A ship trajectory prediction method based on multi-source heterogeneous ship shipping data

Abstract

The present invention discloses a ship trajectory prediction method based on multi-source heterogeneous ship shipping data, comprising: performing data integrity analysis, spatiotemporal trajectory synchronization, and ship trajectory extraction on the multi-source heterogeneous ship shipping data; constructing a three-level trajectory cache based on the ship trajectory state, and designing an incremental trajectory state mining algorithm to learn the prediction model feature parameters and realize the mining of ship motion laws; constructing a ship trajectory prediction algorithm using the mining laws, and outputting the predicted ship trajectory and probability. The present invention fully utilizes the complementary advantages of multi-source heterogeneous ship shipping data information, while taking into account the interpretability of the prediction results. By designing key technologies such as a trajectory extraction framework, an incremental trajectory state mining algorithm, and a trajectory prediction algorithm, the method realizes the prediction of ship trajectories in the medium and long term series in port waters, thereby providing early safety warnings to ships, guiding drivers to make safe driving behaviors such as obstacle avoidance in advance, and ensuring the navigation safety of ships.

Description

Technical Field

[0001] The present invention relates to the technical field of marine ship trajectory prediction, and in particular to a ship trajectory prediction method based on multi-source heterogeneous ship navigation data. Background Art

[0002] With the increasing degree of economic globalization and the frequent economic and trade exchanges between countries, the shipping industry is thriving. However, the increased load on sea lanes and port waters has led to problems such as ship traffic congestion and frequent ship collisions. Therefore, accurately obtaining real-time dynamic information about ship navigation, promptly detecting abnormal ship trajectories, and predicting ship navigation status in advance have become research hotspots for scholars at home and abroad.

[0003] Currently, long-short-term memory (LSTM) networks and other related neural network solutions can effectively predict the short-term (second-level) future motion state of a ship (latitude, longitude, speed, heading, etc.) by integrating traffic context features. However, the black-box nature of neural networks makes interpretation of prediction results difficult, reducing the practicality of the prediction models. Physical-based ship trajectory prediction solutions, on the other hand, face significant difficulties in encoding traffic environment features and are unable to better capture the environmental and interactive factors affecting ship motion. These methods struggle to guarantee accurate predictions for relatively medium- to long-term (tens of minutes or even longer) ship trajectories, and their practicality is also limited. Furthermore, most ship trajectory prediction studies focus on ordinary sea areas, with less research on ship trajectory prediction within busy port waters. Within port areas, ships frequently interact with each other. Due to complex port traffic conditions and ship scheduling, ship motion patterns within ports are no longer uniform, making trajectory prediction more challenging. Summary of the Invention

[0004] In view of this, the present invention provides a ship trajectory prediction method based on multi-source heterogeneous ship navigation data, aiming to establish a ship trajectory prediction method that can achieve relatively medium- and long-term results. By designing a trajectory extraction framework, a cache-based incremental trajectory state mining algorithm, and a trajectory prediction algorithm, on the one hand, the method has the characteristics of interpretability, and on the other hand, it does not require the introduction of complex expert knowledge to construct the ship kinematic equation. In addition, the method can also mine the motion pattern laws of the trajectory, learn the navigation strategy of the target ship, and mine the ship's motion pattern and driving habits, thereby improving the prediction accuracy and efficiency.

[0005] To this end, the present invention provides the following technical solutions:

[0006] The present invention provides a ship trajectory prediction method based on multi-source heterogeneous ship shipping data, comprising:

[0007] Acquire multi-source heterogeneous ship shipping data; and pre-process the multi-source heterogeneous ship shipping data to obtain ship trajectory status; the multi-source heterogeneous ship shipping data includes: ship automatic identification system, vector electronic nautical chart, hydrological and meteorological data, and ship registration data;

[0008] A three-level trajectory cache is constructed based on the ship trajectory state, and an incremental trajectory state mining algorithm is designed to learn hidden Markov model parameters and realize ship motion law mining; the incremental trajectory state mining algorithm includes: a state cache and a state mining function, the state mining function iterates and aggregates the initial state of the three-level trajectory cache according to the observed state and the hidden state, thereby constructing the state cache; if the caches of levels 1 to 3 are updated, the updated amount is iterated and aggregated, and the result is cached in the state cache as an increment, and the training cycle is repeated. If the cache is no longer updated, the parameters of the hidden Markov model after training are output, and then a hidden Markov model containing the ship motion law is obtained as a ship trajectory prediction model;

[0009] The ship trajectory prediction model is used to construct a ship trajectory prediction algorithm, and the predicted ship trajectory and probability are output.

[0010] Furthermore, the multi-source heterogeneous shipping data is pre-processed, including:

[0011] Performing data integrity analysis and UML modeling on the multi-source heterogeneous shipping data;

[0012] Synchronize the longitude, latitude, ground speed, heading, wind speed and current speed in the ship's shipping data in time and space;

[0013] Build a ship trajectory extraction framework to extract the ship trajectory status.

[0014] Furthermore, a parallel spatiotemporal trajectory synchronization method is used to synchronize the longitude, latitude, ground speed, heading, wind speed and current speed in the ship shipping data, including:

[0015] Divide the input synchronized trajectory into blocks;

[0016] Create a spatiotemporal trajectory synchronization thread through the interpolation thread pool;

[0017] Apply for several threads from the thread pool according to the current track block;

[0018] The threads are used to execute interpolation functions in parallel, and the interpolation results are inserted into a skip list, wherein the skip list is sorted according to the positioning time in the ship trajectory point, and finally the complete skip list node is output as the final interpolation result; wherein the interpolation function includes a cubic spline interpolation function for solving longitude and latitude interpolation and a linear interpolation function for solving ground speed, heading, wind, and current interpolation.

[0019] Furthermore, the trajectory extraction framework includes: pluggable trajectory partitioning rule functions, dynamic routing guards, and archiving trajectories;

[0020] The dynamic routing guard receives the input of the divided trajectory point, and then calls back the pluggable trajectory division rule function, which returns a Boolean value of true or false. Then, the route of the current trajectory point is determined according to the callback result:

[0021] If true, the divided track points will be archived according to the ship MMSI and positioning time in the divided track points to form an archived track, and the routing guard space of the current voyage track segment will be cleared;

[0022] If it is false, the route guard will continue to be executed for the current ship's voyage track.

[0023] Furthermore, the ship trajectory state includes an observed state and a hidden state; the observed state includes: the longitude, latitude, speed over ground, heading, wind speed and current speed in the ship trajectory; the hidden state includes: a set of historical trajectories of the target ship aggregated according to user-defined rules;

[0024] The three-level trajectory cache includes: using the trajectory code as the key and the archived trajectory value to construct the first-level cache L1TrainingSet; using the observation state code as the key and the hidden state code as the value to construct the second-level cache L2TrainingSet; using the trajectory code as the key and the hidden state code as the value to perform inverted trajectory, thereby constructing the third-level cache L3TrainingSet.

[0025] Furthermore, the hidden Markov model parameters include an initial state probability vector π, a state transition probability matrix A, and an observation probability matrix B.

[0026] Furthermore, during the ship trajectory prediction model training phase, the observed ships in the AIS are verified based on the ship file registration data, thereby filtering out illegal ships; the ship file registration data includes: ship MMSI, call sign, ship name, ship type and ship size.

[0027] Furthermore, during the ship trajectory prediction model training phase, the spatial latitude and longitude position information of isolated dangerous objects and obstacle targets is extracted based on the electronic chart vector data, thereby reducing the observation state space.

[0028] Furthermore, the ship trajectory prediction algorithm includes: a prediction function and a probability calculation function, the prediction function is used to solve the ship trajectory of the predicted ship within the future time range, and the probability calculation function is used to solve the probability of the predicted trajectory;

[0029] The prediction function takes the observed state sequence, the initial hidden state probability vector π, the hidden state transition probability matrix A, and the observed state probability matrix B in the predicted trajectory as inputs to the additive Viterbi algorithm, calculates the union of the algorithm prediction results for each step length until the final prediction step length is reached, and takes the final union as the final prediction result;

[0030] The probability calculation function takes the observation state sequence in the predicted ship trajectory as well as the initial hidden state probability vector π, the hidden state transition probability matrix A and the observation state probability matrix B as the input of the additive forward algorithm, and recursively solves the observation probability of the observation state at each time step until the final observation sequence step length is reached. The final probability calculation result is anti-logarithmized to obtain the final probability value.

[0031] As can be seen from the above technical solution, the present invention addresses the issues of medium- and long-term prediction accuracy of ship trajectories in port waters and the interpretability of prediction results. First, by fully leveraging the information correlation and complementarity between multi-source heterogeneous ship navigation data, the present invention explores the movement patterns of ships during port entry and exit. Furthermore, the model is continuously trained to learn the potential motion characteristics contained in navigation data over medium- and long-term timeframes (several minutes to tens of minutes), thereby enabling the prediction model to predict and analyze ship trajectories in complex port waters over medium- and long-term timeframes. Secondly, due to the transparency of the algorithm, the calculation process is traceable, avoiding the black box nature of the model, thus allowing for interpretation of the calculation results and improving the practicality of the prediction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0033] Figure 1 This is a flowchart of ship trajectory prediction based on multi-source heterogeneous ship navigation data in an embodiment of the present invention;

[0034] Figure 2 Flowchart of a method for predicting ship trajectory based on multi-source heterogeneous ship navigation data in an embodiment of the present invention;

[0035] Figure 3 This is an ER diagram of ship navigation data in an embodiment of the present invention;

[0036] Figure 4 This is a parallel spatiotemporal trajectory synchronization diagram in an embodiment of the present invention;

[0037] Figure 5 This is a trajectory extraction framework diagram in an embodiment of the present invention;

[0038] Figure 6 This is a trajectory state mining framework diagram in an embodiment of the present invention;

[0039] Figure 7 This is a ship trajectory prediction diagram based on the additive Viterbi algorithm in an embodiment of the present invention;

[0040] Figure 8 This is an example diagram of the ship trajectory prediction results in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0042] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0043] like Figure 1 and Figure 2 As shown, the ship trajectory prediction method based on multi-source heterogeneous ship navigation data according to an embodiment of the present invention is as follows:

[0044] S1. Acquire multi-source heterogeneous shipping data;

[0045] Among them, multi-source heterogeneous ship shipping data includes: ship automatic identification system, vector electronic nautical chart, hydrological and meteorological data and ship registration data;

[0046] S2. Integrity analysis and UML modeling of multi-source heterogeneous ship navigation data:

[0047] In order to fully understand the association and complementarity between AIS, hydrometeorological, vector electronic nautical charts and ship archive data, the embodiment of the present invention performs integrity analysis on the data from the perspectives of data mode, attribute characteristics and semantic relevance.

[0048] like Figure 3 As shown in the figure, there is an information correlation between the static information of the ship in the AIS message and the ship file registration information. The attribute MMSI associated with the two can be used to realize the data consistency analysis of the static information in the AIS message and the ship file registration information. Furthermore, through the correlation between the spatial information in the AIS dynamic information and the geometric spatial information in the electronic nautical chart, the spatial characteristics of the ship during the navigation process can be analyzed. Furthermore, through the correlation between the spatiotemporal information in the AIS dynamic information and the spatiotemporal information in the hydrological and meteorological information, the meteorological environment characteristics during the navigation process can be analyzed.

[0049] Based on the above analysis, we can further conclude that the entities in the above data, namely, AIS static information and AIS dynamic information use the message ID as the logical primary key, thereby uniquely identifying the AIS dynamic and static information; hydrological and meteorological information uses time and observation point as the logical primary key, thereby uniquely identifying the hydrological and meteorological information; feature objects and geometric objects in electronic nautical chart data use their object code as the logical primary key, thereby uniquely identifying the feature objects and geometric objects in the electronic nautical chart; ship file registration data uses the ship MMSI as the logical primary key, thereby uniquely identifying the ship file information.

[0050] Based on the above analysis, we can further establish relationships between these entities. Specifically, AIS dynamic and static information is linked to the logical primary key, the ship MMSI, in the ship profile information through the ship's MMSI logical foreign key. AIS dynamic and static information is linked to electronic chart geometric objects through longitude and latitude spatial information, while electronic chart feature objects are linked to electronic chart geometric objects through the logical foreign key of spatial feature codes. AIS dynamic information is linked to hydrological and meteorological information through spatiotemporal attributes. This enables the complete analysis and modeling of spatiotemporal information and ship feature information during a ship's voyage, thereby further facilitating the construction of predictive models.

[0051] S3. Use the parallel space-time trajectory synchronization method to synchronize the ship's trajectory in space and time:

[0052] In order to solve the problem of uneven intervals between ship trajectory points, a trajectory synchronization method is introduced in the embodiment of the present invention, which includes an interpolation thread pool and an interpolation function. Figure 4 As shown, first the input synchronized trajectory is divided into blocks, then the spatiotemporal trajectory synchronization thread is created through the interpolation thread pool, and then N is requested from the thread pool. segmentThe interpolation function is then executed in parallel. The interpolation function inserts the interpolation result into a skip list, where the skip list is sorted according to the positioning time of the ship trajectory point, so as to obtain the final time-ordered synchronous trajectory points after interpolation.

[0053] Among them, the size of the interpolation thread pool (N threads ) is determined by the number of processor cores (N cpu ), CPU utilization (U cpu ), trajectory block size (N segment ) is given, the specific relationship is as follows:

[0054] N threads = N cpu * U cpu *(1+ N segment ) (1)

[0055] Among them, the interpolation function includes cubic spline interpolation function (used to solve longitude and latitude interpolation) and linear interpolation (used to solve ground speed, heading, wind, and current interpolation);

[0056] The cubic spline interpolation method for longitude and latitude includes: assuming there are n ship track points, the track point sequence {p1, p2, ..., p n}, where p i ={lon i ,lat i ,SOG i ,COG i ,soundings i ,wind i ,...}(i=1,2,3,...,n), then the speed in the meridian direction is The latitude direction speed is Taking the ship's longitude sequence as an example, let the ship's longitude lon(t) be a function of time. When t∈[t i ,t j ], lon(t)=aΔt 3 +bΔt 2 +cΔt+d, taking the derivative of lon(t) we get:

[0057]

[0058] According to the trajectory sequence boundary values lon(t1),lon(t2),SOG lon (t1),SOG lon Substituting (t2) into equation (3) yields the spline function coefficients a, b, c, and d, thereby determining the function expression of longitude with respect to time. Furthermore, the filled longitude value can be obtained based on the interpolation time interval. The latitude value can be solved in the same way.

[0059] The linear interpolation method for ground speed, heading, wind and flow interpolation includes: taking wind speed as an example, assuming that wind speed (t) is a function of time, when t∈[t i ,t j ]hour:

[0060]

[0061] Thus, we can solve the interval [t i ,t j ]Inside p i and p j Interpolation of wind speed between two points. The same method is used to interpolate speed, heading and current.

[0062] S4. Design a cache-based ship trajectory extraction framework to extract ship trajectories;

[0063] like Figure 5 As shown in the figure, the ship trajectory extraction framework specifically includes: plug-in trajectory division rule function, dynamic routing guard and archive trajectory. First, the dynamic routing guard receives the input of the divided trajectory point: P MMSI , then call back the plug-in trajectory division rule function, the rule function returns a Boolean value (true or false), and then decides the route of the current trajectory point according to the callback result: if it is true, then according to P MMSI The relationship between the ship's MMSI and positioning time in P MMSI Archive to form an archive track and clear the routing guard space of the current voyage track segment; if it is false, continue to execute routing guard for the voyage track of the current ship.

[0064] S5. Define trajectory observation state, hidden state, and three-level cache:

[0065] Based on the Hidden Markov Model (HMM), the ship navigation observation state S is defined for the ship trajectory. obs and hidden state S hid , where S obs is the observable feature of the ship’s navigation data (longitude, latitude, speed over ground, heading, wind speed and current speed, etc.), S hid It is a set of historical trajectories of target ships aggregated according to user-defined rules.

[0066] The three-level trajectory cache includes: using the trajectory code as the key and the archived trajectory value to construct the first-level cache L1TrainingSet, further, using the observation state code as the key and the hidden state code as the value to construct the second-level cache L2TrainingSet, further, using the trajectory code as the key and the hidden state code as the value to perform inverted trajectory, thereby constructing the third-level cache L3TrainingSet;

[0067] S6. Based on the three-level cache mining state and state probability matrix, output the trained prediction model:

[0068] like Figure 6 As shown, first the state mining function is based on S obs and S hid Define the initial states of iterative and aggregated L1TrainingSet, L2TrainingSet, and L3TrainingSet to construct the state cache (state probability vector π, state transition probability matrix A, and observation probability matrix B). If the 1st to 3rd level caches are updated, the updated amount is iterated and aggregated, and the result is cached back into the state cache as an increment. The training cycle is repeated. If the cache is no longer updated, the final trained ship trajectory prediction model λ is output.

[0069] The calculation formula of the initial state probability matrix π is as shown in formula (5), where K represents the number of hidden states S, and N represents the size of the set in the first-level cache, that is, the number of original training trajectories of the target modeling ship. Then it represents the trajectory traj n The hidden state s contained in i The number of, and length(traj n ) represents the trajectory traj n The length of , that is, the number of nodes contained in the trajectory.

[0070]

[0071] The calculation formula of the hidden state transition probability matrix A is as shown in formula (6), where K represents the number of elements in the hidden state set S, and N represents the size of the first-level cache set, that is, the number of original training trajectory sets of the target modeling ship. Then it represents the trajectory traj n The hidden state s contained in i The number of It means that on the target trajectory traj n Middle S i The successor state is s j times.

[0072]

[0073] The calculation formula of the observation state probability matrix B is as shown in formula (7), where K represents the number of hidden states S, M represents the number of observation states O, and N represents the size of the first-level cache set, that is, the number of original training trajectory sets of the target modeling ship. Then it represents the trajectory traj n The hidden state s contained in i The number of It means that on the target trajectory traj n Middle S i The hidden state generates the observation state o j times.

[0074]

[0075] S7. Design a ship trajectory prediction algorithm based on the trained prediction model:

[0076] Considering that the state probability matrix contains a large number of floating-point numbers, it will cause intermediate data underflow during the calculation process, resulting in large calculation errors. At the same time, a large number of matrix multiplication operations will greatly consume computing resources. Therefore, the multiplication probability matrix operation is logarithmized to transform it into the addition probability matrix operation. Figure 7 As shown in the figure, the prediction model λ and the trajectory sequence to be predicted are used as the input of the prediction algorithm. The multiplication in the prediction process is logarithmized. Then, after initialization and recursion to obtain the maximum probability and reach termination, the optimal path, i.e. the predicted trajectory sequence, can be obtained by backtracking the variable Ψ.

[0077] S8. Use prediction algorithm to predict trajectory:

[0078] According to the prediction algorithm, the predicted trajectory is predicted step by step, and then the results of each step prediction are combined, and then the prediction model λ is used to calculate the union of each trajectory O in the union. t (t=1,2,...,T)Find the conditional probability P(O t |λ), and finally a predicted trajectory with probability can be obtained, where the probability expresses the possibility of the predicted trajectory occurring.

[0079] After prediction is performed using the method of this embodiment, the following can be obtained: Figure 8 The prediction effect shown in the figure is shown in Table 1. The further ship trajectory prediction experimental results are shown in Table 1, where the predicted trajectory is the trajectory with sequence number "⑩", "①, " is the trajectory number in the prediction result set, combined with Figure 8 It can be found that the trajectory The spatial operation behavior of trajectory ⑩ can be well predicted, and the turning maneuvering behavior and destination of the ship during the port entry process can be predicted. At the same time, the method of this embodiment finally gives the prediction probability, realizing the ship trajectory prediction within a longer time range (several minutes to tens of minutes) during the port entry process.

[0080] Table 1

[0081]

[0082]

[0083] Furthermore, in the prediction process, this embodiment can predict predicted trajectories of different lengths and make corresponding outputs. At the same time, the model can incrementally train trajectory sets of different sizes through three-level cache, and the model has a great advantage in algorithm robustness. Since the model has a time complexity of prediction algorithm: T(n) = O(L obserState *N hiddenState *N hiddenState ), where L obserState Represents the length of the predicted trajectory, N hiddenState represents the number of hidden states. It can be seen that the prediction time mainly depends on the number of hidden states. In the present invention, the number of hidden states is usually not greater than the number of observed states. The number of observed states depends on the feature selection and can be controlled within a constant range. Therefore, the overall model prediction time can be optimized within O(n 3 ) range. And the space complexity: S(n)=S initialData +S runtime , where S initialData represents the initial dataset size, which depends on the real environment, S runtime It represents the size of the space occupied by the algorithm during operation. Its size mainly depends on the 1-3 caches and the state cache (π, A and B). It depends on the size of the actual trajectory segment. It can be reasonably segmented according to the navigation behavior of the ship entering and leaving the port to reduce the size of the entire trajectory.

[0084] Furthermore, this embodiment outputs the prediction results in the form of multiple prediction trajectories in the final prediction result set, which can remain sensitive to potential dangerous situations, making it easier for supervisors to discover unknown traffic danger situations and reduce the occurrence of accidents to a greater extent. At the same time, it is more conducive to the navigation decision-making of unmanned systems, thereby better avoiding unknown navigation risks.

[0085] The embodiments of the present invention utilize multi-source heterogeneous ship navigation data, fully exploit information complementarity, and design a ship trajectory prediction method. This method not only enables medium- and long-term predictions with high accuracy, but also maintains interpretability of the prediction results, avoiding model black-boxing and improving the method's practicality. In particular, the present invention introduces caching to address the learning issues of the prediction model, supporting incremental learning of latent state features and avoiding cold-start model training. Furthermore, the cache's efficient I / O efficiency significantly improves learning efficiency. Combined with the additive prediction algorithm, the real-time trajectory prediction is ensured.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship trajectory prediction method based on multi-source heterogeneous ship shipping data, characterized in that: include: Acquire multi-source heterogeneous shipping data; and preprocessing the multi-source heterogeneous ship shipping data to obtain ship trajectory status; The multi-source heterogeneous ship shipping data includes: ship automatic identification system, vector electronic nautical chart, hydrological and meteorological data and ship registration data; A three-level trajectory cache is constructed based on the ship trajectory state, and an incremental trajectory state mining algorithm is designed to learn the hidden Markov model parameters and realize the mining of ship motion laws. The incremental trajectory state mining algorithm includes: a state cache and a state mining function. The state mining function iterates and aggregates the initial state of the three-level trajectory cache according to the observed state and the hidden state to construct the state cache. If the 1st to 3rd level caches are updated, the updated amount is iterated and aggregated, and the result is cached as an increment in the state cache. The training cycle is repeated. If the cache is no longer updated, the final trained hidden Markov model parameters are output, and then a hidden Markov model containing the ship motion law is obtained as a ship trajectory prediction model. Using the ship trajectory prediction model to construct a ship trajectory prediction algorithm, outputting the predicted ship trajectory and probability; The ship trajectory state includes an observed state and a hidden state. The observed state includes the longitude, latitude, speed over ground, heading, wind speed, and current speed of the ship trajectory. The hidden state includes the historical trajectory set of the target ship aggregated according to user-defined rules. The three-level trajectory cache includes: using the trajectory code as the key and the archived trajectory value to construct the first-level cache L1TrainingSet; using the observation state code as the key and the hidden state code as the value to construct the second-level cache L2TrainingSet; using the trajectory code as the key and the hidden state code as the value to perform inverted trajectory, thereby constructing the third-level cache L3TrainingSet.

2. The ship trajectory prediction method based on multi-source heterogeneous ship shipping data according to claim 1 is characterized in that: Preprocessing the multi-source heterogeneous shipping data includes: Performing data integrity analysis and UML modeling on the multi-source heterogeneous shipping data; Synchronize the longitude, latitude, ground speed, heading, wind speed and current speed in the ship's shipping data in time and space; Build a ship trajectory extraction framework to extract the ship trajectory status.

3. The ship trajectory prediction method based on multi-source heterogeneous ship shipping data according to claim 2 is characterized in that: The parallel spatiotemporal trajectory synchronization method is used to synchronize the longitude, latitude, ground speed, heading, wind speed and current speed in the ship shipping data, including: Divide the input synchronized trajectory into blocks; Create a spatiotemporal trajectory synchronization thread through the interpolation thread pool; Apply for several threads from the thread pool according to the current track block; The threads are used to execute interpolation functions in parallel, and the interpolation results are inserted into a skip list, wherein the skip list is sorted according to the positioning time in the ship trajectory point, and finally the complete skip list node is output as the final interpolation result; wherein the interpolation function includes a cubic spline interpolation function for solving longitude and latitude interpolation and a linear interpolation function for solving ground speed, heading, wind, and current interpolation.

4. The ship trajectory prediction method based on multi-source heterogeneous ship shipping data according to claim 2 is characterized in that: The trajectory extraction framework includes: pluggable trajectory partitioning rule functions, dynamic routing guards and archiving trajectories; The dynamic routing guard receives the input of the divided trajectory point, and then calls back the pluggable trajectory division rule function, which returns a Boolean value of true or false. Then, the route of the current trajectory point is determined according to the callback result: If true, the divided track points will be archived according to the ship MMSI and positioning time in the divided track points to form an archived track, and the routing guard space of the current voyage track segment will be cleared; If it is false, the route guard will continue to be executed for the current ship's voyage track.

5. The ship trajectory prediction method based on multi-source heterogeneous ship shipping data according to claim 1 is characterized in that: The hidden Markov model parameters include the initial state probability vector , state transition probability matrix A and observation probability matrix B.

6. The ship trajectory prediction method based on multi-source heterogeneous ship shipping data according to claim 1 is characterized in that: During the ship trajectory prediction model training phase, the observed ships in the AIS are verified based on the ship registration data to filter out illegal ships; The ship registration data includes: ship MMSI, call sign, ship name, ship type and ship size.

7. The ship trajectory prediction method based on multi-source heterogeneous ship shipping data according to claim 1 is characterized in that: During the ship trajectory prediction model training phase, the spatial latitude and longitude position information of isolated dangerous objects and obstacles is extracted based on the electronic chart vector data, thereby reducing the observation state space.

8. The ship trajectory prediction method based on multi-source heterogeneous ship shipping data according to claim 5 is characterized in that: The ship trajectory prediction algorithm includes: a prediction function and a probability calculation function, the prediction function is used to solve the ship trajectory of the predicted ship within the future time range, and the probability calculation function is used to solve the probability of the predicted trajectory; The prediction function will be predicted by the observation state sequence in the trajectory, the initial hidden state probability vector , the hidden state transition probability matrix A and the observed state probability matrix B are used as inputs of the additive Viterbi algorithm, and the union of the algorithm prediction results for each step is calculated until the final prediction step is reached, and the final union is used as the final prediction result; The probability calculation function will predict the observed state sequence in the ship trajectory and the initial hidden state probability vector , the hidden state transition probability matrix A and the observation state probability matrix B are used as the input of the additive forward algorithm. The observation probability of the observation state at each time step is solved recursively until the final observation sequence step length is reached. The final probability calculation result is anti-logarithmized to obtain the final probability value.

Images