Ship track prediction method and device

Abstract

The present invention discloses a ship track prediction method and device, which discretizes the track distribution range into a number of grid units based on the spatiotemporal distribution density of the ship track within the ship navigation area; obtains the grid unit corresponding to the current ship track node and its adjacent grid units; calculates the cost function of the ship position transfer between the current grid unit and the adjacent grid units respectively; and uses the adjacent grid unit with the smallest cost function as the predicted ship track of the ship position transfer. The solution of the present invention performs reliability assessment and repairs the original data of the training track to restore the real track before prediction, adds a timing factor constraint to the ship navigation network diagram, and takes into account the temporal correlation of the track data while depicting the spatial correlation of adjacent tracks. The size of the grid unit in the ship navigation network is flexibly set based on the spatiotemporal distribution density of the ship track.

Description

Technical Field

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

[0002] Ship track prediction is crucial for maritime traffic regulation, traffic safety risk warnings, business operations management, and environmental protection. Leveraging historical trajectory features for ship track prediction is a proven solution. However, these approaches focus on researching and optimizing prediction methods, while either omitting or oversimplifying the quality assessment and restoration of historical metadata. This creates a bottleneck in improving prediction accuracy.

[0003] Automatic Identification System (AIS) data is one of the most common data sources in the shipping industry. However, AIS data suffers from multiple sources of interference during measurement and transmission, resulting in low data accuracy and, therefore, inability to accurately locate vessels. When making predictions based on raw AIS data, data fluctuations and noise can lead to reduced prediction accuracy. Evaluating and repairing the training trajectory metadata before prediction to restore the true trajectory and achieve accurate, real-time trajectory prediction is a key challenge currently underway in ship track prediction. Summary of the Invention

[0004] The present invention provides a ship track prediction method and device, which can predict the ship track more accurately and quickly, and solve the problems encountered in track prediction such as low data accuracy, weak time correlation, and low prediction accuracy.

[0005] According to one aspect of the present invention, a ship track prediction method is provided, comprising the following steps:

[0006] Discretizing the track distribution range into a plurality of grid cells based on the spatiotemporal distribution density of the ship track within the ship navigation area; each ship track node is represented by a grid cell, and each grid cell includes at least one ship track node;

[0007] Get the grid cell corresponding to the current ship trajectory node and its adjacent grid cells;

[0008] Calculating the cost function of ship position transfer between the current grid unit and the adjacent grid units respectively; the cost function includes probability cost and time cost;

[0009] The adjacent grid cell with the smallest cost function is used as the predicted ship track of the ship position transfer.

[0010] The ship trajectory node is obtained in the following way:

[0011] Obtain ship tracks through the Automatic Identification System (AIS);

[0012] The ship track includes the ship's longitude, latitude, speed, heading and time; the ship track node is obtained by the following formula:

[0013] V={v1,v2,…v T};

[0014] Where i = 1, 2, ..., T is the time index of the ship's track in each navigation period; the state vector v i By latitude and longitude The speed (s) and heading (θ) are expressed as four dynamic variables, which are recorded as:

[0015]

[0016] in, is the latitude, λ i is the longitude, s is the speed, and θ is the heading angle;

[0017] The ship track is evaluated for reliability, abnormal points are eliminated, missing points are supplemented, and the ship track nodes of the ship track are obtained.

[0018] The reliability assessment of the ship track includes:

[0019] Anomaly assessment: including latitude and longitude anomaly assessment, speed anomaly assessment and heading anomaly assessment of the track;

[0020] Track missing assessment: Calculate the time difference between adjacent track points, take the statistical mode of the time difference in the historical track sequence, and use the mode of the time difference as the standard track time difference; determine the size of the AIS reporting time interval and the standard track time difference; if the AIS reporting time interval is greater than the size of the standard track time difference, it is determined that there is a point data missing between the two track points at this time, and the lost track point should appear at the time point where the sum of the previous moment and the standard track time difference is reached.

[0021] The step of removing abnormal points includes:

[0022] Based on real-time position updates, outliers in the ship trajectory time series are removed to obtain a smooth track. Specifically,

[0023] Calculate the five values ​​of longitude and latitude of the track points in the track trajectory respectively; the median M, upper and lower quartiles and upper and lower extreme values ​​of the five values;

[0024] If the longitude value and the latitude value are both greater than the lower limit value and less than the upper limit value, the trajectory point is a normal value; otherwise, the trajectory point is discarded.

[0025] The said supplementation of missing points includes:

[0026] For the missing track points, the cubic spline interpolation method is used to fit and estimate the missing points, including:

[0027] The time series interval [a,b] can be divided into n intervals, namely [(x0,x1), (x1,x2), ..., (x n -1,x n )], the function values ​​corresponding to the n+1 trajectory points are y0,y1,…,y n , if these points satisfy the interpolation function Si(x), then:

[0028] Si(x)=a i x 3 +b i x 2 +c i x+d i

[0029] Among them, a i 、b i 、c i d i is the coefficient to be determined;

[0030] The trajectory points are interpolated within the same time interval.

[0031] The acquisition of the cost function includes:

[0032] The edge weight from position j to position i will be denoted as P (j→i) ; For second-order transitions, each node is represented by n i|j , the probability of i to k is recorded as P (i|j→k) , the number of transfers is recorded as C i|j→k ;

[0033] Given an initial position The previous position is estimated using the haversine equation using the heading angle θ clockwise from north and the radius of the earth R.

[0034]

[0035]

[0036] δ is the angular distance obtained by d / R, d is the navigation distance, which is the distance between the centers of two grid cells;

[0037] When adjacent trajectory nodes occur in non-adjacent grid cells, linear interpolation is used to estimate intermediate transfers;

[0038] The ship position transfer probability can be calculated by the following formula:

[0039]

[0040] The cost function is obtained by calculating the following method:

[0041]

[0042] Among them, α is the weight coefficient used to weigh the probability cost and time cost, and its value range is [0,1]; d i，k is the distance between trajectory node i and trajectory node k, d max is the maximum distance between two grid cells, v i is the average velocity in grid cell i, v min is the minimum average velocity in all grid cells; 1-P(i|j→k) is the probability of an infeasible path.

[0043] When using the PNPoly algorithm to determine whether the acquired track node is located in the grid cell corresponding to the land block, the current adjacent grid cell is eliminated.

[0044] According to another aspect of the present invention, there is provided a ship track prediction device, comprising:

[0045] A grid unit division unit is used to discretize the track distribution range into a plurality of grid units based on the spatiotemporal distribution density of the ship track in the ship navigation area; each ship track node is represented by a grid unit, and each grid unit includes at least one ship track node;

[0046] A trajectory node acquisition unit is used to obtain the grid cell corresponding to the current ship trajectory node and its adjacent grid cells;

[0047] A cost function calculation unit, configured to calculate a cost function for ship position transfer between a current grid unit and an adjacent grid unit; the cost function includes a probability cost and a time cost;

[0048] The track prediction unit is used to take the adjacent grid unit with the smallest cost function as the predicted ship track of the ship position transfer.

[0049] A data reliability assessment unit is configured to assess the acquired AIS track data, including anomaly point assessment and track loss assessment. The anomaly point assessment includes latitude and longitude anomaly assessment, speed anomaly assessment, and heading anomaly assessment of the track. The track loss assessment includes calculating the time difference between adjacent track points, taking the statistical mode of the time difference values ​​in the historical track sequence, and using the mode of the time difference as the standard track time difference; determining the size of the AIS position reporting time interval and the standard track time difference; if the AIS position reporting time interval is greater than the standard track time difference, it is determined that point data is missing between the two track points at this time, and the missing track point should appear at a time equal to the sum of the previous moment and the standard track time difference.

[0050] The trajectory spatiotemporal repair unit is used to remove abnormal points in the trajectory data and fill in missing points. Specifically, it includes:

[0051] Based on real-time position updates, outliers in the ship trajectory time series are removed to obtain a smooth track;

[0052] For the missing track points, the cubic spline interpolation method is used to fit and estimate the missing points, and the track points are interpolated within the same time interval.

[0053] By adopting the technical solution of the present invention, a ship track prediction scheme is proposed, in which the track distribution range is discretized into a plurality of grid cells based on the spatiotemporal distribution density of the ship track within the ship navigation area; each ship track node is represented by a grid cell, and each of the grid cells includes at least one ship track node; the grid cell corresponding to the current ship track node and its adjacent grid cells are obtained; the cost function of the ship position transfer between the current grid cell and the adjacent grid cells is calculated respectively; the cost function includes a probability cost and a time cost; and the adjacent grid cell with the smallest cost function is used as the predicted ship track for the ship position transfer.

[0054] The proposed solution addresses the issue of low trajectory metadata quality by first conducting a reliability assessment of AIS data and then proposing trajectory data repair methods tailored to the characteristics of various issues. A graph-theory-based trajectory prediction modeling approach is proposed, which incorporates timing factor constraints in adjacent node transitions to address the issue of not incorporating temporal correlations in trajectory data into the prediction method. Furthermore, the proposed solution addresses the issue of grid sizing in ship navigation networks by flexibly adjusting the size of each grid based on the spatiotemporal distribution density of ship trajectories, avoiding performance issues associated with dense or sparse grids.

[0055] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0057] Figure 1 This is a flow chart of the ship track prediction principle in an embodiment of the present invention;

[0058] Figure 2 A probability-based ship navigation network diagram according to an embodiment of the present invention;

[0059] Figure 3 This is a diagram showing the effect of a ship collision warning based on track prediction in an embodiment of the present invention;

[0060] Figure 4 Schematic diagram of the structure of a ship track prediction device in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0062] AIS data transmits position reports at a constant frequency, so AIS tracks are discrete sets of points in time and space. Drawing on graph theory, we can abstract these spatiotemporal points of a ship into a network graph, where nodes represent track locations, adjacent nodes are connected by directed edges, and the current track node moves to the next location node with a certain probability. This gridding of track points can characterize the spatial correlation of adjacent tracks through a random process, but research on how to account for the temporal correlation of track points remains lacking. Furthermore, this gridding approach faces the problem of setting the grid size: a sparse grid can easily confuse irrelevant track information; a dense grid can easily fragment the temporal and spatial correlations of tracks and increase the computational effort.

[0063] This paper first assesses the reliability of AIS data and proposes trajectory data repair methods tailored to the characteristics of various issues, improving data accuracy. It also proposes a trajectory prediction method based on graph theory, incorporating a timing factor into the transitions between adjacent nodes to account for the temporal correlation of trajectory data. To address the grid sizing challenges faced by meshing in ship navigation networks, the paper flexibly sets the size of each grid based on the spatiotemporal distribution and temporal correlations of ship trajectories, avoiding the problem of dense or sparse grids that degrade the prediction model's performance.

[0064] Figure 1 FIG. 1 is a flow chart of ship track prediction in an embodiment of the present invention. Figure 1 As shown in FIG, the ship track prediction process includes the following steps:

[0065] Step 101: discretize the track distribution range into a number of grid cells based on the spatiotemporal distribution density of the ship track within the ship navigation area.

[0066] In the embodiment of the present invention, each ship trajectory node is represented by a grid unit, and each grid unit includes at least one ship trajectory node.

[0067] The track distribution range is discretized into multiple grid cells based on the spatiotemporal distribution density of the ship track in the ship navigation area. Each node in the ship navigation network diagram is represented by a cell (grid cell) in the grid. If the nodes are adjacent in space, they are regarded as adjacent grid cells. The track distribution range is represented by I, and the grid cell in I is m. Let N = {n1, n2, ..., n k} represents the adjacent grid cells of m, that is, the surrounding node cells that the predicted node m can reach. It is assumed here that each grid cell has eight possible navigation directions. In fact, each grid cell can have more neighbors, and the size of the grid cells does not need to be uniform. The grid cells are set based on the spatiotemporal distribution density of the ship trajectory. There will be at least one ship trajectory node in a cell. In subsequent calculations, all trajectory nodes in the cell are represented by one cell. For example, the distance between two cells is represented by the distance between the center points of their respective cells. The more grid cells there are, the higher the resolution of the grid, and the more accurate the predicted path should be. However, the more grid cells there are, the greater the amount of calculation.

[0068] Step 102: Obtain the grid cell corresponding to the current ship track node and its adjacent grid cells.

[0069] In the embodiment of the present invention, the current ship trajectory node and its temporally adjacent trajectory nodes form adjacent grid units. The ship trajectory nodes need to be acquired through AIS system data.

[0070] Ship tracks have unique characteristics. The spatial distribution of ship track data is relatively stable, and the navigation status of a ship within a certain area is closely related to the surrounding water environment, which means that ships sailing in the same waters often share common behaviors. In narrow waters, ships typically travel slower than in open waters. When a ship approaches a port, its speed may be very slow. Similarly, when a ship navigates a curved channel, it may frequently perform turning maneuvers. In addition, due to the limitations of a ship's maneuverability, a ship cannot make sharp turns like a pedestrian or robot. Based on the above ship navigation characteristics, it can be seen that the main determining indicators of a ship's navigation status are longitude, latitude, speed, heading, and time. Therefore, the ship's navigation state vector is composed of these five indicators.

[0071] The ship's track is denoted as V, which can be obtained by the corresponding state vector v in T time steps i To express it, write it as:

[0072] V={v1,v2,…v T}

[0073] Where i = 1, 2, ..., T is the time index of each ship track in each navigation period. The state vector v i By latitude and longitude The speed (s) and heading (θ) are expressed as four dynamic variables, which are recorded as:

[0074]

[0075] in, is the latitude, λ i is the longitude, s is the speed (unit can be knots / hour), and θ is the heading, in degrees clockwise from north.

[0076] There are two main problems with ship track prediction based on AIS data: measurement error and inconsistent navigation time intervals. The main causes of these two problems include abnormalities in the AIS built-in position sensor, abnormalities in the AIS signal link, problems with the AIS device itself, or signal delays caused by AIS channel obstruction. Measurement error refers to the fact that the AIS data value received during measurement is not the actual value collected by the sensor, and this deviation can sometimes be very large. Inconsistent navigation time intervals are mainly affected by various factors such as inconsistent AIS broadcast frequencies, packet loss, and channel obstruction. Therefore, to improve the accuracy of ship track prediction, it is crucial to conduct a reliability assessment of the raw AIS data. The quality of AIS data directly determines the accuracy of ship track prediction. AIS data reliability assessment includes outlier assessment and track point missing assessment.

[0077] Outlier Assessment

[0078] AIS track data anomalies may include longitude anomalies, latitude anomalies, heading anomalies, and speed anomalies. When evaluating each track for these types of anomalies, we first prioritize longitude and latitude anomalies. If an outlier value is detected in either longitude or latitude, the corresponding track point is deleted. Next, we evaluate speed anomalies. If the speed remains at zero for a certain period of time, the track point is deleted. Finally, we evaluate heading anomalies.

[0079] Longitude and latitude anomalies primarily include abnormal latitude and longitude data signs and sudden shift anomalies. Longitude data with signs other than E or latitude data with signs other than N are considered outliers and can be deleted. For sudden shift anomalies, data points outside the interval (QL - 1.5 hours, QU + 1.5 hours) are generally considered outliers (wild values).

[0080] Track point missing assessment

[0081] The navigation time series is extracted from the historical trajectory data set, and the time difference between adjacent trajectory points is calculated. The statistical mode of the time difference in the historical trajectory sequence is taken, and the mode of the time difference is used as the standard track time difference. The size of the AIS reporting time interval and the standard track time difference is judged. If the AIS reporting time interval is greater than the size of the standard track time difference, it is determined that there is a point data missing between the two track points at this time. The lost track point should appear at the time point of the sum of the previous moment and the standard track time difference.

[0082] AIS track temporal fixes.

[0083] Outlier removal method

[0084] Based on real-time position updates, outliers in the ship trajectory time series are removed to obtain smoothed trajectories. This outlier removal calculation utilizes a five-number summary method suitable for large sample testing. The "five numbers" are the median (M), the upper and lower quartiles (QU, QL), and the upper and lower extremes (QU+1.5H, QL-1.5H). Data outside the interval (QL-1.5H, QU+1.5H) are generally considered outliers (outliers).

[0085] The specific algorithm is as follows:

[0086] Parameters: Ship longitude and latitude collection

[0087] Functions: Minimum, Q1 (lower quartile), Median, Q3 (upper quartile), Maximum, IQR = Q3 - Q1, lower limit lower_limit = Q1 - 1.5 * IQR, upper limit upper_limit = Q3 + 1.5 * IQR.

[0088] Algorithm steps: 1. Calculate the five values ​​of longitude and latitude in the trajectory respectively; 2. If the longitude value and latitude value are both greater than the lower limit and less than the upper limit, the position is normal, otherwise the position is eliminated.

[0089] Cubic spline interpolation method

[0090] For missing track points, the cubic spline interpolation method is used to fit and estimate the missing points. The cubic spline difference is used to fit the time series data to keep the time interval of the data consistent. Compared with other interpolation methods, the cubic spline interpolation method can obtain a smoother series. The cubic spline interpolation method assumes that the first and second order derivatives of the curve are both zero as constraints of the cubic polynomial. Assume that the time series interval [a,b] can be divided into n intervals, that is, [(x0,x1), (x1,x2), ..., (x n -1,xn )], the function values ​​corresponding to the n+1 data points are y0,y1,…,y n , if these points satisfy the interpolation function Si(x), then:

[0091] Si(x)=a i x 3 +b i x 2 +c i x+d i

[0092] Among them, a i 、b i 、c i d i is an undetermined coefficient that can be obtained by evaluating adjacent data. Then, the AIS data are interpolated within the same time interval.

[0093] Step 103 : Calculate the cost function of the ship position transfer between the current grid cell and the adjacent grid cells respectively.

[0094] In the embodiment of the present invention, see Figure 2 A network is a graph consisting of a number of given points and lines connecting two points. This graph is usually used to describe a specific relationship between certain things. Nodes represent things, and the line connecting two nodes represents the quantitative relationship between the two things. For ship navigation, a ship sails from a position node at time t to the next position node at time t+1. This contains information about the heading, the position of the previous node, the sailing distance, and the sailing time. Therefore, the network graph of ship navigation is a directed weighted graph, which can be used to represent the spatial correlation and temporal correlation of the trajectory. The specific structure is shown in Figure 2 The ship travels from position node A to position node C, denoted as C|A. Node C|A has a certain probability of traveling to node D, denoted as P C|A→D , there is a certain probability that it will drive to node E, denoted as P C|A→E Similarly, a ship heading to node C has a certain probability of going to node A, denoted as C|A; and a certain probability of going to node B, denoted as C|B. Therefore, the weight of the edge connecting the nodes in the ship navigation graph is mainly determined by the transition probability between the nodes.

[0095] In most cases, the constructed ship navigation network diagram avoids land because ships do not travel over land. Therefore, cells surrounded by land should be zero density. However, due to AIS update errors, the ship navigation network diagram may sometimes contain track points within land. In addition, due to sparse AIS data in some areas, some routes may contain grids with zero density values. So zero-density cells should not be excluded from the track prediction algorithm. Land is regarded as a polygon. In order to determine whether a cell is within the land polygon, the PNPoly algorithm is used for calculation. If a cell is completely contained in land, it is marked as within land and is not considered as a neighboring cell of any other grid cell. In this way, cells within land will be excluded from the track prediction algorithm.

[0096] The plot avoidance algorithm is as follows:

[0097] Main parameters: time t, latitude and longitude of the ship to be measured at time t (x i ,y i ), polygon vertex D, the number of intersections between the ray passing through the point to be tested and the polygon edge,

[0098] Main function: the latitude and longitude of the ship to be measured at time t textp = (x i ,y i ), any set of fixed points areapoints(*), the vertices determine the edges of the polygon area lines, the maximum longitude and latitude value of the polygon area max(*), the minimum longitude and latitude value of the polygon area min(*),

[0099] Algorithm steps: 1. Determine whether the track point is outside the polygon area. If x i >x max ∩x i <x min ∩y i >y max ∩y i <y min, then textp is outside the quadrilateral circumscribing the polygon. 2. For each side in the set of polygon sides, if textp is between the two vertices of the side, substitute the longitude and latitude coordinate values of textp into the linear equation of the side \(y = lines(i,2)+(lines(i,4)-lines(i,2)) / (lines(i,3)-lines(i,1))*(testp(1)-lines(i,1))\). If \(y < testp(2)\), then the test point is not on the side, and at this time, increment the number of intersections of the point to be tested with the side by 1. If \(y = testp(2)\), then the test point is on the side and testp is inside the polygon. 3. For each side in the set of polygon sides, if the ray emitted by textp intersects with the polygon vertex, then update the number of intersections of the point to be tested with the side. Additionally, if textp coincides with the polygon vertex, then testp is inside the polygon. 4. If count is odd, then testp is inside the polygon; otherwise, testp is outside the polygon.

[0100] If only first - order transitions are considered, each node in the ship navigation network diagram will be directly mapped to a cell in the grid, and connection edges will be constructed between adjacent grids. The edge weight from position j to position i will be denoted as P (j→i) . For second - order transitions, each node in the graph is represented as n i|j , that is, there is a certain probability that position i is traveled from position j. At this time, the probability from i to k is denoted as P(i|j→k), and the number of transitions is denoted as C i|j→k . If there are two consecutive position information in adjacent grids, then j can be directly found. Otherwise, find j according to the course of the first information found in i. Given the initial position The course angle θ in the clockwise direction from the north and the Earth radius R, then the semi - versed sine equation can be used to estimate the previous position

[0101]

[0102]

[0103] δ is the angular distance obtained by d / R. d is the navigation distance, which is the distance between the centers of two grid cells.

[0104] If there are two consecutive position information in adjacent grids, then it is easy to obtain the number of transitions c between the two grids i|j→k . However, due to the non - fixed update frequency of the AIS transponder, the update may occur in non - adjacent grids. Therefore, when a non - adjacent grid update is observed, linear interpolation is used to estimate the intermediate transition.

[0105] The ship position transfer probability can be calculated by the following formula:

[0106]

[0107] Once the transition probability is determined, the edge weight cost can be calculated. The cost consists of two parts: probability cost and time cost. Probability cost is used to characterize the cost corresponding to the probability of a feasible path. Infeasible paths have higher costs. Time cost is used to penalize unnecessary paths. The cost is calculated as follows:

[0108]

[0109] Where α is a weight used to weigh the probability cost and time cost, and its value range is [0,1]. Giving a higher weight to the probability cost is necessary because in most cases, the physically shorter path may not be a feasible path. For example, a faster route may be dangerous and therefore infeasible, so α is usually set higher than 0.5; d i，k is the distance between node i and node k, d max is the maximum distance between two grid cells, v i is the average velocity within the i grid, v min is the minimum average speed in all grids. 1-P(i|j→k) is used to characterize the probability of an infeasible path. To determine the time cost, the time is calculated by dividing the distance by the speed. Since the cost here is a dimensionless value, the time cost also needs to be set to dimensionless, so by The maximum time value is obtained by taking the maximum distance d between two units. max and minimum speed v min Calculated, the minimum speed v min This is the minimum speed of the ship during cruising. Here the minimum speed can be set to 5 knots.

[0110] Step 104: taking the adjacent grid cell with the minimum cost function as the predicted ship track of the ship position transfer.

[0111] In the embodiment of the present invention, the navigation of a ship with an unknown destination is similar to a Markov random process. Here, the node with the highest weight among the adjacent nodes is used to determine the next node to travel to. The specific algorithm is described as follows:

[0112] Main parameters: initial node n0, prediction time T, current node n i , the next node n i+1 , the arrival time t of the next node.

[0113] Function: A(*) function is used to add the trajectory of the node. T(*) function is used to determine the travel time between two nodes.

[0114] Algorithm steps: assign the initial node to n i , n i Add to the ship track, when the current predicted duration is less than the total predicted duration T, perform four steps: 1. i The adjacent node with the minimum cost is assigned to the next node n i+1 ;2. Set the current node n i Go to the next node n i+1 The navigation time T is assigned to the next node; 3. The next node n i+1 Added to the ship trajectory, denoted as A(n i+1 ); 4. Update the next node to become the current node n i When the current prediction time exceeds the total prediction time T, the loop ends, the calculation ends, and a series of trajectory node data, that is, the trajectory of the ship, is obtained.

[0115] Figure 3 This is a diagram showing the effect of implementing a ship navigation collision warning method based on an embodiment of the present invention.

[0116] In the embodiments of the present invention, the main focus is on solving three major problems in ship track prediction: 1. Due to the existence of multi-source interference in the measurement and transmission process of AIS data, the accuracy of the data itself is not high. When making predictions based on the original AIS data, the fluctuation and noise of the data will lead to a decrease in prediction accuracy. The present invention proposes to conduct reliability assessment and repair the original data of the training trajectory to restore the real trajectory before prediction, which is a prerequisite for achieving accurate track prediction. 2. The present invention adds a timing factor constraint to the ship navigation network diagram, which can not only characterize the spatial correlation of adjacent trajectories, but also take into account the temporal correlation of track data, thereby improving the speed and accuracy of track prediction. 3. The present invention proposes to flexibly set the size of the grid unit in the ship navigation network based on the spatiotemporal distribution density of the ship trajectory, which can effectively avoid the problem of dense or sparse grids that reduce the performance of the prediction model.

[0117] In order to realize the above process, the technical solution of the present invention also provides a ship track prediction device, such as Figure 4 As shown, the ship track prediction device includes:

[0118] A grid unit division unit 21 is used to discretize the track distribution range into a plurality of grid units based on the spatiotemporal distribution density of the ship track within the ship navigation area; each ship track node is represented by a grid unit, and each grid unit includes at least one ship track node;

[0119] The track node acquisition unit 22 is used to acquire the grid cell corresponding to the current ship track node and its adjacent grid cells;

[0120] A cost function calculation unit 23 is used to calculate the cost function of the ship position transfer between the current grid unit and the adjacent grid units respectively; the cost function includes a probability cost and a time cost;

[0121] The track prediction unit 24 is configured to use the adjacent grid unit with the minimum cost function as the predicted ship track for the ship position transfer.

[0122] The device further comprises:

[0123] The data reliability evaluation unit 25 is used to evaluate the acquired AIS track data, including anomaly point evaluation and track loss evaluation. The anomaly point evaluation includes latitude and longitude anomaly evaluation, speed anomaly evaluation, and heading anomaly evaluation of the track. The track loss evaluation includes: calculating the time difference between adjacent track points, taking the statistical mode of the time difference values ​​in the historical track sequence, and using the mode of the time difference as the standard track time difference; determining the size of the AIS position reporting time interval and the standard track time difference; if the AIS position reporting time interval is greater than the standard track time difference, it is determined that point data is missing between the two track points at this time, and the missing track point should appear at the time point equal to the sum of the previous time and the standard track time difference.

[0124] The trajectory spatiotemporal restoration unit 26 is used to remove abnormal points in the trajectory data and fill in missing points. Specifically, it includes:

[0125] Based on real-time position updates, outliers in the ship trajectory time series are removed to obtain a smooth track;

[0126] For the missing track points, the cubic spline interpolation method is used to fit and estimate the missing points, and the track points are interpolated within the same time interval.

[0127] In summary, the technical solution of the present invention proposes a ship track prediction scheme, which discretizes the track distribution range into several grid units based on the spatiotemporal distribution density of the ship track within the ship navigation area; each ship track node is represented by a grid unit, and each of the grid units includes at least one ship track node; the grid unit corresponding to the current ship track node and its adjacent grid units are obtained; the cost function of the ship position transfer between the current grid unit and the adjacent grid units is calculated respectively; the cost function includes a probability cost and a time cost; and the adjacent grid unit with the smallest cost function is used as the predicted ship track of the ship position transfer.

[0128] The solution of the present invention performs reliability assessment and restoration of the original training trajectory data before prediction, which is a prerequisite for achieving accurate track prediction. Adding a timing factor constraint to the ship navigation network diagram not only depicts the spatial correlation of adjacent trajectories, but also takes into account the temporal correlation of the track data, improving the speed and accuracy of track prediction. Flexible setting of the grid cell size in the ship navigation network based on the spatiotemporal distribution density of ship trajectories effectively avoids the problem of dense or sparse grids that degrade the performance of the prediction model.

[0129] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0130] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0131] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0133] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A ship track prediction method, characterized in that: It includes the following steps: Based on the spatio-temporal distribution density of ship trajectories within the ship navigation area, discretize the track distribution range into several grid cells; each ship trajectory node is represented by a grid cell, and each of the grid cells includes at least one ship trajectory node; Obtain the grid cell corresponding to the current ship trajectory node and its adjacent grid cells; Calculate the cost functions of ship position transfer between the current grid cell and the adjacent grid cells respectively; the cost functions include probability cost and time cost; Take the adjacent grid cell with the minimum cost function as the predicted ship track of the ship position transfer; The method further includes a land parcel avoidance process: When using the PNPoly algorithm to determine that the obtained track trajectory node is located in the grid cell corresponding to the land parcel, eliminate the current adjacent grid cell. The specific process is as follows: The land parcel is regarded as a polygon; If a cell is completely contained in the land, it is marked as being within the land and will not be regarded as an adjacent cell of any other grid cell; The land parcel avoidance algorithm is as follows: Main parameters: time t, latitude and longitude of the ship to be tested at time t (x i ,y i ), polygon vertex D, the number of intersections between the ray passing through the point to be measured and the polygon edge count; Main function: the latitude and longitude of the ship to be measured at time t textp = (x i ,y i ), arbitrary fixed point set areapoints(*), vertices determine the edges of the polygon area lines, the maximum longitude and latitude value of the polygon area max(*), the minimum longitude and latitude value of the polygon area min(*); Determine whether the track point is outside the polygonal area. If x i >x max ∩x i <x min ∩y i >y max ∩y i <y min , then textp is outside the quadrilateral circumscribed by the polygon; For each side in the set of sides of the polygon, if textp is between the two vertices of the side, substitute the longitude and latitude coordinate values of textp into the linear equation of the side y = lines(i, 2)+(lines(i, 4)-lines(i, 2)) / (lines(i, 3)-lines(i, 1))*(testp(1)-lines(i, 1)); if y < testp(2), then textp is not on the side, and update the number of intersections of the point to be tested with the side by 1; if y = testp(2), then textp is on the side and textp is inside the polygon; For each side in the set of sides of the polygon, if the ray emitted by textp intersects with the polygon vertex, update the number of intersections of textp with the side; If textp coincides with the polygon vertex, then textp is inside the polygon; if count is odd, then textp is inside the polygon, otherwise, textp is outside the polygon.

2. A ship track prediction method according to claim 1, characterized in that: The ship trajectory node is obtained through the following method: Obtain the ship track through the Automatic Identification System (AIS) of the ship; The ship track includes the longitude, latitude, speed, course and time of the ship; the ship trajectory node is obtained through the following formula: V={v1,v2,…v T }; Where i = 1, 2, ..., T is the time index of the ship's track in each navigation period; the state vector v i By latitude and longitude It is represented by four dynamic variables: λ, speed s and heading θ, which can be expressed as: in, is the latitude, λ i is the longitude, s is the speed, and θ is the heading angle; Perform reliability evaluation on the ship track, eliminate abnormal points, and supplement missing points to obtain the ship trajectory nodes of the ship track.

3. A ship track prediction method according to claim 2, characterized in that: The reliability evaluation of the ship track includes: Abnormal point evaluation: including longitude and latitude abnormal evaluation, speed abnormal evaluation and course abnormal evaluation of the track; Track missing evaluation: calculate the time difference between adjacent track points, take the mode of the statistical values of the time differences in the historical track sequence as the standard track time difference; judge the size of the AIS reporting position time interval and the standard track time difference; if the AIS reporting position time interval is greater than the size of the standard track time difference, it is determined that there is missing position data between two track points at this time, and the missing track point should appear at the time point of the sum of the time of the previous moment and the standard track time difference.

4. A ship track prediction method according to claim 3, characterized in that: The abnormal point elimination includes: Based on real-time position updates, remove the outliers in the ship trajectory time series to obtain a smooth trajectory; specifically including: Calculate the five-number summary of the longitude and the five-number summary of the latitude of the trajectory points in the trajectory respectively; the five-number summary includes the median M, the upper and lower quartiles, and the upper and lower extremes. If both the longitude value and the latitude value are greater than the lower limit value and less than the upper limit value, then the trajectory point is a normal value, otherwise the trajectory point is eliminated.

5. A ship track prediction method according to claim 4, characterized in that: The missing point filling includes: For the missing trajectory points, use the cubic spline interpolation method to perform fitting and estimation of the missing points, specifically including: The time series interval [a,b] can be divided into n intervals, namely [(x0,x1), (x1,x2), ..., (x n -1,x n )], the function values ​​corresponding to the n+1 trajectory points are y0,y1,…,y n , if these points satisfy the interpolation function Si(x), then: Si(x)=a i x 3 +b i x 2 +c i x+d i Among them, a i 、b i 、c i d i is the coefficient to be determined; Interpolate the trajectory points at the same time interval.

6. A ship track prediction method according to claim 1, characterized in that: The acquisition of the cost function includes: The edge weight from position j to position i will be denoted as P (j→i) ; For second-order transitions, each node is represented by n i|j , the probability of i to k is recorded as P (i|j→k) , the number of transfers is recorded as C i|j→k ; Given an initial position The previous position is estimated using the haversine equation using the heading angle θ clockwise from north and the radius of the earth R. δ is the angular distance obtained by d / R, where d is the navigation distance, which is the distance between the centers of two grid cells. When adjacent trajectory nodes appear in non-adjacent grid cells, use linear interpolation to estimate the intermediate transition. The ship position transfer probability can be calculated by the following formula:

7. A ship track prediction method according to claim 6, characterized in that: The acquisition of the cost function is calculated by the following method: Among them, α is the weight coefficient used to weigh the probability cost and time cost, and its value range is [0,1]; d i，k is the distance between trajectory node i and trajectory node k, d max is the maximum distance between two grid cells, v i is the average velocity in grid cell i, v min is the minimum average velocity in all grid cells; 1-P(i|j→k) is the probability of an infeasible path.

8. A ship track prediction device, characterized in that: Including: A grid cell division unit for discretizing the track distribution range into several grid cells based on the spatio-temporal distribution density of the ship trajectory in the ship navigation area; each ship trajectory node is represented by a grid cell, and each of the grid cells includes at least one ship trajectory node. A trajectory node acquisition unit for acquiring the grid cell corresponding to the current ship trajectory node and its adjacent grid cells. A cost function calculation unit for calculating the cost function of the ship position transfer between the current grid cell and the adjacent grid cells respectively; the cost function includes a probability cost and a time cost. It also includes a land block avoidance process: When it is determined by the PNPoly algorithm that the obtained track trajectory node is located in the grid cell corresponding to the land block, eliminate the current adjacent grid cell. The specific process is as follows: The land block is regarded as a polygon. If a cell is completely contained in the land, it is marked as being inside the land and will not be regarded as an adjacent cell of any other grid cell. The land block avoidance algorithm is as follows: Main parameters: time t, latitude and longitude of the ship to be tested at time t (x i ,y i ), polygon vertex D, the number of intersections between the ray passing through the point to be measured and the polygon edge count; Main function: the latitude and longitude of the ship to be measured at time t textp = (x i ,y i ), arbitrary fixed point set areapoints(*), vertices determine the edges of the polygon area lines, the maximum longitude and latitude value of the polygon area max(*), the minimum longitude and latitude value of the polygon area min(*); Determine whether the track point is outside the polygonal area. If x i >x max ∩x i <x min ∩y i >y max ∩y i <y min , then textp is outside the quadrilateral circumscribed by the polygon; For each edge in the set of all sides of the polygon, if textp is between the two vertices of the edge, substitute the longitude and latitude coordinate values of textp into the linear equation of the edge y = lines(i,2)+(lines(i,4)-lines(i,2)) / (lines(i,3)-lines(i,1))*(testp(1)-lines(i,1)); if y < testp(2), then textp is not on the edge, and update the number of intersections of the test point and the edge by adding 1; if y = testp(2), then textp is on the edge and testp is inside the polygon. For each edge in the set of all sides of the polygon, if the ray emitted by textp intersects with the polygon vertex, update the number of intersections of textp and the edge. If textp coincides with the polygon vertex, then textp is inside the polygon; if count is odd, then textp is inside the polygon, otherwise, textp is outside the polygon. The track prediction unit is used to take the adjacent grid unit with the smallest cost function as the predicted ship track for ship position transfer.

9. A ship track prediction device according to claim 8, characterized in that: The device further comprises: A data reliability assessment unit is configured to assess the acquired AIS track data, including anomaly point assessment and track loss assessment. The anomaly point assessment includes latitude and longitude anomaly assessment, speed anomaly assessment, and heading anomaly assessment of the track. The track loss assessment includes calculating the time difference between adjacent track points, taking the statistical mode of the time difference values ​​in the historical track sequence, and using the mode of the time difference as the standard track time difference; determining the size of the AIS position reporting time interval and the standard track time difference; if the AIS position reporting time interval is greater than the standard track time difference, it is determined that point data is missing between the two track points at this time, and the missing track point should appear at a time equal to the sum of the previous moment and the standard track time difference. The trajectory spatiotemporal repair unit is used to remove abnormal points in the trajectory data and fill in missing points. Specifically, it includes: Based on real-time position updates, outliers in the ship trajectory time series are removed to obtain a smooth track. For the missing track points, the cubic spline interpolation method is used to fit and estimate the missing points, and the track points are interpolated within the same time interval.

Images