Large model-based ship abnormal behavior prediction method

Abstract

The invention discloses a ship abnormal behavior prediction method based on a large model, and the method comprises the following steps: S1, sensing a water area target object and environment information through the 5G-A sensing cooperative sensing capability, achieving the positioning, distance measurement and speed measurement of a ship, and carrying out the recognition and tracking of the ship in real time; s2, constructing a ship abnormal behavior detection model based on positioning, distance measurement and speed measurement information of the ship; s3, adopting the ship abnormal behavior detection model to predict abnormal behaviors of the ship; according to the method, sensing of ship behaviors is achieved through the 5G-A sensing technology, the accuracy of ship abnormal behavior recognition is improved in combination with large model reasoning, intelligent early warning and decision support are achieved, and the collaboration and effectiveness of supervision and prediction of the ship abnormal behaviors can be enhanced.

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular relates to a method for predicting abnormal behavior of a ship based on a large model. Background Art

[0002] Based on multi-source data fusion and intelligent algorithm model building, the accuracy of ship abnormal behavior identification can be improved, abnormal behavior analysis of ships in waterways, abnormal supervision of ships in rare marine species protection areas and operation areas, and abnormal behavior detection of bridge anti-ship collisions can be realized, thus ensuring the safe navigation of ships and the intelligent management of marine traffic, and it has broad development prospects.

[0003] The existing analysis and prediction of abnormal behavior of ships in waterways, such as the measured data from Yantai Port to Dalian Bay, adopts the construction of mixed Gaussian model and principal component analysis model to analyze the target trajectory behavior, pre-process the suspected abnormal track information, fit the track curve using mixed Gaussian model, build a track recognition probability model, obtain characteristic values ​​with the help of principal component analysis method, preset channel restriction parameters and abnormal behavior types, analyze ship behavior from four perspectives: motion abnormality, position abnormality, AIS equipment abnormality and other parameter abnormalities, build an evaluation index system for ship abnormal behavior, and use hierarchical analysis method and fuzzy comprehensive evaluation method to evaluate and predict ship abnormal behavior.

[0004] However, in terms of abnormal behavior analysis of waterway ships, the mixed Gaussian model and the principal component analysis model are not accurate and flexible enough when dealing with ship trajectories in complex environments. The mixed Gaussian model has limited fitting capabilities for nonlinear trajectories and cannot accurately capture the characteristics of abnormal tracks. The principal component analysis method loses some important information when extracting eigenvalues, which affects the judgment of abnormal behavior. There is a certain degree of subjectivity in the evaluation of abnormal ship behavior by the hierarchical analysis method and the fuzzy comprehensive evaluation method. The evaluation results are affected by expert opinions and weight settings, and lack objectivity and accuracy. Therefore, how to improve the ability to analyze and predict abnormal ship behavior has become an urgent problem that the industry wants to solve. Summary of the invention

[0005] To solve the above problems, the present invention proposes a method for predicting abnormal behavior of ships based on a large model. The method uses 5G-A synaesthesia technology to realize the perception of ship behavior, combines large model reasoning to improve the accuracy of identifying abnormal behavior of ships, realizes intelligent early warning and decision support, and can enhance the coordination and effectiveness of supervision and prediction of abnormal behavior of ships.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A method for predicting abnormal behavior of a ship based on a large model comprises the following steps: S1. Use 5G-A's synaesthesia and collaborative perception capabilities to perceive water target objects and environmental information, achieve ship positioning, distance measurement and speed measurement, and identify and track ships in real time; S2. Build a ship abnormal behavior detection model based on the ship's positioning, ranging and speed measurement information; The specific process of step S2 is: S21, collect the ship's positioning, ranging and speed measurement information to obtain a data sequence, recorded as { }; Clean the data to remove outliers and noise; S22. Construct a ship abnormal behavior detection model based on GRU recurrent neural network. In the ship abnormal behavior detection model, Reset the gate to decide how new input is combined with previous memory: ( ),in, is the output of the reset gate at time step t; σ(·) is the sigmoid function; is the weight matrix of the reset gate; is the bias term for resetting the gate; is the hidden state at the previous moment; is the input at the current moment; Update gate, decide to keep the previous memory and add new information: ,in, is the output of the update gate at time step t; is the weight matrix of the update gate; is the bias term of the update gate; The candidate hidden states are: ), current hidden state: ,in, is the candidate hidden state; (·) is the hyperbolic tangent function; is the concatenated vector Perform linear transformation; is element-wise multiplication; b is the bias term when calculating the candidate hidden state; For the current moment The hidden state of S3, using a ship abnormal behavior detection model to predict the abnormal behavior of the ship; The specific process of step S3 is: S31, the position coordinates, velocity vector and detection results of the past n time steps are used as model input, and the model input is expressed as X={ }; where the model input at the tth time step is , is the position coordinates of the ship; is the velocity vector of the ship; is the detection result of the ship; the model input X is input into the GRU layer in sequence, and after processing for multiple time steps, the final hidden state is obtained ; Then the final hidden state Input into the fully connected layer and output the predicted future position coordinate sequence Y of the ship. The prediction formula is: ,Y=FC( ), where GRU (·) represents the calculation of the GRU layer; FC (·) represents the calculation of the fully connected layer; the output of the ship position coordinates in the next m time steps is expressed as Y = { }; S32, using the sigmoid activation function, outputs a probability value between 0 and 1, which is used to indicate the probability of abnormal behavior of the ship. The classification formula is: p = σ (FC ( )), where p is the probability of abnormal behavior of the ship; σ (·) is the sigmoid function; If p is greater than a certain threshold, it is predicted that the ship has abnormal behavior; if p is less than or equal to a certain threshold, it is predicted that the ship does not have abnormal behavior; S33. The ship abnormal behavior detection model outputs a binary classification result of whether the ship has abnormal behavior, that is, Q={0,1}, where Q=0 indicates that the ship behavior is normal; Q=1 indicates that the ship behavior is abnormal.

[0007] Preferably, in step S1, the specific process of using the arrival time difference to locate the ship is as follows: There are multiple base stations, and the coordinates of the base stations are (x1, y1, z1), (x2, y2, z2), ..., (x n ,y n ,z n ), the ship's position coordinates are (x, y, z); The propagation time of the signal from the ship to the i-th base station is , the propagation speed is v, then the distance from the ship to the i-th base station is ; According to the distance formula: , for two base stations i and j, subtracting the two distance formulas yields: , simplifying to get: ,in, , and are the horizontal coordinate, vertical coordinate and vertical coordinate of base station i respectively; , and are the horizontal coordinate, vertical coordinate and vertical coordinate of base station j respectively; Given the position and distance difference between base stations i and j, the position of the ship can be determined by solving a set of linear equations.

[0008] Preferably, in step S1, the specific process of using array measurement angles to locate the ship is as follows: The signals transmitted by the ship are received by multiple base stations, and the arrival angle of the signals is measured by the antenna array to determine the position of the ship; the coordinates of base station i and base station j are respectively ( , , )and( , , ), the measured angles of the ship signal reaching base station i and base station j are ( )and( ); The coordinate system is established with base station i as the origin, according to the angle ( ) to obtain the direction vector of the ship's position in the coordinate system for: , establish a coordinate system with base station j as the origin, according to the angle ( ) to obtain the direction vector of the ship's position in the coordinate system for: , the difference between the position coordinates of the ship and base station i is parallel to the direction vector from the ship to base station i, and the difference between the position coordinates of the ship and base station j is parallel to the direction vector from the ship to base station j, that is: , ,in, is the difference between the position coordinates of the ship and base station i; is the difference between the position coordinates of the ship and base station j; and μ are respectively known proportionality coefficients; Given the location coordinates of base station i and base station j, list the equations: , , solve the equations to get the position of the ship .

[0009] Preferably, in step S1, the process of measuring the distance from the ship to the base station is: According to the signal propagation model, there is a relationship between the received signal strength and the distance. The distance is calculated by measuring the received signal strength. The logarithmic distance path loss model is: ,in, To measure distance The received signal strength measured at is the transmission signal strength; is the transmitting antenna gain; is the receiving antenna gain; is the path loss index; To measure distance; is the reference distance; is a Gaussian random variable with zero mean; According to the reference distance The received signal strength measured at , calculated at the reference distance Path loss at: , the known parameters and the received signal strength at the measurement distance d are measured Substitute into the logarithmic distance path loss model: , has been calculated at the reference distance The value of After substituting the value of , we get: , simplifying to get: , solve to get the measured distance d: .

[0010] Preferably, in step S1, the speed measurement process from the ship to the base station is: Based on the Doppler effect principle, in ship speed measurement, the signal transmitting source is at the base station or on the ship, and the signal receiving source is at the other side; when there is relative motion between the ship and the base station, the frequency of the received signal changes, and the relative speed between the ship and the base station is calculated by measuring the frequency change; the ship speed measurement model is: A continuous wave signal with a transmission frequency of , when the ship moves with a velocity vector v t When moving relative to the base station, the frequency of the received signal changes according to the Doppler effect. for; , where c is the propagation speed of electromagnetic waves in the medium; is the speed of the receiving end relative to the medium, is the velocity of the transmitter relative to the medium; In the case of ship speed measurement, it is assumed that the base station is stationary, that is, =0, the transmitter is on the ship, and the speed of the ship relative to the base station is v t ; When the ship is far away from the base station, , after derivation, the velocity vector of the ship is: ,in, = is the Doppler frequency shift term; When the ship approaches the base station, , after derivation, the velocity vector of the ship is: ,in, = is the Doppler shift term.

[0011] After adopting the above technical solution, the present invention has the following beneficial effects: 1. The present invention uses 5G-A synaesthesia technology to achieve high-precision perception of ship behavior, and combines it with a large model algorithm to realize multi-source data fusion analysis to improve the accuracy of abnormal behavior identification. For example, the position, speed, heading and other information of the ship are transmitted in real time through the 5G network, and then a comprehensive analysis is performed using a large model algorithm to accurately determine whether the ship has abnormal behavior.

[0012] 2. The present invention can continuously train and optimize the ship abnormal behavior detection model so that it can better handle nonlinear and dynamically changing ship trajectories and environmental factors, improve the accuracy and reliability of abnormal behavior identification, and improve the adaptability to complex ship behaviors and marine environments.

[0013] 3. The present invention predicts the future behavior trend of ships by analyzing historical data and real-time perception information, can evaluate potential risk factors, take preventive measures, reduce the probability of accidents, and achieve all-round coordinated supervision of ship behavior. It also uses the intelligent analysis capabilities of large models to predict ship behavior and assess risks, and can achieve intelligent early warning and decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0016] like Figure 1 As shown, a method for predicting abnormal behavior of a ship based on a large model comprises the following steps: S1. Use 5G-A's synaesthesia and collaborative perception capabilities to perceive water target objects and environmental information, achieve ship positioning, distance measurement and speed measurement, and identify and track ships in real time; In step S1, the specific process of using arrival time difference to locate the ship is as follows: There are multiple base stations, and the coordinates of the base stations are (x1, y1, z1), (x2, y2, z2), ..., (x n ,y n ,z n ), the ship's position coordinates are (x, y, z); The propagation time of the signal from the ship to the i-th base station is , the propagation speed is v, then the distance from the ship to the i-th base station is ; According to the distance formula: , for two base stations i and j, subtracting the two distance formulas yields: , simplifying to get: ,in, , and are the horizontal coordinate, vertical coordinate and vertical coordinate of base station i respectively; , and are the horizontal coordinate, vertical coordinate and vertical coordinate of base station j respectively; Given the position and distance difference between base stations i and j, the position of the ship is determined by solving a set of linear equations; Alternatively, in step S1, the specific process of using the array measurement angle to locate the ship is: The signals transmitted by the ship are received by multiple base stations, and the arrival angle of the signals is measured by the antenna array to determine the position of the ship; the coordinates of base station i and base station j are respectively ( , , )and( , , ), the measured angles of the ship signal reaching base station i and base station j are ( )and( ); The coordinate system is established with base station i as the origin, according to the angle ( ) to obtain the direction vector of the ship's position in the coordinate system for: , establish a coordinate system with base station j as the origin, according to the angle ( ) to obtain the direction vector of the ship's position in the coordinate system for: , the difference between the position coordinates of the ship and base station i is parallel to the direction vector from the ship to base station i, and the difference between the position coordinates of the ship and base station j is parallel to the direction vector from the ship to base station j, that is: , ,in, is the difference between the position coordinates of the ship and base station i; is the difference between the position coordinates of the ship and base station j; and μ are respectively known proportionality coefficients; Given the location coordinates of base station i and base station j, list the equations: , , solve the equations to get the position of the ship ; In step S1, the distance measurement process from the ship to the base station is: According to the signal propagation model, there is a relationship between the received signal strength and the distance. The distance is calculated by measuring the received signal strength. The logarithmic distance path loss model is: ,in, To measure distance The received signal strength measured at is the transmission signal strength; is the transmitting antenna gain; is the receiving antenna gain; is the path loss index; To measure distance; is the reference distance; is a Gaussian random variable with zero mean; According to the reference distance The received signal strength measured at , calculated at the reference distance Path loss at: , the known parameters and the received signal strength at the measurement distance d are measured Substitute into the logarithmic distance path loss model: , has been calculated at the reference distance The value of After substituting the value of , we get: , simplifying to get: , solve to get the measured distance d: ; In step S1, the speed measurement process from the ship to the base station is: Based on the Doppler effect principle, in ship speed measurement, the signal transmitting source is at the base station or on the ship, and the signal receiving source is at the other side; when there is relative motion between the ship and the base station, the frequency of the received signal changes, and the relative speed between the ship and the base station is calculated by measuring the frequency change; the ship speed measurement model is: A continuous wave signal with a transmission frequency of , when the ship moves with a velocity vector v t When moving relative to the base station, the frequency of the received signal changes according to the Doppler effect. for; , where c is the propagation speed of electromagnetic waves in the medium; is the speed of the receiving end relative to the medium, is the velocity of the transmitter relative to the medium; In the case of ship speed measurement, it is assumed that the base station is stationary, that is, =0, the transmitter is on the ship, and the speed of the ship relative to the base station is v t ; When the ship is far away from the base station, , after derivation, the velocity vector of the ship is: ,in, = is the Doppler shift term; When the ship approaches the base station, , after derivation, the velocity vector of the ship is: ,in, = is the Doppler frequency shift term; S2. Build a ship abnormal behavior detection model based on the ship's positioning, ranging and speed measurement information; The specific process of step S2 is: S21, collect the ship's positioning, ranging and speed measurement information to obtain a data sequence, recorded as { }; Clean the data to remove outliers and noise; S22. Construct a ship abnormal behavior detection model based on GRU recurrent neural network. In the ship abnormal behavior detection model, Reset the gate to decide how new input is combined with previous memory: ( ),in, is the output of the reset gate at time step t; σ(·) is the sigmoid function; is the weight matrix of the reset gate; is the bias term for resetting the gate; is the hidden state at the previous moment; is the input at the current moment; Update gate, decide to keep the previous memory and add new information: ,in, is the output of the update gate at time step t; is the weight matrix of the update gate; is the bias term of the update gate; The candidate hidden states are: ), current hidden state: ,in, is the candidate hidden state; (·) is the hyperbolic tangent function; is the concatenated vector Perform linear transformation; is element-wise multiplication; b is the bias term when calculating the candidate hidden state; For the current moment The hidden state of S3, using a ship abnormal behavior detection model to predict the abnormal behavior of the ship; The specific process of step S3 is: S31, the position coordinates, velocity vector and detection results of the past n time steps are used as model input, and the model input is expressed as X={}; where the model input at the tth time step is , is the position coordinates of the ship; is the velocity vector of the ship; is the detection result of the ship; the model input X is input into the GRU layer in sequence, and after processing for multiple time steps, the final hidden state is obtained ; Then the final hidden state Input into the fully connected layer and output the predicted future position coordinate sequence Y of the ship. The prediction formula is: ,Y=FC( ), where GRU (·) represents the calculation of the GRU layer; FC (·) represents the calculation of the fully connected layer; the output of the ship position coordinates in the next m time steps is expressed as Y = { }; S32, using the sigmoid activation function, outputs a probability value between 0 and 1, which is used to indicate the probability of abnormal behavior of the ship. The classification formula is: p = σ (FC ( )), where p is the probability of abnormal behavior of the ship; σ (·) is the sigmoid function; If p is greater than a certain threshold, it is predicted that the ship has abnormal behavior; if p is less than or equal to a certain threshold, it is predicted that the ship does not have abnormal behavior; S33. The ship abnormal behavior detection model outputs a binary classification result of whether the ship has abnormal behavior, that is, Q={0,1}, where Q=0 indicates that the ship behavior is normal; Q=1 indicates that the ship behavior is abnormal.

[0017] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting abnormal behavior of a ship based on a large model, characterized in that: The following steps are involved: S1. Use 5G-A's synaesthesia and collaborative perception capabilities to perceive water target objects and environmental information, achieve ship positioning, distance measurement and speed measurement, and identify and track ships in real time; S2. Build a ship abnormal behavior detection model based on the ship's positioning, ranging and speed measurement information; The specific process of step S2 is: S21, collect the ship's positioning, ranging and speed measurement information to obtain a data sequence, recorded as { }; Clean the data to remove outliers and noise; S22. Construct a ship abnormal behavior detection model based on GRU recurrent neural network. In the ship abnormal behavior detection model, Reset the gate to decide how new input is combined with previous memory: ( ),in, is the output of the reset gate at time step t; σ(·) is the sigmoid function; is the weight matrix of the reset gate; is the bias term for resetting the gate; is the hidden state at the previous moment; is the input at the current moment; Update gate, decide to keep the previous memory and add new information: ,in, is the output of the update gate at time step t; is the weight matrix of the update gate; is the bias term of the update gate; The candidate hidden states are: ), current hidden state: ,in, is the candidate hidden state; (·) is the hyperbolic tangent function; is the concatenated vector Perform linear transformation; is element-wise multiplication; b is the bias term when calculating the candidate hidden state; For the current moment The hidden state of S3, using a ship abnormal behavior detection model to predict the abnormal behavior of the ship; The specific process of step S3 is: S31, the position coordinates, velocity vector and detection results of the past n time steps are used as model input, and the model input is expressed as X={ }; where the model input at the tth time step is , is the position coordinates of the ship; is the velocity vector of the ship; is the detection result of the ship; the model input X is input into the GRU layer in sequence, and after processing for multiple time steps, the final hidden state is obtained ; Then the final hidden state Input into the fully connected layer and output the predicted future position coordinate sequence Y of the ship. The prediction formula is: ,Y=FC( ), where GRU (·) represents the calculation of the GRU layer; FC (·) represents the calculation of the fully connected layer; the output of the ship position coordinates in the next m time steps is expressed as Y = { }; S32, using the sigmoid activation function, outputs a probability value between 0 and 1, which is used to indicate the probability of abnormal behavior of the ship. The classification formula is: p = σ (FC ( )), where p is the probability of abnormal behavior of the ship; σ (·) is the sigmoid function; If p is greater than a certain threshold, it is predicted that the ship has abnormal behavior; if p is less than or equal to a certain threshold, it is predicted that the ship does not have abnormal behavior; S33. The ship abnormal behavior detection model outputs a binary classification result of whether the ship has abnormal behavior, that is, Q={0,1}, where Q=0 indicates that the ship behavior is normal; Q=1 indicates that the ship behavior is abnormal.

2. A method for predicting abnormal behavior of a ship based on a large model as claimed in claim 1, characterized in that: In step S1, the specific process of using arrival time difference to locate the ship is as follows: There are multiple base stations, and the coordinates of the base stations are (x1, y1, z1), (x2, y2, z2), ..., (x n ,y n ,z n ), the ship's position coordinates are (x, y, z); The propagation time of the signal from the ship to the i-th base station is , the propagation speed is v, then the distance from the ship to the i-th base station is ; According to the distance formula: , for two base stations i and j, subtracting the two distance formulas yields: , simplifying to get: ,in, , and are the horizontal coordinate, vertical coordinate and vertical coordinate of base station i respectively; , and are the horizontal coordinate, vertical coordinate and vertical coordinate of base station j respectively; Given the position and distance difference between base stations i and j, the position of the ship can be determined by solving a set of linear equations.

3. A method for predicting abnormal behavior of a ship based on a large model as claimed in claim 1, characterized in that: In step S1, the specific process of using array measurement angle to locate the ship is: receiving the signal transmitted by the ship through multiple base stations, using the antenna array to measure the arrival angle of the signal, and determining the position of the ship; wherein the coordinates of base station i and base station j are respectively ( , , )and( , , ), the measured angles of the ship signal reaching base station i and base station j are ( )and( ); The coordinate system is established with base station i as the origin, according to the angle ( ) to obtain the direction vector of the ship's position in the coordinate system for: , establish a coordinate system with base station j as the origin, according to the angle ( ) to obtain the direction vector of the ship's position in the coordinate system for: , the difference between the position coordinates of the ship and base station i is parallel to the direction vector from the ship to base station i, and the difference between the position coordinates of the ship and base station j is parallel to the direction vector from the ship to base station j, that is: , ,in, is the difference between the position coordinates of the ship and base station i; is the difference between the position coordinates of the ship and base station j; and μ are known proportionality coefficients respectively; Given the location coordinates of base station i and base station j, list the equations: , , solve the equations to get the position of the ship .

4. A method for predicting abnormal behavior of a ship based on a large model as claimed in claim 1, characterized in that: In step S1, the distance measurement process from the ship to the base station is: According to the signal propagation model, there is a relationship between the received signal strength and the distance. The distance is calculated by measuring the received signal strength. The logarithmic distance path loss model is: ,in, To measure distance The received signal strength measured at is the transmission signal strength; is the transmitting antenna gain; is the receiving antenna gain; is the path loss index; To measure distance; is the reference distance; is a Gaussian random variable with zero mean; According to the reference distance The received signal strength measured at , calculated at the reference distance Path loss at: , the known parameters and the received signal strength at the measurement distance d are measured Substitute into the logarithmic distance path loss model: , has been calculated at the reference distance The value of After substituting the value of , we get: , simplifying to get: , solve to get the measured distance d: .

5. A method for predicting abnormal behavior of a ship based on a large model as claimed in claim 1, characterized in that: In step S1, the speed measurement process from the ship to the base station is: Based on the Doppler effect principle, in ship speed measurement, the signal transmitting source is at the base station or on the ship, and the signal receiving source is at the other side; when there is relative motion between the ship and the base station, the frequency of the received signal changes, and the relative speed between the ship and the base station is calculated by measuring the frequency change; the ship speed measurement model is: A continuous wave signal with a transmission frequency of , when the ship moves with a velocity vector v t When moving relative to the base station, the frequency of the received signal changes according to the Doppler effect. for; , where c is the propagation speed of electromagnetic waves in the medium; is the speed of the receiving end relative to the medium, is the velocity of the transmitter relative to the medium; In the case of ship speed measurement, it is assumed that the base station is stationary, that is, =0, the transmitter is on the ship, and the speed of the ship relative to the base station is v t ; When the ship is far away from the base station, , after derivation, the velocity vector of the ship is: ,in, = is the Doppler shift term; When the ship approaches the base station, , after derivation, the velocity vector of the ship is: ,in, = is the Doppler shift term.

Images