Adaptive multi-ship encounter recognition method based on optimal encounter distance in complex waters
By fusing radar and Automatic Identification System (AIS) data and utilizing deep neural networks and adaptive density clustering algorithms, the problem of quantifying ship encounter situations in complex waters was solved, enabling precise monitoring and early warning of maritime traffic safety and avoiding collisions.
Patent Information
- Application Number
- CN202110989694.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-26
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-08-26
AI Technical Summary
Existing technologies are insufficient to accurately quantify the optimal encounter distance for ships in complex waters, resulting in low efficiency in maritime traffic safety supervision and an inability to effectively identify and warn of ship encounter behavior.
By fusing radar data and Automatic Identification System (AIS) data, a dynamic and static dataset of ship navigation behavior is constructed. A nonlinear mapping relationship is established using a deep neural network, and combined with an adaptive density clustering algorithm, real-time identification of multi-ship encounters and calculation of the optimal encounter distance are achieved.
Accurately identify vessel encounter behavior in complex waters, quantify the optimal encounter distance, improve the efficiency of maritime traffic safety supervision, avoid collisions, and ensure maritime traffic safety.
Smart Images

Figure CN113762367B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of marine intelligent transportation technology, and in particular to an adaptive multi-ship encounter recognition method based on an optimal encounter distance in complex waters. Background Art
[0002] Ship encounter is an important concept in marine traffic engineering. It is a necessary condition for ships to have a collision accident. When ships are sailing in complex waters at sea, if the distance between the ships is less than a certain threshold and both parties do not take correct measures to avoid collision, a collision accident will occur. By studying ship encounters, we can reflect the degree of danger of water traffic to a certain extent.
[0003] An accurate identification model for multiple ship encounters in complex waters is a prerequisite for the perception and early warning of collision risks between ships. At present, the identification models proposed in relative research all assume that all ships have the same encounter distance. However, in reality, whether a large number of ships sailing in complex waters have an effective encounter situation is affected by factors such as the dynamic and static characteristics of the ship itself, the navigation of the waters, the geographical environment, and the hydrological and meteorological factors. In addition, there are interactions between the factors themselves, which makes it difficult to quantify the optimal encounter distance for the ship encounter situation. Summary of the invention
[0004] In view of the problems existing in the prior art, the purpose of the present invention is to provide an adaptive multi-ship encounter identification method in complex waters based on the optimal encounter distance, which can accurately and intuitively identify ships that encounter in complex waters at sea, quantify the optimal encounter distance of ship encounter situations, provide maritime duty personnel with ship encounter and warning information, and improve the efficiency of maritime traffic safety supervision.
[0005] To achieve the above object, the technical solution adopted by the present invention is:
[0006] An adaptive multi-ship encounter recognition method based on optimal encounter distance in complex waters comprises the following steps:
[0007] S1: Based on the electronic nautical chart containing environmental information, the radar data and the ship automatic identification system data containing static and dynamic information of the ship are fused and calculated, and the dynamic and static data sets of the ship's navigation behavior are constructed;
[0008] S2: Analyze the influencing factors of ship encounters in complex waters, mine the characteristic values of ship collision avoidance actions from the dynamic and static data sets of ship navigation behaviors, and construct a data set of characteristic values of ship collision avoidance actions;
[0009] S3: A nonlinear mapping relationship between the ship encounter distance and its influencing factors is established through a deep neural network, and a calculation model for the optimal ship encounter distance under the influence of multiple coupling factors is constructed;
[0010] S4: Real-time identification of multiple ship encounters is performed through an adaptive density clustering algorithm. In this algorithm, the encounter distance of each ship under the influence of multiple factors is used as the distance threshold of density clustering. Ship clusters with effective encounter behaviors are extracted from the sailing ships in real time, and the ships sailing in complex waters are divided into multiple encounter ship subsystems composed of encounter ships.
[0011] Furthermore, step S1 specifically implements the following sub-steps:
[0012] S11: Spatiotemporal registration of AIS and radar data; converting AIS track data containing latitude and longitude information and radar track data containing distance and azimuth information into the same data format, and then transforming the two track data into the same rectangular reference system and aligning them in time;
[0013] S12: Correlate the two track data after time-space registration; transform the rectangular reference system into parameter space by Hough transform, and then calculate the correlation between the automatic identification system track data and the radar track data by fuzzy clustering and membership function method;
[0014] S13: The track data of the automatic identification system of ships and the radar track data after track association are fused by using the set evidence theory and the adaptive multi-factor track fusion algorithm to obtain high-precision ship track data and construct a dynamic and static data set of ship navigation behavior.
[0015] Furthermore, step S2 specifically implements the following sub-steps:
[0016] S21: Select the dynamic and static data sets of navigation behaviors in a specific water area within a period of time, use the set discrimination method to obtain the samples of ships that have an impact on each other and need to perform collision avoidance maneuvers in the water area, calculate the real-time distance and relative orientation between the two ships, and then discriminate and record the encounter situation based on the position and orientation relationship between the two ships;
[0017] S22: Tracking the collision avoidance process of the ship samples, and according to the change of the heading, identifying and mining the ship collision avoidance action characteristic values of the ship samples that perform the collision avoidance behavior; the ship collision avoidance action characteristic values include the distance between the give-way ship and the target ship when the give-way ship starts to take the collision avoidance action in various encounter situations, the avoidance angle taken by the give-way ship, and the distance between the give-way ship and the target ship when the two ships resume the original heading;
[0018] S23: By mining the samples of ships that have performed collision avoidance behaviors in multiple time windows in specific waters, a dataset of ship collision avoidance action feature values for different encounter ship combinations in various encounter situations is formed.
[0019] Furthermore, step S3 specifically implements the following sub-steps:
[0020] S31: Assume that the influencing factor of ship length on the encounter distance is L, the influencing factor of ship width is W, the influencing factor of navigation environment is E, the influencing factor of geographical environment is G, and the influencing factor of hydrological and meteorological is M. Then the encounter distance d of two ships that have an effective encounter behavior is a and d b As shown in the following formula,
[0021] d a =f(L a ,W a ,E a ,G a ,M a )
[0022] d b =f(L b ,W b ,E b ,G b ,M b )
[0023] Among them, L a , W a 、E a , G a and M a They represent the factors affecting the optimal encounter distance of one of the encountering ships, such as the length and width of the encountering ship, the navigation environment, the geographical environment and the hydrological and meteorological environment; L b , W b 、E b , G b and M b They represent the factors affecting the optimal encounter distance of another encountering ship, such as the length and width of the ship, the navigation environment, the geographical environment, and the hydrological and meteorological environment;
[0024] S32: Assuming that the distance at which the ship starts to take avoidance action is the optimal encounter distance, the length, width, navigation environment, geographical environment and hydrological and meteorological conditions are used as the input parameters of the deep neural network, and the optimal encounter distance is used as the output. By training on encounter scene samples, a calculation model for the optimal encounter distance of ships affected by multiple coupling factors is constructed.
[0025] Furthermore, step S4 specifically implements the following sub-steps:
[0026] S41: Based on the ship free encounter distance calculation model, when the distance between ships is less than their respective encounter distances, they are considered to be encountering ships; assuming that any two ships V o and V t The dynamic and static data set of ship navigation behavior includes the ship's real-time navigation longitude, latitude, ground speed, and ground heading, and is expressed as and According to the static and dynamic information of ships received by the ship automatic identification system, the Mercator algorithm calculation process of the distance between ships is as follows:
[0027] D λ =λ t -λ0
[0028]
[0029]
[0030] D MP =M P_t -M P_0
[0031] Bearing=arctan(D λ / D MP )
[0032]
[0033] Among them, D λ represents the longitude difference, Indicates the latitude difference, M p Denotes the latitude growth rate, D MP It represents the difference in latitude elongation rate between different latitudes, Bearing represents the direction, and D is the distance between the two ships. By calculating the distances between all ships, the distance matrix is obtained;
[0034] S42: Execute an adaptive density clustering algorithm, the input of which is a data set containing n ship track data, a distance matrix between ships, a field distance of each ship, and a density threshold for forming an encounter cluster, wherein the optimal encounter distance of ships is used as the field distance, and the minimum number of encounter ships is used as the density threshold for forming an encounter cluster; the output is the category label to which the encounter ship track data belongs.
[0035] After adopting the above scheme, the present invention can realize high-precision fusion of ship track data by fusing radar data and static and dynamic information of ships received by the ship automatic identification system, and then can mine the characteristic values of collision avoidance actions under various encounter situations from the historical data of specific waters, and then introduce a deep neural network to establish a nonlinear mapping between the ship encounter distance and its influencing factors, and construct a ship optimal encounter distance calculation model with multi-factor coupling influence. Finally, based on the adaptive density clustering algorithm, real-time identification of multiple ship encounters is realized, and ship clusters with effective encounter behaviors are extracted in real time from a large number of sailing ships. Ships sailing in complex waters are divided into multiple collision risk subsystems composed of encountering ships, providing a theoretical basis for maritime traffic safety monitoring.
[0036] The present invention can quantify the optimal encounter distance of ship encounter situations and accurately and intuitively identify ships that encounter in complex marine waters. It is of great significance for improving the monitoring and early warning of ship encounter and collision risks in the marine traffic monitoring system, avoiding ship collision accidents, and ensuring marine traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 This is a principle block diagram of step S1.
[0038] Figure 2a , Figure 2b and Figure 2c Schematic diagrams of the collision avoidance process of ship samples in chronological order.
[0039] Figure 3 Schematic diagram of different ship encounter distance ranges.
[0040] Figure 4 Schematic diagram of the calculation model of the optimal encounter distance of ships under the influence of multiple coupling factors.
[0041] Figure 5a and Figure 5b They are schematic diagrams of display of electronic nautical charts before and after real-time identification of multiple ship encounters. DETAILED DESCRIPTION
[0042] The present invention discloses an adaptive multi-ship encounter recognition method based on an optimal encounter distance in complex waters, which comprises the following steps:
[0043] Step S1: Based on the electronic chart containing environmental information, radar data and ship automatic identification system (AIS) data containing static and dynamic information of ships are fused, and high-precision ship track data can be obtained by using time-space registration, track association, and track fusion algorithms, thereby constructing a dynamic and static data set of ship navigation behavior. The electronic chart is an electronic chart of the marine traffic monitoring waters under the E-Navigation system. Figure 1 As shown, step S1 is specifically implemented as follows:
[0044] S11: Spatiotemporal registration of AIS and radar data. First, convert the AIS track data containing longitude and latitude information and the radar track data containing distance and azimuth information into the same data format, and then transform the two track data into the same rectangular reference system. Then further align the two sets of track data in time to ensure that the data points of AIS and radar are in the same spatiotemporal dimension. Only the space and time that have completed the spatiotemporal registration can proceed to the next step of track association.
[0045] S12: Correlate the two track data after time-space registration. In order to achieve accurate fusion of massive radar and automatic identification system data, Hough transform is used to transform the rectangular reference system into parameter space, and then the correlation between the automatic identification system track data and radar track data is calculated through fuzzy clustering and membership function method.
[0046] S13: The track data of the ship automatic identification system and the radar track data after track association are fused by using the collective evidence theory and the adaptive multi-factor track fusion algorithm to obtain high-precision ship track data and construct a dynamic and static data set of ship navigation behavior. The use of the collective evidence theory and the adaptive multi-factor track fusion algorithm can overcome the limitation that the weight value and threshold value in the existing fusion algorithm are obtained according to the empirical value, further improve the accuracy of the precise fusion of radar and ship automatic identification system data, and provide a reliable source of ship data for subsequent ship encounter identification.
[0047] Step S2: Analyze the influencing factors of ship encounters in complex waters, and use the set discriminant method to mine the ship collision avoidance action feature values including the distance from the target ship when the give-way ship takes collision avoidance actions in various encounter situations, the avoidance angle taken, and the actual distance passed by the two ships from the dynamic and static data sets of ship navigation behaviors, and construct a ship collision avoidance action feature value data set. The specific implementation is as follows:
[0048] S21: Figure 2a , Figure 2b and Figure 2c As shown in the figure, a dynamic and static data set of navigation behavior within a period of time and in a specific water area is selected, and the set discriminant method is used to obtain the samples of ships that have an impact on each other and need to perform collision avoidance maneuvers in the water area, and the real-time distance and relative orientation between the two ships are calculated. Then, according to the position and orientation relationship between the two ships, the encounter situation is judged and recorded, such as encounter, crossing and overtaking.
[0049] S22: Record the distance between the two ships when they start to turn sharply to the right, track the collision avoidance process of the ship samples, and identify and mine the ship collision avoidance action feature values of the ship samples that perform collision avoidance behavior according to the change in heading. The ship collision avoidance action feature values include the distance from the target ship when the give-way ship starts to take collision avoidance actions in various encounter situations, the avoidance angle taken by the give-way ship, the closest encounter distance, and the distance between the give-way ship and the target ship when they resume their original headings.
[0050] S23: By mining the samples of ships that have performed collision avoidance behaviors in multiple time windows in specific waters, a dataset of ship collision avoidance action feature values for different encounter ship combinations in various encounter situations is formed.
[0051] Step S3: A nonlinear mapping between the ship encounter distance and its influencing factors is established through a deep neural network, and a ship optimal encounter distance calculation model under the influence of multiple factors is constructed. Due to different dynamic and static characteristics and navigation environments, ships have different encounter distances, such as Figure 3 In the scenario shown, taking the encountering ships ShipA and ShipB as an example, the two ships are of different lengths, and the surrounding ships have different degrees of acceptance of the collision risk to themselves. Large ships are difficult to maneuver and require a larger safety margin, while small ships are better at maneuvering and can accept a smaller safety margin, resulting in different optimal encounter distances for different ships. Step S3 is specifically implemented as follows:
[0052] S31: Assume that the influencing factor of ship length on the encounter distance is L, the influencing factor of ship width is W, the influencing factor of navigation environment is E, the influencing factor of geographical environment is G, and the influencing factor of hydrological and meteorological is M. Then the encounter distance d of two ships that have an effective encounter behavior is a and d b As shown in the following formula,
[0053] d a =f(L a ,W a ,E a ,G a ,M a )
[0054] d b =f(L b ,W b ,E b ,G b ,M b )
[0055] Among them, L a , W a 、E a , G a and M a They represent the factors affecting the optimal encounter distance of one of the encountering ships, such as the length and width of the encountering ship, the navigation environment, the geographical environment and the hydrological and meteorological environment; L b , W b 、E b , G b and M b They represent the factors affecting the optimal encounter distance of another encountering ship, such as the length and width of the ship, the navigation environment, the geographical environment, and the hydrological and meteorological environment;
[0056] S32: Since the influence of ship characteristics, navigation environment, geographical factors and hydrological and meteorological factors on the optimal encounter distance of ships is nonlinear, it is difficult to express it with a data formula with clear physical meaning. Therefore, it is assumed that the distance at which the ship starts to take avoidance action is the optimal encounter distance. The characteristic value of the ship's collision avoidance behavior obtained through mining is a discrete combination of scenes and distances. Figure 4 As shown in the figure, in order to construct the optimal meeting distance model of ships under the influence of ship characteristics and environmental factors, a deep neural network is introduced. Based on the powerful nonlinear and adaptive self-organizing learning ability of the deep neural network, the length, width, navigation environment, geographical environment and hydrological and meteorological conditions are used as the input parameters of the deep neural network, and the optimal meeting distance is used as the output. By training the meeting scene samples, a calculation model for the optimal meeting distance of ships affected by multiple factors is constructed.
[0057] The trained deep neural network model is used to calculate the real-time ship encounter distance under subsequent scenarios and influencing factors, providing an accurate and effective quantitative basis for the subsequent real-time determination of multiple ship encounters in the waters.
[0058] Step S4: Real-time identification of multiple ship encounters is performed through an adaptive density clustering algorithm. In this algorithm, the encounter distance of each ship under the influence of multiple factors is used as the distance threshold of density clustering, and ship clusters with effective encounter behaviors are extracted in real time from a large number of sailing ships, and ships sailing in complex waters are divided into multiple encounter ship subsystems composed of encountering ships. Ship encounter is a prerequisite for the risk of collision. On the basis of obtaining the calculation model of the optimal encounter distance of ships, when the distance between ships is less than their respective encounter distances, they can be regarded as encountering ships. Therefore, the ship encounter identification modeling problem can be abstracted into a mathematical problem of adaptive density clustering of spatial data in data mining. Step S4 is specifically implemented as follows:
[0059] S41: Assume that there are any two ships V o and V t The dynamic and static data set of ship navigation behavior includes the ship's real-time navigation longitude, latitude, ground speed, and ground heading, and is expressed as and According to the static and dynamic information of ships received by the ship automatic identification system, the Mercator algorithm calculation process of the distance between ships is as follows:
[0060] D λ =λ t -λ0
[0061]
[0062]
[0063] D MP=M P_t -M P_0
[0064] Bearing=arctan(D λ / D MP )
[0065]
[0066] Among them, D λ represents the longitude difference, Indicates the latitude difference, M p Denotes the latitude growth rate, D MP represents the difference in latitude elongation rate between different latitudes, Bearing represents the direction, B in cos(B) is the abbreviation of Bearing, and D is the distance between the two ships. The distances between all ships are calculated by the above method to obtain the distance matrix;
[0067] S42: Execute the adaptive density clustering algorithm, with the input being a data set containing n ship track data, the distance matrix between ships, the field distance of each ship and the density threshold of the encounter cluster, wherein the optimal encounter distance of ships is used as the field distance (the field distance is the field radius), set to ε; the minimum number of encountering ships is used as the density threshold of the encounter cluster, set to M; the output is the category label to which the encounter ship track data belongs. The specific operations are as follows:
[0068] S4201: randomly select a track point p of a ship that is not a processed object in the distance matrix;
[0069] S4202: Mark p as the processed data 'visited' and continue processing in the ship distance matrix;
[0070] S4203: If there are at least M ship track objects in the ε-neighborhood of p;
[0071] S4204: Create a new meeting cluster C and add p to C;
[0072] S4205: Let N be the set of objects in the ε-neighborhood of p;
[0073] S4206: For each p' in N
[0074] S4207: If p' is unvisited,
[0075] S4208: mark p' as visited;
[0076] S4209: If there are at least M points in the ε-neighborhood of p', add these points to N,
[0077] S4210: If p' is not yet a member of any cluster, add p' to N,
[0078] S4211:End For
[0079] S4212: The output will encounter cluster C;
[0080] S4213: Else mark p as noise;
[0081] S4214: End until there is no object marked as unvisited.
[0082] The key of the present invention is that by fusing radar data and static and dynamic information of ships received by the ship automatic identification system, high-precision fusion of ship track data can be achieved, and then the characteristic values of collision avoidance actions under various encounter situations can be mined from the historical data of specific waters, and then a deep neural network is introduced to establish a nonlinear mapping between the ship encounter distance and its influencing factors, and a ship optimal encounter distance calculation model under the influence of multi-factor coupling is constructed. The encounter distance of each ship under the influence of multi-feature coupling is used as the distance threshold of density clustering, and finally, based on the adaptive density clustering algorithm, real-time recognition of multiple ship encounters is realized, and ship clusters with effective encounter behaviors are extracted in real time from a large number of sailing ships. Ships sailing in complex waters are divided into multiple collision risk subsystems composed of encountering ships, providing a theoretical basis for maritime traffic safety monitoring.
[0083] The present invention can quantify the optimal encounter distance of ship encounter situations and accurately and intuitively identify ships that encounter in complex marine waters. It is of great significance for improving the monitoring and early warning of ship encounter and collision risks in the marine traffic monitoring system, avoiding ship collision accidents, and ensuring marine traffic safety.
[0084] The above description is merely an embodiment of the present invention and does not limit the technical scope of the present invention. Therefore, any minor modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.
Claims
1. Adaptive multi-ship encounter recognition method based on optimal encounter distance in complex waters, It is characterized in that The following steps are involved: S1: Based on the electronic nautical chart containing environmental information, the radar data and the ship automatic identification system data containing static and dynamic information of the ship are fused and calculated, and the dynamic and static data sets of the ship's navigation behavior are constructed; S2: Analyze the influencing factors of ship encounters in complex waters, mine the characteristic values of ship collision avoidance actions from the dynamic and static data sets of ship navigation behaviors, and construct a data set of characteristic values of ship collision avoidance actions; S3: A nonlinear mapping relationship between the ship encounter distance and its influencing factors is established through a deep neural network, and a calculation model for the optimal ship encounter distance under the influence of multiple coupling factors is constructed; S4: Real-time identification of multiple ship encounters is performed through an adaptive density clustering algorithm. In this algorithm, the encounter distance of each ship under the influence of multiple factors is used as the distance threshold of density clustering, and the ship clusters with effective encounter behaviors are extracted from the sailing ships in real time, and the sailing ships in complex waters are divided into multiple encounter ship subsystems composed of encounter ships; Wherein, step S3 specifically implements the following sub-steps: S31: Assume that the influencing factor of ship length on the encounter distance is L, the influencing factor of ship width is W, the influencing factor of navigation environment is E, the influencing factor of geographical environment is G, and the influencing factor of hydrological and meteorological is M. Then the encounter distance d of two ships that have an effective encounter behavior is a and d b As shown in the following formula, d a =f(L a ,W a ,E a ,G a ,M a ) d b =f(L b ,W b ,E b ,G b ,M b ) Among them, L a , W a 、E a , G a and M a They represent the factors affecting the optimal encounter distance of one of the encountering ships, such as the length and width of the encountering ship, the navigation environment, the geographical environment and the hydrological and meteorological environment; L b , W b 、E b , G b and M b They represent the factors affecting the optimal encounter distance of another encountering ship, such as the length and width of the ship, the navigation environment, the geographical environment and the hydrological and meteorological environment; S32: Assuming that the distance at which the ship starts to take avoidance action is the optimal encounter distance, the length, width, navigation environment, geographical environment and hydrological and meteorological conditions are used as the input parameters of the deep neural network, and the optimal encounter distance is used as the output. By training on encounter scene samples, a calculation model for the optimal encounter distance of ships affected by multiple coupling factors is constructed.
2. The adaptive multi-ship encounter recognition method based on optimal encounter distance in complex waters according to claim 1, It is characterized in that The specific implementation of step S1 is as follows: S11: Spatiotemporal registration of AIS and radar data; converting AIS track data containing latitude and longitude information and radar track data containing distance and azimuth information into the same data format, and then transforming the two track data into the same rectangular reference system and aligning them in time; S12: Correlate the two track data after time-space registration; transform the rectangular reference system into parameter space by Hough transform, and then calculate the correlation between the automatic identification system track data and the radar track data by fuzzy clustering and membership function method; S13: The track data of the automatic identification system of ships and the radar track data after track association are fused by using the set evidence theory and the adaptive multi-factor track fusion algorithm to obtain high-precision ship track data and construct a dynamic and static data set of ship navigation behavior.
3. The adaptive multi-ship encounter recognition method based on optimal encounter distance in complex waters according to claim 1, It is characterized in that The specific implementation of step S2 is as follows: S21: Select the dynamic and static data sets of navigation behaviors in a specific water area within a period of time, use the set discrimination method to obtain the samples of ships that have an impact on each other and need to perform collision avoidance maneuvers in the water area, calculate the real-time distance and relative orientation between the two ships, and then discriminate and record the encounter situation based on the position and orientation relationship between the two ships; S22: Tracking the collision avoidance process of the ship samples, and according to the change of the heading, identifying and mining the ship collision avoidance action characteristic values of the ship samples that perform the collision avoidance behavior; the ship collision avoidance action characteristic values include the distance between the give-way ship and the target ship when the give-way ship starts to take the collision avoidance action in various encounter situations, the avoidance angle taken by the give-way ship, and the distance between the give-way ship and the target ship when the two ships resume the original heading; S23: By mining the samples of ships that have performed collision avoidance behaviors in multiple time windows in specific waters, a dataset of ship collision avoidance action feature values for different encounter ship combinations in various encounter situations is formed.
4. The adaptive multi-ship encounter recognition method based on optimal encounter distance in complex waters according to claim 1, It is characterized in that The specific implementation of step v4 is as follows: S41: Based on the ship free encounter distance calculation model, when the distance between ships is less than their respective encounter distances, they are considered to be encountering ships; assuming that any two ships V o and V t The dynamic and static data set of ship navigation behavior includes the ship's real-time navigation longitude, latitude, ground speed, and ground heading, and is expressed as and According to the static and dynamic information of ships received by the ship automatic identification system, the Mercator algorithm calculation process of the distance between ships is as follows: D λ =λ t -l 0 D MP =M P_t -M P_0 Bearing=arctan(D λ / D MP ) D=D φ / cos(B) Among them, D λ represents the longitude difference, Indicates the latitude difference, M p Denotes the latitude growth rate, D MP It represents the difference in latitude elongation rate between different latitudes, Bearing represents the direction, and D is the distance between the two ships. By calculating the distances between all ships, the distance matrix is obtained; S42: Execute an adaptive density clustering algorithm, the input of which is a data set containing n ship track data, a distance matrix between ships, a field distance of each ship, and a density threshold for forming an encounter cluster, wherein the optimal encounter distance of ships is used as the field distance, and the minimum number of encounter ships is used as the density threshold for forming an encounter cluster; the output is the category label to which the encounter ship track data belongs.
