Abnormal behavior detection method, system and storage medium for inland waterway vessel navigation

Abstract

The present invention discloses a method, system, and storage medium for detecting abnormal navigation behavior of inland vessels, which are applied to the field of maritime intelligent supervision technology and can realize real-time detection of abnormal navigation behavior of ships at a microscopic scale, reduce the occurrence of ship navigation accidents, and improve the accuracy of detecting abnormal behavior of ships. The method includes: extracting ship trajectory motion behavior characteristics and preset interaction characteristics based on ship navigation trajectory data, fusing them through an attention mechanism autoencoder model to obtain intermediate layer information, inputting a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory cluster clusters; using a grid partitioning method to calculate the motion state probability density distribution data of each ship motion mode based on the trajectory cluster clusters; constructing a corresponding local autoencoder model based on the trajectory cluster clusters; performing motion pattern recognition through the local autoencoder model to obtain the current ship motion mode; and obtaining anomaly detection results based on the current ship trajectory data and the motion state probability density distribution data.

Description

Technical Field

[0001] The present invention relates to the field of maritime intelligent supervision technology, and in particular to a method, system and storage medium for detecting abnormal navigation behavior of inland vessels. Background Art

[0002] With the surge in the number of inland waterway vessels, accidents such as ship deviation and groundings have become increasingly frequent, posing an increasingly prominent challenge to ship safety supervision. Related technologies primarily employ a "human-monitored" supervision model. This involves monitoring key estuary areas through manual observation and expert discussion to assess the behavior of vessels in the waters, and communicating alerts to vessels that deviate from their routes or exhibit abnormal movements. However, this manual supervision approach lacks real-time performance and granularity. In busy waters, ship supervision is challenging, labor-intensive, and prone to missed or misjudgment. Summary of the Invention

[0003] In order to solve at least one of the above technical problems, the present invention proposes a method, system and storage medium for detecting abnormal navigation behavior of inland ships, which can realize real-time detection of abnormal navigation behavior of ships, reduce the occurrence of ship navigation accidents, and effectively improve the accuracy of abnormal behavior detection of ships.

[0004] In one aspect, an embodiment of the present invention provides a method for detecting abnormal navigation behavior of an inland waterway vessel, comprising the following steps:

[0005] Extracting ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters; wherein the preset interaction characteristics are interaction characteristics of the distance between the ship and the preset buoy of the waterway;

[0006] Build an attention mechanism autoencoding model;

[0007] The ship trajectory motion behavior characteristics and the preset interaction characteristics are fused through the attention mechanism autoencoder model to obtain intermediate layer information; wherein the intermediate layer information is trajectory feature information obtained by aggregating the ship trajectory motion behavior characteristics and the preset interaction characteristics;

[0008] Inputting the intermediate layer information into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters; wherein each trajectory cluster represents a ship motion mode;

[0009] According to the trajectory clusters, the motion state probability density distribution data of each ship motion mode is statistically analyzed by a grid division method;

[0010] Constructing a corresponding local autoencoder model according to the trajectory cluster; wherein the local autoencoder model corresponds to the ship motion mode;

[0011] Performing motion pattern recognition using the local autoencoder model based on the current trajectory data of the ship to obtain the current motion pattern of the ship; wherein the current trajectory data of the ship is the ship trajectory data acquired in real time;

[0012] An abnormal behavior analysis of the ship is performed based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current motion mode of the ship to obtain an abnormality detection result.

[0013] A method for detecting abnormal inland vessel navigation behavior according to an embodiment of the present invention has at least the following beneficial effects: First, the method extracts vessel trajectory motion behavior characteristics and preset interaction characteristics, namely, interaction characteristics of the distance between a vessel and a preset buoy in the waterway, from vessel navigation data in a predetermined water area. The method then fuses these characteristics with an attention-based autoencoder model to obtain intermediate-layer information, enabling the intermediate-layer information to reflect the differences and similarities between different vessel navigation trajectories. Next, the method inputs the intermediate-layer information into a clustering analysis algorithm to perform cluster analysis on the vessel trajectories, generating trajectory clusters. Each trajectory cluster corresponds to a vessel motion mode, thereby enabling identification of different vessel motion modes. Furthermore, the method uses a grid partitioning method to calculate the motion state probability density distribution data for each vessel motion mode based on the trajectory clusters. Simultaneously, local autoencoder models are constructed based on the trajectory clusters, corresponding to each vessel motion mode. The method then identifies the corresponding vessel motion mode by reconstructing the differences in the vessel navigation trajectories using different local autoencoder models. This enables dynamic, real-time vessel motion mode recognition, alleviating the problem of traditional trajectory clustering analysis and motion mode recognition requiring the input of all trajectory points. Furthermore, after the current ship motion mode is obtained through the local autoencoder model identification based on the current trajectory data of the ship, this embodiment performs an abnormal behavior analysis of the ship to obtain a detection result by comparing the current trajectory data of the ship with the motion state probability density distribution data corresponding to the current ship motion mode, thereby realizing real-time, micro-scale detection of abnormal ship navigation behavior, which is conducive to reducing the occurrence of ship navigation accidents and effectively improving the granularity of abnormal ship behavior detection.

[0014] According to some embodiments of the present invention, extracting ship trajectory motion behavior characteristics and preset interaction characteristics based on ship navigation trajectory data in a preset water area includes:

[0015] Construct a ship trajectory motion behavior feature extraction model and a preset interaction feature extraction model based on a time sliding window;

[0016] Extracting features from the ship's navigation trajectory data using the ship's trajectory motion behavior feature extraction model to obtain the ship's trajectory motion behavior features;

[0017] The preset interactive feature extraction model is used to extract features from the ship navigation track data and the waterway landmark data to obtain the preset interactive features.

[0018] According to some embodiments of the present invention, the attention mechanism autoencoder model includes a multidimensional measurement motion feature sequence attention mechanism submodule and a ship and environment interaction state attention submodule;

[0019] The ship trajectory motion behavior characteristics and the preset interaction features are fused through the attention mechanism autoencoder model to obtain intermediate layer information, including:

[0020] Determining a trajectory motion feature weight set by the multi-dimensional measurement motion feature sequence attention mechanism submodule;

[0021] Determining an interaction feature attention weight set by the ship and environment interaction state attention submodule;

[0022] The ship trajectory motion behavior characteristics and the preset interaction features are fused according to the trajectory motion feature weight set and the interaction feature attention weight set to obtain the intermediate layer information.

[0023] According to some embodiments of the present invention, inputting the intermediate layer information into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters includes:

[0024] The intermediate layer information is input into a k-means clustering algorithm to perform cluster analysis on the ship trajectory to obtain the trajectory clusters.

[0025] According to some embodiments of the present invention, the calculating of the motion state probability density distribution data of each ship motion mode by a grid partitioning method based on the trajectory clusters includes:

[0026] Dividing the waterway by a grid division method to obtain waterway grid data under each ship motion mode;

[0027] Gaussian probability density distribution data of the ship motion state in the grid under each ship motion mode is statistically calculated based on the channel grid data and the trajectory clusters.

[0028] According to some embodiments of the present invention, performing motion mode recognition using the local autoencoder model based on the current trajectory data of the ship to obtain the current motion mode of the ship includes:

[0029] Filling the current ship trajectory data through a masking mechanism to obtain expected ship trajectory data; wherein the expected ship trajectory data has the same data length as the ship navigation trajectory data in the preset waters;

[0030] Inputting the expected ship trajectory data into each of the local autoencoder models to obtain corresponding reconstruction errors;

[0031] According to some embodiments of the present invention, the abnormal behavior analysis of the ship is performed based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current ship motion mode to obtain the abnormality detection result, including:

[0032] Generate an abnormal alarm threshold value through Gaussian truncation probability according to the current trajectory motion mode of the ship and the corresponding motion state probability density distribution data; wherein the abnormal alarm threshold value includes an upper alarm level and a lower alarm level;

[0033] An abnormal behavior analysis is performed on the current trajectory data of the ship according to the abnormal alarm threshold to obtain the detection result.

[0034] On the other hand, an embodiment of the present invention further provides a system for detecting abnormal navigation behavior of inland waterway vessels, comprising:

[0035] A feature extraction module is used to extract ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters; wherein the preset interaction characteristics are interaction characteristics of the distance between the ship and the preset buoy of the channel;

[0036] The first model building module is used to build an attention mechanism autoencoding model;

[0037] a feature fusion module, configured to fuse the ship trajectory motion behavior features and the preset interaction features through the attention mechanism autoencoder model to obtain intermediate layer information; wherein the intermediate layer information is trajectory feature information obtained by aggregating the ship trajectory motion behavior features and the preset interaction features;

[0038] A cluster analysis module, configured to input the intermediate layer information into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters; wherein each trajectory cluster represents a ship motion mode;

[0039] A grid division module is used to calculate the motion state probability density distribution data of each ship motion mode by a grid division method according to the trajectory clusters;

[0040] A second model building module is used to build a corresponding local autoencoder model according to the trajectory cluster; wherein the local autoencoder model corresponds to the ship motion mode;

[0041] a motion pattern recognition module, configured to perform motion pattern recognition using the local autoencoder model based on the current trajectory data of the ship to obtain the current motion pattern of the ship; wherein the current trajectory data of the ship is the ship trajectory data acquired in real time;

[0042] The abnormal behavior analysis module is used to perform abnormal behavior analysis on the ship based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current motion mode of the ship to obtain an abnormality detection result.

[0043] On the other hand, an embodiment of the present invention further provides a system for detecting abnormal navigation behavior of inland waterway vessels, comprising:

[0044] at least one processor;

[0045] at least one memory for storing at least one program;

[0046] When the at least one program is executed by the at least one processor, the at least one processor implements the method for detecting abnormal navigation behavior of inland vessels as described in the above embodiment.

[0047] On the other hand, an embodiment of the present invention further provides a computer storage medium storing a program executable by a processor, wherein the program executable by the processor is used to implement the method for detecting abnormal navigation behavior of inland vessels as described in the above embodiment when executed by the processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a method for detecting abnormal navigation behavior of inland waterway vessels provided by an embodiment of the present invention;

[0049] Figure 2 This is a principle block diagram of the inland river vessel navigation abnormal behavior detection system provided by an embodiment of the present invention;

[0050] Figure 3 Schematic diagram of the structure of the attention mechanism autoencoder model provided by an embodiment of the present invention;

[0051] Figure 4 Schematic diagram of six types of ship motion modes provided in an embodiment of the present invention;

[0052] Figure 5 is a schematic diagram of a grid division method provided in an embodiment of the present invention;

[0053] Figure 6 1 is a schematic diagram of the difference in reconstruction error trends of the local autoencoder model provided in an embodiment of the present invention;

[0054] Figure 7 Schematic diagram of detection points of a real-time trajectory of a ship provided in an embodiment of the present invention;

[0055] Figure 8 2 is a schematic diagram of the real-time abnormal behavior detection results of a ship provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The embodiments described in the embodiments of this application should not be regarded as limitations of this application. All other embodiments obtained by ordinary technicians in this field without making any creative work are within the scope of protection of this application.

[0057] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0059] With the surge in the number of inland waterway vessels, accidents such as ship deviation and grounding have become more frequent, and the issue of ship safety supervision has become increasingly prominent. Among related technologies, the main "human monitoring" supervision model is adopted. During the supervision process in key estuary areas, the current behavior of ships in the water is analyzed through manual observation and expert discussion. Then, communication alerts are issued to ships that deviate from their routes or have abnormal movement status. However, this manual supervision method is not highly real-time and lacks granularity. In busy waters, manual ship supervision is difficult and labor-intensive, and is prone to missed and misjudgment.

[0060] An embodiment of the present invention provides a method, system and storage medium for detecting abnormal navigation behavior of inland waterways ships, which can realize real-time detection of abnormal navigation behavior of ships, reduce the occurrence of navigation accidents, and effectively improve the accuracy of abnormal behavior detection of ships. Figure 1 The method of the embodiment of the present invention includes but is not limited to step S110, step S120, step S130, step S140, step S150, step S160, step S170 and step S180.

[0061] Specifically, the application process of the method of the embodiment of the present invention includes but is not limited to the following steps:

[0062] S110: Extracting ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters, wherein the preset interaction characteristics are interaction characteristics of the distance between the ship and the preset buoy of the waterway.

[0063] S120: Construct an attention mechanism autoencoding model.

[0064] S130: The ship trajectory motion behavior characteristics and the preset interaction characteristics are integrated through the attention mechanism autoencoder model to obtain intermediate layer information. The intermediate layer information is the trajectory feature information obtained by aggregating the ship trajectory motion behavior characteristics and the preset interaction characteristics.

[0065] S140: Inputting the intermediate layer information into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters, wherein each trajectory cluster represents a ship motion mode.

[0066] S150: Counting the motion state probability density distribution data of each ship motion mode through a grid division method according to the trajectory clusters.

[0067] S160: Constructing a corresponding local autoencoder model based on the trajectory clusters, wherein the local autoencoder model corresponds to the ship motion mode.

[0068] S170: Perform motion pattern recognition using a local autoencoder model based on the current ship trajectory data to obtain the current ship motion pattern. The current ship trajectory data is the ship trajectory data acquired in real time.

[0069] S180: Analyze abnormal behavior of the ship based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current motion mode of the ship to obtain an abnormality detection result.

[0070] During the operation of this specific embodiment, this embodiment first extracts the ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters. Specifically, the preset interaction characteristics are the interaction characteristics of the distance between the ship and the preset buoy of the channel. Among them, the preset buoy of the channel is the key buoy set in the channel. This embodiment obtains the ship navigation trajectory data in the preset waters, and the preset waters can be any waters where abnormal ship behavior detection is required. This embodiment extracts the ship's motion behavior characteristics, i.e., the ship trajectory motion behavior characteristics, and the interaction characteristics of the distance between the ship and the preset buoy of the channel, i.e., the preset interaction characteristics, based on the obtained ship navigation trajectory data in the preset waters. Among them, the interaction characteristics of the distance between the ship and the preset buoy of the channel are represented by the distance change characteristics between the ship and each key buoy of the channel within the jurisdiction of the navigation area, i.e., the preset buoy of the channel. Further, this embodiment constructs an attention mechanism self-encoding model. At the same time, this embodiment fuses the ship trajectory motion behavior characteristics and the preset interaction characteristics through the attention mechanism self-encoding model to obtain intermediate layer information. Specifically, this embodiment reconstructs the motion behavior characteristics and preset interaction features of ship trajectories by establishing an attention mechanism autoencoder model, and uses the intermediate layer information of the attention mechanism autoencoder model to represent the trajectory features. The intermediate layer information can serve as a vector representation of the deep encoding of the ship trajectory, because the reconstructed sequence of the ship trajectory can be reconstructed by the encoder based on the intermediate layer information. The intermediate layer information reflects the differences and similarities between different ship trajectories and contains all the trajectory information. Next, this embodiment inputs the intermediate layer information into a clustering analysis algorithm to perform cluster analysis on the ship trajectories, obtaining trajectory clusters. In this embodiment, the intermediate layer information of the historical ship trajectories (i.e., the navigation trajectory data of ships in the preset waters) is extracted as input to the clustering analysis algorithm, and the clustering analysis algorithm is used to perform cluster analysis on the ship trajectories to obtain trajectory clusters. Each trajectory cluster represents a ship motion mode. It should be noted that in this embodiment, the ship motion mode is a collection of different ship navigation trajectories. Most ships navigate according to a corresponding motion mode, and the corresponding ship motion mode is obtained through cluster analysis. Furthermore, this embodiment uses a grid partitioning method to calculate the motion state probability density distribution data of each ship motion mode based on the trajectory clusters. This embodiment first performs grid division, and then collects statistics on the probability density distribution data of trajectory clusters in each grid under different ship motion modes.

[0071] To identify the motion patterns of real-time ship trajectories, this embodiment constructs a corresponding local autoencoder model based on trajectory clusters. This local autoencoder model shares the same structure as the attention mechanism autoencoder model. The complete ship trajectory under the corresponding motion pattern serves as training data for the local autoencoder model, enabling a more accurate reconstruction of the ship trajectory under each motion pattern.

[0072] It is easy to understand that during actual navigation, the real-time navigation trajectory of a ship within the waterway gradually increases, and the shape and motion characteristics of this trajectory gradually approach the historical trajectory cluster corresponding to its inherent motion pattern, that is, the trajectory cluster. Therefore, the reconstruction of the local autoencoder model corresponding to the inherent motion pattern of the ship will gradually decrease, that is, the reconstruction error of the local autoencoder model corresponding to the inherent motion pattern of the real-time acquired ship trajectory data will gradually decrease, while the reconstruction error of other local autoencoder models will gradually increase or fluctuate. This embodiment uses the method of rolling calculation of the trajectory reconstruction mechanism error to more accurately realize the automatic identification of the ship's motion pattern by utilizing the differences in the trajectory reconstruction error trend. Furthermore, this embodiment obtains the ship's navigation trajectory data in real time, that is, the current ship trajectory data of the ship, and after identifying the current ship motion mode based on the current ship trajectory data, analyzes and compares the current ship motion mode with the motion state probability density distribution data corresponding to the motion mode to obtain grid-level anomaly detection results, realizing real-time detection of abnormal ship navigation behavior, reducing the occurrence of ship navigation accidents, and effectively improving the granularity of ship abnormal behavior detection.

[0073] In some embodiments of the present invention, the ship trajectory motion behavior characteristics and preset interaction characteristics are extracted based on the ship navigation trajectory data in the preset waters, including but not limited to:

[0074] A ship trajectory motion behavior feature extraction model and a preset interaction feature extraction model are constructed based on the time sliding window.

[0075] The ship trajectory motion behavior feature extraction model is used to extract features from the ship navigation trajectory data and channel landmark data to obtain the ship trajectory motion behavior features.

[0076] The preset interactive feature extraction model is used to extract features from the ship's navigation trajectory data to obtain preset interactive features.

[0077] In this specific embodiment, this embodiment first constructs a ship trajectory motion behavior feature extraction model and a preset interaction feature extraction model based on a time sliding window. Among them, the preset interaction feature extraction model is the interaction feature of the distance between the ship and the preset buoy of the channel. Then, the ship trajectory motion behavior feature extraction model is used to extract the characteristics of the ship's navigation trajectory data to obtain the ship trajectory motion behavior characteristics, and the preset interaction feature extraction model is used to extract the characteristics of the ship's navigation trajectory data and the channel landmark data to obtain the preset interaction features. Specifically, the ship trajectory dataset Tr={Tr i |i=1,2,...,n}, where Tr i ={P t |t=1,2,...m}, t is the timestamp, P t={(lat, lon, sog, cog)}, where lat represents latitude, lon represents longitude, sog represents speed, and cog represents heading. Set the time sliding window to c and the step size to s, where s < c, Tr i The trajectory segment behavior feature sequence under each sliding window is composed as follows:

[0078]

[0079] Where r represents the number of trajectory segments, represents the average velocity of the trajectory points in the sliding window, represents the upper quartile of speed, represents the lower quartile value of the speed, Δx1 represents the latitude change of the trajectory point in the sliding window, and Δy1 represents the longitude change. Represents the latitude coordinate of the midpoint of the trajectory point in the sliding window, Represents the midpoint longitude coordinate. Using the mean and quartile values of motion features within each window can effectively reduce feature extraction errors caused by uneven point density. This method also eliminates the need for trajectory interpolation, reducing model computational costs.

[0080] At the same time, the distance change between the ship and multiple key buoys of the channel within the jurisdiction of the navigation area, that is, the preset buoys of the channel, is a temporal and spatial evolution process. o , o=1,2,...,g, the distance between the midpoint of the ship in each time window and all the preset buoys of the channel is used to dynamically express the interaction state between the ship and the environment, as shown in the following formula (2):

[0081]

[0082] Among them, r represents the number of sliding windows, Indicates the midpoint of trajectory i in the first sliding window and the preset buoy a of the channel o The Euclidean distance of .

[0083] In some embodiments of the present invention, the attention mechanism autoencoder model includes a multi-dimensional motion feature sequence attention mechanism submodule and a ship-environment interaction state attention submodule. The attention mechanism autoencoder model fuses the ship trajectory motion behavior characteristics and the preset interaction features to obtain intermediate layer information, including but not limited to:

[0084] The trajectory motion feature weight set is determined by the multi-dimensional measurement motion feature sequence attention mechanism sub-module.

[0085] The interaction feature attention weight set is determined through the ship and environment interaction state attention submodule.

[0086] According to the trajectory motion feature weight set and the interaction feature attention weight set, the ship trajectory motion behavior characteristics and the preset interaction features are fused to obtain the intermediate layer information.

[0087] In this specific embodiment, the attention mechanism autoencoder model constructed in this embodiment includes a multi-dimensional measurement motion feature sequence attention mechanism submodule and a ship and environment interaction state attention submodule. Specifically, this embodiment determines the trajectory motion feature weight set through the multi-dimensional measurement motion feature sequence attention mechanism submodule, and determines the interaction feature attention weight set through the ship and environment interaction state attention submodule. Then, according to the trajectory motion feature weight set and the interaction feature attention weight set, the ship trajectory motion behavior characteristics and the preset interaction features are fused to obtain the intermediate layer information. For example, referring to Figure 3 In this embodiment, the multidimensional motion feature sequence attention mechanism submodule (attention module 1) converts r window trajectory segments into a weight set of length r and value 0-1 through the Dense layer and the softmax layer. The model learns parameters through the Adam optimizer to determine the multidimensional motion feature sequence weight, that is, the trajectory motion feature weight set, as shown in the following formula (3):

[0088]

[0089] Among them, Representative trajectory characteristics (ship trajectory motion behavior characteristics) E i The set of weights for all windows.

[0090] In addition, the construction method of the ship-environment interaction state attention submodule (attention module 2) in this embodiment is the same as the construction method of the multidimensional measurement motion feature sequence attention mechanism submodule. The obtained ship-environment interaction state attention weight set, that is, the interaction feature attention weight set, is shown in the following formula (4):

[0091]

[0092] Among them, Represents the preset interaction feature P i The set of weights for all windows.

[0093] Furthermore, this embodiment integrates the preset interactive features and the ship trajectory motion behavior features, and splices the trajectory motion feature weight set with the interactive feature attention weight set to obtain the intermediate layer information. Among them, this embodiment uses the obtained intermediate layer information as the input of the model decoder, and the decoder adopts a single-layer long short-term memory (LSTM) module. At the same time, in this embodiment, the number of input units r is 30, and the length of the intermediate layer unit is h = 150, that is, the size of the high-dimensional feature vector of the trajectory is 150, and an attention mechanism trajectory self-encoding model is constructed. The encoder adopts the Bilstm module and the decoder adopts a single-layer LSTM module. The decoder can restore the reconstructed features with the same size as the original input according to the connection of the high-dimensional feature vector. The model is based on the loss function Loss and is optimized accordingly using the Adam optimizer, as shown in the following formula (5):

[0094]

[0095] Among them, E i 、 The distribution represents the motion characteristics of the actual trajectory segments and the characteristics of the reconstructed trajectory segments, and the loss is the trajectory reconstruction error.

[0096] In some embodiments of the present invention, the intermediate layer information is input into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters, including but not limited to:

[0097] The intermediate layer information is input into the k-means clustering algorithm to perform cluster analysis on the ship trajectories and obtain trajectory clusters.

[0098] In this specific embodiment, the present embodiment performs cluster analysis on the ship trajectory using the K-means clustering algorithm. Specifically, the present embodiment first obtains the intermediate layer high-dimensional feature vector of the trained attention mechanism autoencoder model, that is, the intermediate layer information. Then, the present embodiment uses the intermediate layer information as the extracted features of the trajectory and uses it as the input of the K-means clustering analysis to obtain the trajectory clusters, thereby determining the ship motion pattern. The clustering analysis process is shown in the following formula (6):

[0099] L n =Kmeans(H|H={h′1,h′2,...,h′ n}) (6)

[0100] Among them, L n is the set of ship motion mode category labels of n trajectories, and h is the intermediate layer information. For example, the cluster analysis results are as follows: Figure 4As shown, in this embodiment, A, B, C, and D are used to represent key points in the detection water area, or four estuaries. There are 6 types of movement behaviors in this water area, namely from A to B, from B to A, from A to C, from C to A, from D to B, and from B to D. The ship trajectories in the detection water area are clustered and analyzed by the k-means clustering algorithm, and 6 clusters are obtained, which correspond to the 6 movement patterns of the preset water area.

[0101] In some embodiments of the present invention, the motion state probability density distribution data of each ship motion mode is statistically analyzed based on the trajectory clusters by grid division method, including but not limited to:

[0102] The waterway is divided by the grid division method to obtain the waterway grid data under each ship motion mode.

[0103] According to the channel grid data and trajectory clustering, the Gaussian probability density distribution data of the ship motion state in each ship motion mode within the grid is statistically analyzed.

[0104] In this specific embodiment, this embodiment first divides the waterway into grids, and then calculates the Gaussian probability density distribution data of the ship motion state under each ship motion mode in the grid based on the obtained waterway grid data and trajectory clustering clusters. Specifically, this embodiment first divides the waterway of the preset water area by the grid division method to obtain the waterway grid data. Then, this embodiment calculates the Gaussian probability density distribution of each type of ship motion state in the grid. This embodiment calculates the corresponding Gaussian probability density distribution data based on the distribution of each trajectory cluster, that is, the ship motion state corresponding to different ship motion modes in the waterway grid. Exemplarily, the grid division process refers to Figure 5 .

[0105] In some embodiments of the present invention, motion pattern recognition is performed using a local autoencoder model based on the current trajectory data of the ship to obtain the current motion pattern of the ship, including but not limited to:

[0106] The current ship trajectory data is filled in through the masking mechanism to obtain the expected ship trajectory data. The expected ship trajectory data has the same data length as the ship's navigation trajectory data in the preset waters.

[0107] The expected trajectory data of the ship is input into each local autoencoder model to obtain the corresponding reconstruction error.

[0108] The current ship motion mode is obtained by matching the changes in the reconstruction error.

[0109] In this specific embodiment, the present embodiment first fills the acquired current trajectory data of the ship through the Mask mechanism in the Keras deep learning environment so that the data length of the expected ship trajectory data obtained is the same as the data length of the ship navigation trajectory data in the preset waters in the training data. Then, by inputting the expected ship trajectory data into each local autoencoder model respectively, the corresponding reconstruction error is obtained, and the current trajectory motion pattern of the ship is obtained according to the reconstruction error matching. Further, an abnormal behavior analysis is performed based on the current trajectory motion pattern of the ship and the probability density distribution data of the motion state to obtain the detection result of abnormal ship navigation behavior. For example, assuming that this embodiment selects a trajectory as the test trajectory l, its trajectory length len(l) is gradually increased. In order to ensure the stability of the model and the stable reconstruction of the trajectory, the Mask mechanism (masking mechanism) in the Keras deep learning environment is used to keep the length consistent with the input length of the training model. Among them, the Mask mechanism does not perform operations on the filled data. Further, this embodiment inputs the trajectory into different local autoencoder models at a fixed frequency to obtain the corresponding reconstruction error. It is easy to understand that the trajectory reconstruction error generated by the local autoencoder model in the inherent motion mode of the ship will gradually decrease, while the trajectory reconstruction error generated by the local autoencoder model other than the inherent motion mode will gradually increase. Figure 6 and Figure 7 This embodiment obtains the current motion pattern of the ship's trajectory based on reconstruction error matching and can quickly identify the current motion pattern category as motion pattern 4, i.e., C to A. This embodiment then performs abnormal behavior analysis based on the current motion pattern and motion state probability density distribution data to obtain detection results, enabling real-time detection of abnormal ship navigation behavior and reducing the occurrence of ship navigation accidents.

[0110] In some embodiments of the present invention, an abnormal behavior analysis of a ship is performed based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current motion mode of the ship to obtain abnormality detection results, including but not limited to:

[0111] Based on the ship's current trajectory motion mode and the corresponding motion state probability density distribution data, the abnormal alarm threshold is generated through Gaussian truncation probability. Among them, the abnormal alarm threshold includes the upper alarm level and the lower alarm level.

[0112] The abnormal behavior analysis of the ship's current trajectory data is performed according to the abnormal alarm threshold to obtain the detection results.

[0113] In this specific embodiment, this embodiment first generates an alarm abnormality alarm threshold including an alarm upper term and an alarm lower term based on the current trajectory motion mode of the ship and the motion state probability density distribution data corresponding to the motion mode through the Gaussian truncation probability. Then, this embodiment performs abnormal behavior analysis on the current trajectory data of the ship according to the abnormal alarm threshold to obtain a detection result. Specifically, after identifying the motion mode to which the current trajectory belongs, that is, obtaining the current trajectory motion mode of the ship according to the reconstruction error matching, this embodiment determines whether the current trajectory point conforms to the probability density distribution of the motion mode category to which it belongs under the current grid, that is, the motion state probability density distribution data. Then, the possibility of abnormality of the current trajectory of the ship is judged according to the Gaussian truncation probability, and the alarm upper term and the alarm lower term are generated as shown in the following formula (7):

[0114]

[0115] Where μ i is the mean value of the motion state within the grid, σ i is the variance of the motion state within the grid. It is easy to understand that if the current trajectory data of the ship exceeds the upper and lower boundaries of the current grid, the trajectory is considered abnormal, the detection result is obtained and an early warning is issued. It should be noted that in some embodiments of the present invention, the latitude and longitude points of abnormal behavior are displayed in the form of a heat map. For example, refer to Figure 7 .

[0116] An embodiment of the present invention further provides a system for detecting abnormal navigation behavior of inland waterway vessels, comprising:

[0117] The feature extraction module is used to extract the ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters. The preset interaction characteristics are the interaction characteristics of the distance between the ship and the preset buoy of the channel.

[0118] The first model building module is used to build an attention mechanism autoencoding model.

[0119] The feature fusion module is used to fuse the ship trajectory motion behavior features and preset interaction features through the attention mechanism autoencoder model to obtain intermediate layer information. The intermediate layer information is the feature information after the ship trajectory motion behavior features and the preset interaction features are aggregated.

[0120] The cluster analysis module is used to input the intermediate layer information into the cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters, wherein each trajectory cluster represents a ship motion mode.

[0121] The grid division module is used to calculate the probability density distribution data of the motion state of each ship motion mode through the grid division method according to the trajectory clustering clusters.

[0122] The second model building module is used to build a corresponding local autoencoder model based on the trajectory clusters, wherein the local autoencoder model corresponds to the ship's motion mode.

[0123] The motion pattern recognition module is used to perform motion pattern recognition based on the current ship trajectory data using a local autoencoder model to obtain the current ship motion pattern. The current ship trajectory data is the ship trajectory data obtained in real time.

[0124] The abnormal behavior analysis module is used to analyze the abnormal behavior of the ship based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current ship motion mode to obtain the abnormal detection results.

[0125] Reference Figure 2 One embodiment of the present invention further provides a system for detecting abnormal navigation behavior of inland waterway vessels, comprising:

[0126] At least one processor 210 .

[0127] At least one memory 220 is configured to store at least one program.

[0128] When at least one program is executed by at least one processor 210, the at least one processor 210 implements the method for detecting abnormal navigation behavior of inland vessels as described in the above embodiment.

[0129] An embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. The computer-executable instructions are executed by one or more control processors, for example, to execute the steps described in the above embodiment.

[0130] Those skilled in the art will appreciate that all or some of the steps and systems in the method disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include computer storage media (or non-transitory media) and communication media (or temporary media). As known to those skilled in the art, the term computer storage media is included in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data) and is volatile and non-volatile, removable, and non-removable. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (DVD), or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage, or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0131] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above implementation. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A method for detecting abnormal navigation behavior of inland waterway vessels, characterized in that: The following steps are involved: Extracting ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters; wherein the preset interaction characteristics are interaction characteristics of the distance between the ship and the preset buoy of the waterway; Build an attention mechanism autoencoding model; The ship trajectory motion behavior characteristics and the preset interaction characteristics are fused through the attention mechanism autoencoder model to obtain intermediate layer information; wherein the intermediate layer information is trajectory feature information obtained by aggregating the ship trajectory motion behavior characteristics and the preset interaction characteristics; Inputting the intermediate layer information into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters; wherein each trajectory cluster represents a ship motion mode; According to the trajectory clusters, the motion state probability density distribution data of each ship motion mode is statistically analyzed by a grid division method; Constructing a corresponding local autoencoder model according to the trajectory cluster; wherein the local autoencoder model corresponds to the ship motion mode; Performing motion pattern recognition using the local autoencoder model based on the current trajectory data of the ship to obtain the current motion pattern of the ship; wherein the current trajectory data of the ship is the ship trajectory data acquired in real time; An abnormal behavior analysis of the ship is performed based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current motion mode of the ship to obtain an abnormality detection result.

2. The method for detecting abnormal navigation behavior of inland waterway vessels according to claim 1, characterized in that: The extraction of ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters includes: Construct a ship trajectory motion behavior feature extraction model and a preset interaction feature extraction model based on a time sliding window; Extracting features from the ship's navigation trajectory data using the ship's trajectory motion behavior feature extraction model to obtain the ship's trajectory motion behavior features; The preset interactive feature extraction model is used to extract features from the ship navigation track data and the waterway landmark data to obtain the preset interactive features.

3. The method for detecting abnormal navigation behavior of inland waterway vessels according to claim 1, characterized in that: The attention mechanism autoencoder model includes a multi-dimensional measurement motion feature sequence attention mechanism submodule and a ship and environment interaction state attention submodule; The ship trajectory motion behavior characteristics and the preset interaction features are fused through the attention mechanism autoencoder model to obtain intermediate layer information, including: Determining a trajectory motion feature weight set by the multi-dimensional measurement motion feature sequence attention mechanism submodule; Determining an interaction feature attention weight set by the ship and environment interaction state attention submodule; The ship trajectory motion behavior characteristics and the preset interaction features are fused according to the trajectory motion feature weight set and the interaction feature attention weight set to obtain the intermediate layer information.

4. The method for detecting abnormal navigation behavior of inland waterway vessels according to claim 1, characterized in that: The step of inputting the intermediate layer information into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters includes: The intermediate layer information is input into a k-means clustering algorithm to perform cluster analysis on the ship trajectory to obtain the trajectory clusters.

5. The method for detecting abnormal navigation behavior of inland waterway vessels according to claim 1, characterized in that: The method of statistically analyzing the motion state probability density distribution data of each ship motion mode by a grid division method according to the trajectory clustering clusters includes: Dividing the waterway by a grid division method to obtain waterway grid data under each ship motion mode; Gaussian probability density distribution data of the ship motion state in the grid under each ship motion mode is statistically calculated based on the channel grid data and the trajectory clusters.

6. The method for detecting abnormal navigation behavior of inland waterway vessels according to claim 1, characterized in that: The performing motion mode recognition by the local autoencoder model according to the current trajectory data of the ship to obtain the current motion mode of the ship includes: Filling the current ship trajectory data through a masking mechanism to obtain expected ship trajectory data; wherein the expected ship trajectory data has the same data length as the ship navigation trajectory data in the preset waters; Inputting the expected ship trajectory data into each of the local autoencoder models to obtain corresponding reconstruction errors; The current ship motion mode is obtained according to the change difference matching of the reconstruction error.

7. The method for detecting abnormal navigation behavior of inland waterway vessels according to claim 1, characterized in that: The abnormal behavior analysis of the ship is performed based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current motion mode of the ship to obtain an abnormality detection result, including: Generate an abnormal alarm threshold value through Gaussian truncation probability according to the current trajectory motion mode of the ship and the corresponding motion state probability density distribution data; wherein the abnormal alarm threshold value includes an upper alarm level and a lower alarm level; An abnormal behavior analysis is performed on the current trajectory data of the ship according to the abnormal alarm threshold to obtain the detection result.

8. A system for detecting abnormal navigation behavior of inland waterway vessels, characterized in that: include: A feature extraction module is used to extract ship trajectory motion behavior characteristics and preset interaction characteristics based on the ship navigation trajectory data in the preset waters; wherein the preset interaction characteristics are interaction characteristics of the distance between the ship and the preset buoy of the channel; The first model building module is used to build an attention mechanism autoencoding model; a feature fusion module, configured to fuse the ship trajectory motion behavior features and the preset interaction features through the attention mechanism autoencoder model to obtain intermediate layer information; wherein the intermediate layer information is trajectory feature information obtained by aggregating the ship trajectory motion behavior features and the preset interaction features; A cluster analysis module, configured to input the intermediate layer information into a cluster analysis algorithm to perform cluster analysis on the ship trajectory to obtain trajectory clusters; wherein each trajectory cluster represents a ship motion mode; A grid division module is used to calculate the motion state probability density distribution data of each ship motion mode by a grid division method according to the trajectory clusters; A second model building module is used to build a corresponding local autoencoder model according to the trajectory cluster; wherein the local autoencoder model corresponds to the ship motion mode; a motion pattern recognition module, configured to perform motion pattern recognition using the local autoencoder model based on the current trajectory data of the ship to obtain the current motion pattern of the ship; wherein the current trajectory data of the ship is the ship trajectory data acquired in real time; The abnormal behavior analysis module is used to perform abnormal behavior analysis on the ship based on the current trajectory data of the ship and the motion state probability density distribution data corresponding to the current motion mode of the ship to obtain an abnormality detection result.

9. A system for detecting abnormal navigation behavior of inland waterway vessels, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method for detecting abnormal navigation behavior of inland vessels as described in any one of claims 1 to 7.

10. A computer storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the method for detecting abnormal navigation behavior of inland vessels as described in any one of claims 1 to 7 when executed by the processor.

Images