multiple vSVR model individuals with different genes, vSVR (v-Support Vector Regression) is an improved support vector regression technology, which is a prior art and will not be described in detail here;

[0060] As an implementation method, the parameters to be optimized by the vSVR model are c for controlling the model complexity and learning ability, and γ for affecting the generalization ability of the model, and the value ranges of c and γ are both 0-100;

[0061] As an implementation method, the input of the vSVR model is the longitude, latitude, speed, and heading of the previous 72 consecutive collection moments, and the output is the longitude and latitude of the next 12 collection moments. Of course, other settings such as 48 to 12 and 36 to 6 are also possible. The specific settings are made by the staff according to the actual situation and will not be elaborated here.

[0062] S22, input the AI ​​S data in the training set into each vSVR model individual for training, and calculate the similarity coefficients of different vSVR model individuals according to the similarities and differences of the longitude coordinate prediction values ​​and latitude coordinate prediction values ​​output by each vSVR model individual after training. The calculation formula is as follows:

[0063]

[0064] It should be noted that the actual values ​​corresponding to the predicted values ​​of the vSVR model individuals in the training set are numbered according to the numbering method of 1, 2, 3, ..., n. x'(j, k1) represents the predicted longitude coordinate output by the k1th vSVR model individual for the jth longitude coordinate, and x'(j, k2) represents the predicted longitude coordinate output by the k2th vSVR model individual for the jth longitude coordinate. j∈(1,n), n represents the total number of actual values ​​corresponding to the predicted values ​​of the vSVR model individuals in the training set, and n∈N+ , k represents the number of different vSVR model individuals, and k = 1, 2, 3, ..., Z, Z is the total number of vSVR model individuals, and Z∈N + , and k1∈(1,2,…,k,…,Z), k2∈(1,2,…,k,…,Z), k1≠k2;

[0065] It should be noted that Sx(k1, k2) represents the longitude prediction similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual, and the closer the longitude coordinate prediction values ​​output by the k1th vSVR model individual and the k2th vSVR model individual for the same longitude coordinate are, the more similar the k1th vSVR model individual and the k2th vSVR model individual are in longitude coordinate prediction, and the smaller the longitude prediction similarity coefficient Sx(k1, k2) is.

[0066] It should be noted that y'(j, k1) represents the latitude coordinate prediction value output by the k1th vSVR model individual for the jth latitude coordinate, and y'(j, k2) represents the latitude coordinate prediction value output by the k2th vSVR model individual for the jth latitude coordinate;

[0067] It should be noted that Sy(k1, k2) represents the latitude prediction similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual, and the closer the latitude coordinate prediction values ​​output by the k1th vSVR model individual and the k2th vSVR model individual for the same latitude coordinate are, the more similar the k1th vSVR model individual and the k2th vSVR model individual are in latitude coordinate prediction, and the smaller the latitude prediction similarity coefficient Sy(k1, k2) is.

[0068] It should be noted that S(k1, k2) represents the similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual, which is obtained by adding the longitude prediction similarity coefficient and the latitude prediction similarity coefficient and then calculating the average value, thereby comprehensively considering the longitude coordinate prediction similarity and latitude coordinate prediction similarity of the k1th and k2th vSVR model individuals, thereby achieving the effect of comprehensively evaluating the comprehensive prediction similarity of the k1th and k2th vSVR model individuals;

[0069] S23, based on the similarity coefficient, the K-means clustering algorithm is used to group the Z vSVR model individuals in the initial population into multiple evolutionary groups;

[0070] It should be noted that, with the similarity coefficient as the reference benchmark, the K-means clustering algorithm is used to divide the Z vSVR model individuals into multiple evolutionary groups, so that the vSVR model individuals in each evolutionary group have high similarity, which helps to maintain diversity and explore different optimization parameters in the search space, providing a better initial population and diversity for the subsequent operation of the genetic algorithm;

[0071] It should be noted that the K-means clustering algorithm can be used to group the Z vSVR model individuals using existing technologies, as follows:

[0072] 1) Determine the number of similar groups to be divided, such as marking the number of similar groups as z, and z <Z,z∈N + ;

[0073] 2) Randomly select z vSVR model individuals as the initial clustering centers;

[0074] 3) Calculate the correlation coefficient between the vSVR model individual and the cluster center. The smaller the correlation coefficient, the closer the vSVR model individual is to the cluster center. In this way, each vSVR model individual is assigned to the group with the closest cluster center.

[0075] 4) For each group, calculate the average correlation coefficient between all vSVR model individuals and the cluster center, and update the cluster center according to the average value;

[0076] 5) Repeat steps 3) and 4) until the cluster center no longer changes or the predetermined number of iterations is reached. At this time, each group is an evolutionary group, and the vSVR model individuals in each evolutionary group have a high similarity.

[0077] S3, based on the environmental data, the longitude coordinate prediction value and the deviation of the longitude coordinate, the latitude coordinate prediction value and the deviation of the latitude coordinate, calculate the fitness value of the vSVR model individual in each evolution group, the fitness value is used to evaluate the quality of the vSVR model individual, including the following steps:

[0078] S31, based on the deviation between the longitude coordinate prediction value and the longitude coordinate, and the deviation between the latitude coordinate prediction value and the latitude coordinate, a distance deviation coefficient and a deviation extreme value coefficient are generated. The distance deviation coefficient is used to evaluate the accuracy of the vSVR model individual, and the deviation extreme value coefficient is used to evaluate the stability of the vSVR model individual. The calculation formula is as follows:

[0079]

[0080] It should be noted that x(j) represents the longitude coordinate corresponding to the longitude coordinate prediction value x'(j,k), y(j) represents the latitude coordinate corresponding to the latitude coordinate prediction value x'(j,k), JLpc(j,k) represents the distance deviation value between the predicted point output by the k-th vSVR model individual for the j-th actual coordinate point and the actual coordinate point, and the larger the distance deviation value JLpc(j,k), the larger the deviation between the predicted point output by the k-th vSVR model individual for the j-th actual coordinate point and the actual coordinate point;

[0081] It should be noted that It represents the distance deviation coefficient of the k-th vSVR model individual, which is the average of n distance deviation values ​​and is used to measure the prediction accuracy of the k-th vSVR model individual. The larger the distance deviation coefficient, the worse the prediction accuracy of the k-th vSVR model individual.

[0082] It should be noted that JLpc(j,k) max represents the maximum value of n distance deviation values ​​in the kth vSVR model individual, JLpc(j,k) min represents the minimum value of the n distance deviation values ​​in the k-th vSVR model individual, PCjz(j,k) represents the deviation extreme value coefficient of the k-th vSVR model individual, which is calculated by subtracting the minimum distance deviation value from the maximum distance deviation value and dividing it by the distance deviation coefficient. It is used to measure the stability of the prediction of the k-th vSVR model individual. The larger the deviation extreme value coefficient, the worse the stability of the prediction of the k-th vSVR model individual.

[0083] S32, combining the environmental data and the distance deviation value in step S31, generating a precision ring ratio coefficient, which is used to evaluate the robustness of the vSVR model individual, and the calculation formula is as follows:

[0084]

[0085] It should be noted that v f (j), v l (j) H l (j) represents the wind speed, current speed and wave height at the same time as the longitude coordinate x(j), HJ(j) represents the environmental coefficient at the same time as the longitude coordinate x(j), and the environmental coefficient HJ(j) is composed of three factors: wind speed, current speed and wave height. By comprehensively considering wind speed, current speed and wave height, the final environmental coefficient can reflect the external environment from a comprehensive level;

[0086] It should be noted that λ1, λ2, and λ3 are all preset proportional coefficients, and λ1 represents the wind speed v f (j) is the weight in the calculation of the environmental factor HJ(j), and λ2 represents the flow velocity vl (j) is the weight in the calculation of the environmental factor HJ(j), and λ3 represents the wave height H l (j) The weight in the calculation of the environmental factor HJ(j). Since the flow velocity has a direct impact on the navigation track of the ship, the wave has a significant impact on the stability and heading of the ship, and the wind speed has little impact on the navigation of modern motor ships, on the basis of λ1+λ2+λ3=1, let 0<λ1<λ3<λ2<1, and the specific values ​​of λ1, λ2, and λ3 are generally determined by those skilled in the art according to actual conditions;

[0087] As an implementation mode, the value range of λ1 is 0.1-0.2, the value range of λ2 is 0.5-0.7, and the value range of λ3 is 0.2-0.3;

[0088] It should be noted that HJ(j1) represents the maximum value of the environmental coefficient among the n environmental coefficients, HJ(j2) represents the minimum value of the environmental coefficient among the n environmental coefficients, JLpc(j1,k) represents the distance deviation between the predicted point and the actual coordinate point output by the kth vSVR model individual for the actual coordinate point at the moment when the environmental coefficient is the maximum value HJ(j1), and JLpc(j2,k) represents the distance deviation between the predicted point and the actual coordinate point output by the kth vSVR model individual for the actual coordinate point at the moment when the environmental coefficient is the minimum value HJ(j2). The distance deviation value between points, JHB(k) represents the precision ring ratio coefficient of the k-th vSVR model individual, which is the ratio of |JLpc(j1,k)-JLpc(j2,k)| to HJ(j1)-HJ(j2), and the smaller the ratio of |JLpc(j1,k)-JLpc(j2,k)| to HJ(j1)-HJ(j2), it means that when the k-th vSVR model individual outputs the predicted value, it can still maintain good output stability under the condition of large changes in the external environment, and the k-th vSVR model individual is more robust;

[0089] S33, scoring the distance deviation coefficient, the deviation extreme value coefficient and the precision ring ratio coefficient according to the expert scoring method, generating an accuracy evaluation coefficient, a stability evaluation coefficient and a robustness evaluation coefficient between 0 and 1, and decreasing as the distance deviation coefficient, the deviation extreme value coefficient and the precision ring ratio coefficient increase, and determining the weights of the accuracy evaluation coefficient, the stability evaluation coefficient and the robustness evaluation coefficient in the fitness value calculation according to the expert scoring method;

[0090] It should be noted that the experts invited in the expert scoring method include but are not limited to navigation technology experts, marine meteorology experts, marine geographic information system experts, data analysis experts, artificial intelligence and machine learning experts, and the number of experts invited is no less than 10. When determining the evaluation coefficient and weight, the maximum and minimum values ​​are removed and the average of other values ​​is calculated;

[0091] S34, combining the accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient, calculates the fitness value of the vSVR model individuals in each evolution group. The calculation formula is as follows:

[0092] SYD(k)=ω1*JD(k)+ω2*WDX(k)+ω3*LBX(k)

[0093] It should be noted that JD(k), WDX(k), and LBX(k) represent the accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient of the kth vSVR model individual, and the accuracy evaluation coefficient JD(k), stability evaluation coefficient WDX(k), and robustness evaluation coefficient LBX(k) are related to the distance deviation coefficient The deviation extreme value coefficient PCjz(j,k) and the precision ring ratio coefficient JHB(k) correspond one to one. SYD(k) represents the fitness value of the kth vSVR model individual, which is composed of three evaluation coefficients: accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient. This comprehensively considers the accuracy, stability, and robustness of the kth vSVR model individual, and achieves the effect of comprehensively evaluating the kth vSVR model individual. The larger the accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient, the better the comprehensive performance of the kth vSVR model individual, and the larger the fitness value SYD(k).

[0094] It should be noted that ω1, ω2, and ω3 are all weights determined by experts, ω1 represents the weight of the accuracy evaluation coefficient JD(k) in the calculation of the fitness value SYD(k), ω2 represents the weight of the stability evaluation coefficient WDX(k) in the calculation of the fitness value SYD(k), ω3 represents the weight of the robustness evaluation coefficient LBX(k) in the calculation of the fitness value SYD(k), and ω1+ω2+ω3=1.

[0095] S4, each group of evolutionary groups is subjected to a round of independent evolution according to the standard operation of the genetic algorithm, and the excellent vSVR model individuals are determined based on the fitness values ​​of the vSVR model individuals after evolution, and the excellent vSVR model individuals are transferred to other evolutionary groups;

[0096] It should be noted that both the independent evolution and the overall evolution of the vSVR model individuals described below need to be retrained using the data in the training set. After the training is completed, the fitness value is calculated by the method of step S3. The standard operations of the genetic algorithm include selection, crossover, and mutation. The selection operation can be performed by roulette selection based on the fitness value. The crossover operation is to exchange some genes of the selected vSVR model individual to generate a new vSVR model individual. The mutation operation is to randomly change some sites of the gene on the vSVR model individual. This is a prior art and will not be described in detail here.

[0097] Among them, the judgment standard of excellent vSVR model individuals is: sort the fitness values ​​of vSVR model individuals in all evolutionary groups from large to small, and take the vSVR model individuals with fitness values ​​in the top row as excellent vSVR model individuals. Specifically, the vSVR model individuals with fitness values ​​in the top three, top five, top ten, top 5% or top 10% can be regarded as excellent vSVR model individuals, and there is no restriction here. The fitness value calculation method of the evolved vSVR model individuals is the same as that in step S3, and will not be repeated here.

[0098] S5, using the standard operation of the genetic algorithm to carry out a round of overall evolution on all the evolutionary groups after independent evolution, and placing the newly generated vSVR model individuals into the evolutionary group where their parent vSVR model individuals are located;

[0099] It should be noted that the specific logic of placing the newly generated vSVR model individual into the evolutionary group where its parent vSVR model individual is located is as follows: if a newly generated vSVR model individual is generated by two vSVR model individuals in the same evolutionary group, then the newly generated vSVR model individual is placed in the evolutionary group where the two vSVR model individuals coexist; if a newly generated vSVR model individual is generated by two vSVR model individuals in different evolutionary groups, then the newly generated vSVR model individual is copied and pasted and placed in the respective evolutionary groups of the two vSVR model individuals.

[0100] S6, repeating steps S3-S5 until the optimal vSVR model individual is found, or after reaching a predetermined number of evolutions, the vSVR model individual with the highest fitness value in all evolution groups is taken as the optimal vSVR model individual;

[0101] It should be noted that the specific logic of repeating steps S3-S5 until the optimal vSVR model individual is found is: a fitness threshold is set in advance, and when the fitness of a vSVR model individual reaches the fitness threshold, the vSVR model individual is used as the optimal vSVR model individual. The fitness threshold can be set by the staff based on past experience, such as 0.8, 0.9, etc., and there is no restriction here.

[0102] S7, inputting the AIS data in the test set into the optimal vSVR model individual for testing, and taking the optimal vSVR model individual as the navigation trajectory prediction model after the test is qualified, and returning to step S2 if the test is unqualified, and re-determining the value range of the parameters to be optimized for the vSVR model;

[0103] It should be noted that the standard for passing the test is that the prediction accuracy is above 90%, 95% or 98%, which is set by the staff according to actual conditions and is not restricted here.

[0104] S8, pre-processing the AIS data of the ship to be tested and inputting it into a navigation track prediction model to predict the navigation track of the ship to be tested;

[0105] It should be noted that the method for preprocessing the AIS data of the ship to be tested is the same as step S1, and also includes conventional removal of outliers, supplementation of missing values, normalization processing, etc.

[0106] Embodiment 2:

[0107] See also Figure 2 The present invention provides a ship navigation track prediction system based on AIS, which is used to execute the ship navigation track prediction method based on AIS in the above-mentioned embodiment 1, comprising:

[0108] A data processing module is used to obtain AIS data arranged in a continuous time series and environmental data aligned with the AIS data time series. After preprocessing the AIS data and environmental data, the AIS data is randomly divided into a training set and a test set. The preprocessing includes removing outliers, supplementing missing values, and normalizing.

[0109] The grouping module is used to establish a vSVR model with parameters to be optimized. The parameters to be optimized by the vSVR model are used as genes and the initial population is randomly generated. The initial population consists of multiple vSVR model individuals with different genes. The vSVR model individuals are trained using the AIS data in the training set, and the individuals in the initial population are divided into multiple evolutionary groups according to the size of similarity.

[0110] The fitness evaluation module is used to calculate the fitness value of the vSVR model individuals in each evolution group based on environmental data, the longitude coordinate prediction value and the deviation of the longitude coordinate, and the latitude coordinate prediction value and the deviation of the latitude coordinate. The fitness value is used to evaluate the quality of the vSVR model individuals;

[0111] The independent evolution module is used to perform a round of independent evolution on each evolution group according to the standard operation of the genetic algorithm, and determine the excellent vSVR model individuals based on the fitness value of the vSVR model individuals after evolution, and transfer the excellent vSVR model individuals to other evolution groups;

[0112] The overall evolution module is used to use the standard operation of the genetic algorithm to carry out a round of overall evolution on all the evolutionary groups after independent evolution, and put the newly generated vSVR model individuals into the evolutionary group where their parent vSVR model individuals are located;

[0113] An iterative judgment module is used to repeat steps S3-S5 until the optimal vSVR model individual is found, or after reaching a predetermined number of evolution times, the vSVR model individual with the highest fitness value in all evolution groups is taken as the optimal vSVR model individual;

[0114] A testing module is used to input the AIS data in the test set into the optimal vSVR model individual for testing, and use the optimal vSVR model individual as the navigation trajectory prediction model after the test is qualified, and return to step S2 when the test is unqualified to re-determine the value range of the parameters that need to be optimized in the vSVR model;

[0115] The navigation track prediction module is used to pre-process the AIS data of the ship to be tested and input it into the navigation track prediction model to predict the navigation track of the ship to be tested.

[0116] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0117] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0118] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0119] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A ship navigation trajectory prediction method based on AIS, characterized in that: The steps include: S1, obtain AIS data arranged in a continuous time series and environmental data aligned with the AIS data time series, preprocess the AIS data and environmental data, and randomly divide the AIS data into a training set and a test set. The preprocessing includes removing outliers, supplementing missing values, and normalizing. S2, establish a vSVR model with parameters to be optimized, use the parameters to be optimized by the vSVR model as genes and randomly generate an initial population, the initial population consists of multiple vSVR model individuals with different genes, use the AIS data in the training set to train the vSVR model individuals, and divide the individuals in the initial population into multiple evolutionary groups according to the size of similarity; S3, based on environmental data, the deviation between the longitude coordinate prediction value and the longitude coordinate, and the deviation between the latitude coordinate prediction value and the latitude coordinate, the fitness value of the vSVR model individual in each evolution group is calculated, and the fitness value is used to evaluate the quality of the vSVR model individual; S4, each evolution group is operated according to the standard of genetic algorithm to perform a round of independent evolution, and the excellent vSVR model individuals are determined based on the fitness values ​​of the vSVR model individuals after evolution, and the excellent vSVR model individuals are transferred to other evolution groups; S5, using the standard operation of the genetic algorithm to carry out a round of overall evolution on all the evolutionary groups after independent evolution, and placing the newly generated vSVR model individuals into the evolutionary group where their parent vSVR model individuals are located; S6, repeating steps S3-S5 until the optimal vSVR model individual is found, or after reaching a predetermined number of evolutions, the vSVR model individual with the highest fitness value in all evolution groups is taken as the optimal vSVR model individual; S7, inputting the AIS data in the test set into the optimal vSVR model individual for testing, and taking the optimal vSVR model individual as the navigation trajectory prediction model after the test is qualified, and returning to step S2 if the test is unqualified, and re-determining the value range of the parameters to be optimized for the vSVR model; S8, after pre-processing the AIS data of the ship to be tested, input it into the navigation track prediction model to predict the navigation track of the ship to be tested.

2. The method for predicting a ship's navigation trajectory based on AIS according to claim 1, characterized in that: The AIS data includes longitude coordinates, latitude coordinates, speed and heading, and the environmental data includes wind speed, current speed and wave height at the ship.

3. The method for predicting a ship's navigation trajectory based on AIS according to claim 2, characterized in that: A vSVR model with parameters to be optimized is established. The parameters to be optimized by the vSVR model are used as genes and the initial population is randomly generated. The initial population consists of multiple vSVR model individuals with different genes. The vSVR model individuals are trained using the AIS data in the training set, and the individuals in the initial population are divided into multiple evolutionary groups according to their similarity. The specific logic is as follows: S21, establishing a vSVR model with parameters to be optimized, wherein the input of the vSVR model is the longitude coordinates, latitude coordinates, speed, and heading of the previous multiple consecutive acquisition moments, and the output is the longitude coordinates and latitude coordinates of the subsequent multiple consecutive moments, taking the parameters to be optimized by the vSVR model as genes, determining the value range of the parameters, and determining the binary code length of the genes according to the value range of the parameters, and randomly generating an initial population according to the determined code length, wherein the initial population consists of multiple vSVR model individuals with different genes; S22, input the AIS data in the training set into each vSVR model individual for training, and calculate the similarity coefficients of different vSVR model individuals according to the similarities and differences of the longitude coordinate prediction values ​​and latitude coordinate prediction values ​​output by each vSVR model individual after training. The calculation formula is as follows: Among them, the actual values ​​corresponding to the predicted values ​​of the vSVR model individuals in the training set are numbered according to the numbering method of 1, 2, 3, ..., n, x'(j, k1) represents the predicted longitude coordinate output by the k1th vSVR model individual for the jth longitude coordinate, x'(j, k2) represents the predicted longitude coordinate output by the k2th vSVR model individual for the jth longitude coordinate, j∈(1,n), n represents the total number of actual values ​​corresponding to the predicted values ​​of the vSVR model individuals in the training set, and n∈N + , k represents the number of different vSVR model individuals, and k = 1, 2, 3, ..., Z, Z is the total number of vSVR model individuals, and Z∈N + , and k1∈(1,2,…,k,…,Z), k2∈(1,2,…,k,…,Z), k1≠k2, Sx(k1,k2) represents the longitude prediction similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual; Wherein, y'(j,k1) represents the latitude coordinate prediction value output by the k1th vSVR model individual for the jth latitude coordinate, y'(j,k2) represents the latitude coordinate prediction value output by the k2th vSVR model individual for the jth latitude coordinate, Sy(k1,k2) represents the latitude prediction similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual, and S(k1,k2) represents the similarity coefficient between the k1th vSVR model individual and the k2th vSVR model individual; S23, based on the similarity coefficient, the K-means clustering algorithm is used to group the Z vSVR model individuals in the initial population to divide them into multiple evolutionary groups.

4. The method for predicting a ship's navigation trajectory based on AIS according to claim 3, characterized in that: Based on environmental data, the deviation between the predicted longitude coordinate and the longitude coordinate, and the deviation between the predicted latitude coordinate and the latitude coordinate, the fitness value of the vSVR model individuals in each evolution group is calculated. The fitness value is used to evaluate the specific logic of the quality of the vSVR model individuals: S31, based on the deviation between the longitude coordinate prediction value and the longitude coordinate, and the deviation between the latitude coordinate prediction value and the latitude coordinate, a distance deviation coefficient and a deviation extreme value coefficient are generated. The distance deviation coefficient is used to evaluate the accuracy of the vSVR model individual, and the deviation extreme value coefficient is used to evaluate the stability of the vSVR model individual. The calculation formula is as follows: Among them, x(j) represents the longitude coordinate corresponding to the longitude coordinate prediction value x'(j,k), y(j) represents the latitude coordinate corresponding to the latitude coordinate prediction value x'(j,k), JLpc(j,k) represents the distance deviation value between the predicted point and the actual coordinate point output by the kth vSVR model individual for the jth actual coordinate point, represents the distance deviation coefficient of the kth vSVR model individual, JLpc(j,k) max represents the maximum value of n distance deviation values ​​in the kth vSVR model individual, JLpc(j,k) min represents the minimum value of n distance deviation values ​​in the k-th vSVR model individual, PCjz(j,k) represents the deviation extreme value coefficient of the k-th vSVR model individual; S32, combining the environmental data and the distance deviation value in step S31, generating a precision ring ratio coefficient, which is used to evaluate the robustness of the vSVR model individual, and the calculation formula is as follows: Among them, v f (j), v l (j) H l (j) represents the wind speed, current speed and wave height at the same time as the longitude coordinate x(j), HJ(j) represents the environmental coefficient at the same time as the longitude coordinate x(j), λ1, λ2 and λ3 are all preset proportional coefficients, and on the basis of λ1+λ2+λ3=1, and 0<λ1<λ3<λ2<1, HJ(j1) represents the maximum value of the environmental coefficient among the n environmental coefficients, HJ(j2) represents the minimum value of the environmental coefficient among the n environmental coefficients, JLpc(j1,k) represents the distance deviation value between the predicted point and the actual coordinate point output by the kth vSVR model individual for the actual coordinate point at the moment of the maximum value of the environmental coefficient HJ(j1), JLpc(j2,k) represents the distance deviation value between the predicted point and the actual coordinate point output by the kth vSVR model individual for the actual coordinate point at the moment of the minimum value of the environmental coefficient HJ(j2), and JHB(k) represents the precision ring ratio coefficient of the kth vSVR model individual; S33, scoring the distance deviation coefficient, the deviation extreme value coefficient and the precision ring ratio coefficient according to the expert scoring method, generating an accuracy evaluation coefficient, a stability evaluation coefficient and a robustness evaluation coefficient between 0 and 1, and decreasing as the distance deviation coefficient, the deviation extreme value coefficient and the precision ring ratio coefficient increase, and determining the weights of the accuracy evaluation coefficient, the stability evaluation coefficient and the robustness evaluation coefficient in the fitness value calculation according to the expert scoring method; S34, combining the accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient, calculates the fitness value of the vSVR model individuals in each evolution group. The calculation formula is as follows: SYD(k)=ω1*JD(k)+ω2*WDX(k)+ω3*LBX(k) Among them, JD(k), WDX(k), and LBX(k) represent the accuracy evaluation coefficient, stability evaluation coefficient, and robustness evaluation coefficient of the kth vSVR model individual, SYD(k) represents the fitness value of the kth vSVR model individual, ω1, ω2, and ω3 are all weights determined by experts, and ω1+ω2+ω3=1.

5. The method for predicting a ship's navigation trajectory based on AIS according to claim 1, characterized in that: The criteria for determining the excellent vSVR model individuals are as follows: the fitness values ​​of the vSVR model individuals in all evolutionary groups are sorted from large to small, and the vSVR model individuals with the highest fitness values ​​are regarded as excellent vSVR model individuals.

6. The method for predicting a ship's navigation trajectory based on AIS according to claim 1, characterized in that: The specific logic of placing the newly generated vSVR model individual into the evolutionary group where its parent vSVR model individual is located is as follows: if a newly generated vSVR model individual is generated by two vSVR model individuals in the same evolutionary group, then this newly generated vSVR model individual is placed in the evolutionary group where the two vSVR model individuals coexist; if a newly generated vSVR model individual is generated by two vSVR model individuals in different evolutionary groups, then this newly generated vSVR model individual is copied and pasted and placed in the respective evolutionary groups of the two vSVR model individuals.

7. A ship navigation track prediction system based on AIS, used to execute the ship navigation track prediction method based on AIS according to any one of claims 1 to 6, characterized in that: include: A data processing module is used to obtain AIS data arranged in a continuous time series and environmental data aligned with the AIS data time series. After preprocessing the AIS data and environmental data, the AIS data is randomly divided into a training set and a test set. The preprocessing includes removing outliers, supplementing missing values, and normalizing. The grouping module is used to establish a vSVR model with parameters to be optimized. The parameters to be optimized by the vSVR model are used as genes and the initial population is randomly generated. The initial population consists of multiple vSVR model individuals with different genes. The vSVR model individuals are trained using the AIS data in the training set, and the individuals in the initial population are divided into multiple evolutionary groups according to the size of similarity. The fitness evaluation module is used to calculate the fitness value of the vSVR model individuals in each evolution group based on environmental data, the longitude coordinate prediction value and the deviation of the longitude coordinate, and the latitude coordinate prediction value and the deviation of the latitude coordinate. The fitness value is used to evaluate the quality of the vSVR model individuals; The independent evolution module is used to perform a round of independent evolution on each evolution group according to the standard operation of the genetic algorithm, and determine the excellent vSVR model individuals based on the fitness value of the vSVR model individuals after evolution, and transfer the excellent vSVR model individuals to other evolution groups; The overall evolution module is used to use the standard operation of the genetic algorithm to carry out a round of overall evolution on all the evolutionary groups after independent evolution, and put the newly generated vSVR model individuals into the evolutionary group where their parent vSVR model individuals are located; An iterative judgment module is used to repeat steps S3-S5 until the optimal vSVR model individual is found, or after reaching a predetermined number of evolution times, the vSVR model individual with the highest fitness value in all evolution groups is taken as the optimal vSVR model individual; A testing module is used to input the AIS data in the test set into the optimal vSVR model individual for testing, and use the optimal vSVR model individual as the navigation trajectory prediction model after the test is qualified, and return to step S2 when the test is unqualified to re-determine the value range of the parameters that need to be optimized in the vSVR model; The navigation track prediction module is used to pre-process the AIS data of the ship to be tested and input it into the navigation track prediction model to predict the navigation track of the ship to be tested.

Images