Maritime target trajectory prediction method and device based on multi-source data

Abstract

The present invention proposes a method and device for predicting the trajectory of maritime targets based on multi-source data. The prediction method includes: obtaining multi-source perception data of maritime targets; performing trajectory grid mapping encoding on maritime targets based on the multi-source perception data; performing data fusion on the multi-source perception data within a preset time period in the trajectory grid; obtaining an input sequence based on the perception data after data fusion and inputting it into a prediction model; and calculating an output sequence for maritime target trajectory prediction using a preset model. This invention addresses the problem of being unable to continuously track ship targets under the expansion of incomplete observation data information. It fully utilizes the complementarity of multi-source data in ship target monitoring, while taking into account the confidence of multi-source means. Through key technologies such as trajectory grid mapping encoding, multi-source data fusion, and prediction models, it realizes the existence probability mapping of maritime target trajectories in long time series and large-scale space under global grid subdivision, thereby realizing the prediction and analysis of maritime target trajectories.

Description

Technical Field

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

[0002] To predict ship trajectories, a recurrent neural network-based ship trajectory prediction method has been proposed. To ensure the regularity of trajectory prediction, commercial or civilian ships with non-adversarial behavior were selected as experimental subjects. This method constructs two prediction models based on recurrent neural networks: the Long Short-Term Memory (LSTM) model and the Gated Recurrent Unit (GRU) model. These models are trained using AIS datasets as input, ultimately achieving accurate and efficient prediction of a ship's future longitude and latitude coordinates. Comparative experiments using a large amount of AIS data have demonstrated the effectiveness of the proposed ship trajectory prediction method in practical applications.

[0003] Another related technique proposed a mobile object trajectory prediction algorithm based on a fuzzy long-short-term neural network. However, due to the limited spatiotemporal granularity of the experimental data, the algorithm's prediction accuracy was only verified using data with a 30-minute time granularity. This method proposes a mobile object trajectory prediction algorithm based on a fuzzy long-short-term memory neural network model. This method introduces the concept of fuzzy trajectories to address the sharp boundary problem caused by fixed grid subdivision. It also improves the traditional LSTM by leveraging the proximity and periodic travel characteristics of historical mobile object trajectories to improve the accuracy of mobile object trajectory position prediction. Finally, the method was experimentally analyzed using a mobile communication signaling trajectory dataset of 100,000 users in a city for 15 consecutive working days. The results show that the method achieves an average prediction accuracy of 83.98% within a 30-minute prediction period, which is significantly higher than the classic prediction model.

[0004] The above-mentioned related technologies have the following defects:

[0005] (1) Single data source:

[0006] Existing ship trajectory prediction algorithms only use AIS data. Special-purpose ships will turn off AIS signals after entering or leaving the port, resulting in sparse and discontinuous ship trajectories. However, remote sensing, public opinion, radar, radio and other data can effectively increase the density of ship tracks. Existing algorithms that simply use AIS data as a data source will fail to accurately predict ship trajectories.

[0007] (2) The prediction time and space range is limited:

[0008] The existing ship trajectory prediction algorithm using neural networks can only make relatively accurate predictions of ships in short time series and small spatial ranges, and cannot predict ship trajectories in long time series and large-scale spaces.

[0009] (3) Limited application scenarios:

[0010] Existing technologies can only be applied when the monitored object has turned on the AIS device. They are not applicable to scenarios where the monitored object in special areas turns off the AIS or turns on the AIS device intermittently. The latter is more commonly used in the field of actively turning off the AIS behavior. Summary of the Invention

[0011] The technical problem to be solved by the present invention is how to improve the reliability of ship track prediction. The present invention proposes a method and device for predicting marine target tracks based on multi-source data.

[0012] A method for predicting maritime target trajectories based on multi-source data according to an embodiment of the present invention includes:

[0013] Acquire multi-source perception data of maritime targets;

[0014] Based on multi-source perception data, performing trajectory grid mapping encoding on the maritime target;

[0015] Perform data fusion on multi-source sensing data within a preset time period in the trajectory grid;

[0016] Obtain input sequence based on the perception data after data fusion and input it into the prediction model;

[0017] The output sequence of the maritime target trajectory prediction is calculated by the preset model.

[0018] According to some embodiments of the present invention, the method further comprises:

[0019] The acquired multi-source perception data is annotated and stored.

[0020] In some embodiments of the present invention, multi-source perception data are acquired using AIS, remote sensing, public opinion, radar, and radio, and each of the perception data includes the position, speed, confidence, acquisition time, and data source of the maritime target.

[0021] According to some embodiments of the present invention, performing data fusion on multi-source sensing data within a preset time period in a trajectory grid includes:

[0022] Grid the target area according to the specified longitude and latitude, and perform unique geocoding based on the row and column information of each grid;

[0023] Arrange all the trajectory geocodes of the maritime targets into a text sequence in order, where each code in the text sequence contains speed, position, and confidence information;

[0024] Add the departure and destination at the beginning and end of the text sequence, and add a colon to the text sequence to generate a complete trajectory encoding text.

[0025] In some embodiments of the present invention, performing data fusion on multi-source sensing data within a preset time period in a trajectory grid includes:

[0026] When grid mapping is used for coding, if there is perception data obtained from multiple sources within a preset time period within a grid, a multi-source data fusion method based on Kalman filtering is used to generate unique geocoding speed, position, and confidence information.

[0027] A device for predicting maritime target trajectories based on multi-source data according to an embodiment of the present invention includes:

[0028] Data acquisition module, used to obtain multi-source perception data of maritime targets;

[0029] A grid mapping module is used to perform trajectory grid mapping encoding on the maritime target based on multi-source perception data;

[0030] The data fusion module is used to fuse the multi-source perception data within a preset time period in the trajectory grid;

[0031] The calculation module is used to obtain an input sequence based on the perception data after data fusion, and input it into the prediction model, and calculate the output sequence of the maritime target trajectory prediction through the preset model.

[0032] According to some embodiments of the present invention, the apparatus further comprises:

[0033] The standardization processing module is used to annotate and store the acquired multi-source perception data.

[0034] In some embodiments of the present invention, the data acquisition module uses AIS, remote sensing, public opinion, radar, and radio to obtain multi-source perception data, and each of the perception data includes the position, speed, confidence, acquisition time, and data source of the maritime target.

[0035] According to some embodiments of the present invention, the grid mapping module is specifically configured to:

[0036] Grid the target area according to the specified longitude and latitude, and perform unique geocoding based on the row and column information of each grid;

[0037] Arrange all the trajectory geocodes of the maritime targets into a text sequence in order, where each code in the text sequence contains speed, position, and confidence information;

[0038] Add the departure and destination at the beginning and end of the text sequence, and add a colon to the text sequence to generate a complete trajectory encoding text.

[0039] In some embodiments of the present invention, the data fusion module is specifically used to:

[0040] When grid mapping is used for coding, if there is perception data obtained from multiple sources within a preset time period within a grid, a multi-source data fusion method based on Kalman filtering is used to generate unique geocoding speed, position, and confidence information.

[0041] The present invention addresses the problem of being unable to continuously track ship targets under the expansion of incomplete observation data information. It fully utilizes the complementarity of multi-source data such as AIS, remote sensing, public opinion, radar, and radio in ship target monitoring, while taking into account the confidence of multi-source means. Through key technologies such as ship trajectory grid mapping encoding, multi-source data fusion based on Kalman filtering, and sequence-to-sequence ship trajectory prediction model, it realizes the existence probability mapping of long-term series and large-scale space ship trajectories under global grid subdivision, thereby realizing the predictive analysis of the ship trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Flowchart of a method for predicting maritime target trajectories based on multi-source data according to an embodiment of the present invention;

[0043] Figure 2 Flowchart of a method for predicting maritime target trajectories based on multi-source data according to an embodiment of the present invention;

[0044] Figure 3 Flowchart of a method for gridding the predicted probability of maritime target position and trajectory in a maritime target trajectory prediction method based on multi-source data according to an embodiment of the present invention;

[0045] Figure 4 Schematic diagram of a ship trajectory grid mapping encoding method in a method for predicting maritime target trajectory based on multi-source data according to an embodiment of the present invention;

[0046] Figure 5 Schematic diagram of a sequence-to-sequence ship trajectory prediction model in a method for predicting maritime target trajectories based on multi-source data according to an embodiment of the present invention.

[0047] Reference numerals:

[0048] Prediction device 100,

[0049] Data acquisition module 10, standardization processing module 20, grid mapping module 30, data fusion module 40, calculation module 50. DETAILED DESCRIPTION

[0050] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose, the present invention is described in detail below with reference to the accompanying drawings and preferred embodiments.

[0051] like Figure 1 and Figure 2 As shown, the method for predicting maritime target trajectories based on multi-source data according to an embodiment of the present invention includes:

[0052] S100, acquires multi-source perception data of maritime targets;

[0053] S200, based on multi-source perception data, performs trajectory grid mapping encoding on maritime targets;

[0054] S300, performing data fusion on multi-source sensing data within a preset time period in the trajectory grid;

[0055] S400, obtaining an input sequence based on the perception data after data fusion, and inputting it into the prediction model;

[0056] S500, obtaining an output sequence of maritime target trajectory prediction through calculation using a preset model.

[0057] like Figure 1 and Figure 2 As shown, according to some embodiments of the present invention, the method further includes:

[0058] S150: annotate and store the acquired multi-source perception data.

[0059] In some embodiments of the present invention, in step S100, multi-source perception data are acquired using AIS, remote sensing, public opinion, radar, and radio, and each perception data includes the position, speed, confidence, acquisition time, and data source of the maritime target.

[0060] According to some embodiments of the present invention, Figure 3 As shown, step S300, fusing multi-source sensing data within a preset time period in the trajectory grid, includes:

[0061] S310, gridding the target area according to the specified longitude and latitude, and performing unique geocoding based on the row and column information of each grid;

[0062] S320, arranging all trajectory geocodes of the maritime target in order into a text sequence, where each code in the text sequence contains speed, position, and confidence information;

[0063] S330 , adding the departure place and the destination at the beginning and end of the text sequence, and adding a colon to the text sequence to generate a complete trajectory encoding text.

[0064] In some embodiments of the present invention, step S400 of fusing multi-source sensing data within a preset time period in a trajectory grid includes:

[0065] When grid mapping is used for coding, if there is perception data obtained from multiple sources within a preset time period within a grid, a multi-source data fusion method based on Kalman filtering is used to generate unique geocoding speed, position, and confidence information.

[0066] like Figure 5 As shown, the maritime target trajectory prediction device 100 based on multi-source data according to an embodiment of the present invention includes: a data acquisition module 10, a grid mapping module 30, a data fusion module 40 and a calculation module 50.

[0067] The data acquisition module 10 is used to acquire multi-source perception data of maritime targets;

[0068] The grid mapping module 30 is used to perform trajectory grid mapping encoding on the maritime target based on multi-source perception data;

[0069] The data fusion module 40 is used to perform data fusion on multi-source sensing data within a preset time period in the trajectory grid;

[0070] The calculation module 50 is used to obtain an input sequence based on the perception data after data fusion, and input it into the prediction model, and obtain an output sequence of the maritime target trajectory prediction through calculation of the preset model.

[0071] According to some embodiments of the present invention, the apparatus further includes: a standardization processing module 20 for annotating and storing the acquired multi-source perception data.

[0072] In some embodiments of the present invention, the data acquisition module uses AIS, remote sensing, public opinion, radar, and radio to obtain multi-source perception data, and each perception data includes the position, speed, confidence, acquisition time and data source of the maritime target.

[0073] According to some embodiments of the present invention, Figure 3 As shown, the grid mapping module 30 is specifically used for:

[0074] Grid the target area according to the specified longitude and latitude, and perform unique geocoding based on the row and column information of each grid;

[0075] Arrange all the trajectory geocodes of the maritime targets into a text sequence in order, where each code in the text sequence contains speed, position, and confidence information;

[0076] Add the departure and destination at the beginning and end of the text sequence, and add a colon to the text sequence to generate a complete trajectory encoding text.

[0077] In some embodiments of the present invention, the data fusion module 40 is specifically configured to:

[0078] When grid mapping is used for coding, if there is perception data obtained from multiple sources within a preset time period within a grid, a multi-source data fusion method based on Kalman filtering is used to generate unique geocoding speed, position, and confidence information.

[0079] The present invention addresses the problem of being unable to continuously track ship targets under the expansion of incomplete observation data information. It fully utilizes the complementarity of multi-source data such as AIS, remote sensing, public opinion, radar, and radio in ship target monitoring, while taking into account the confidence of multi-source means. Through key technologies such as ship trajectory grid mapping encoding, multi-source data fusion based on Kalman filtering, and sequence-to-sequence ship trajectory prediction model, it realizes the existence probability mapping of long-term series and large-scale space ship trajectories under global grid subdivision, thereby realizing the predictive analysis of the ship trajectory.

[0080] The following describes in detail the method and apparatus for predicting maritime target trajectory based on multi-source data according to the present invention with reference to the accompanying drawings. It should be understood that the following description is merely exemplary and should not be construed as limiting the present invention.

[0081] like Figure 1 and Figure 2 As shown in the figure, the maritime target trajectory prediction method based on multi-source data uses AIS, remote sensing, public opinion, radar, radio and other technical means in their respective fields to perceive ship targets and store the results in a unified format. The ship trajectory obtained by multi-source data is grid mapped and encoded, and the ship position, speed and confidence information in the grid are obtained through a multi-source data fusion method based on Kalman filtering. Finally, the geographic coding and grid attribute information are input into the sequence ship trajectory prediction model to realize the probability prediction of ship trajectory.

[0082] like Figure 2 As shown, the specific steps of the prediction method are as follows:

[0083] A1, ship target perception;

[0084] The position and speed information of ship targets are extracted using technical methods in their respective fields such as AIS, remote sensing, public opinion, radar, and radio, and the confidence level of the perceived targets is obtained.

[0085] A2, standardization of perception data format;

[0086] The perception data will be stored in the format of [position, speed, confidence, acquisition time, data source] and will be used as prior information for ship position prediction.

[0087] A3, ship track grid mapping code;

[0088] like Figure 3 As shown in Figure 1, the ship trajectory grid mapping code maps the ship target perceived by multi-source data into a unique coding grid. The mapping method is as follows:

[0089] A31, grid the target area according to the specified longitude and latitude, and perform unique geocoding based on the row and column information of each grid;

[0090] A32, arranges all the track geocodes of a ship into a text sequence in order, and each code contains speed, position, and confidence information;

[0091] A33, adds the departure and destination at the beginning and end of the text sequence, and adds a colon to the text sequence to generate a complete trajectory encoding text;

[0092] A34, when mapping grids, if there is a trajectory point obtained from multiple sources within a specified time period t in a grid, a multi-source data fusion method based on Kalman filtering is used to generate unique geocoded speed, position, and confidence information;

[0093] A35, selects a suitable sliding window to generate source sequence samples. The sliding window length represents the length of the historical trajectory that needs to be taken into account.

[0094] A4, multi-source data fusion based on Kalman filtering;

[0095] In a grid, within a specified time period t, ship targets may be simultaneously sensed by multiple sources such as multiple remote sensing satellites, AIS, public opinion, radar, and radio. The ship target information that integrates l types of data is expressed by weighted average as follows:

[0096]

[0097] Where, represents the fusion estimation result of the Cartesian coordinate position of the ship in the time period t; represents the fusion estimation result of the ship speed in the reference system during the time period t; A represents the fusion estimation result of ship confidence within time period t; i (t) represents the fusion weight matrix; It represents the Cartesian coordinate estimation result of the ship position in time period t when the Kalman filter is used to perform the i-th (1≤i≤l) multi-source data observation; It represents the estimated result of the ship speed in the reference system during the time period t when the Kalman filter is used to perform the i-th (1≤i≤l) multi-source data observation; It represents the built-in confidence estimation result in time period t when using Kalman filtering for the i-th (1≤i≤l) multi-source data observation.

[0098] The multi-source data fusion weight matrix is ​​determined by the following formula:

[0099]

[0100] Where, P ij (t) represents the cross-covariance matrix of the filtering errors between the i-th and j-th multi-source observation data.

[0101] The cross-covariance matrix of the filter errors is determined by this formula:

[0102] P ij (t) = [I5-K i (t)·H]·[φ·P ij (t-1)·φ T ]·[I5-K j (t)·H] T ;

[0103] Where I5 ​​represents the 5×5 unit matrix; K i (t)(K j (t)) represents the Kalman filter gain matrix subsystem (1≤i≤l,1≤j≤l) of the i-th (j-th) sensor in the specified time period t; H, φ represent the constant matrix of the system:

[0104]

[0105] The multi-source observation data fusion process defined by the above formula is optimal because it minimizes the trace of the fusion estimator error matrix.

[0106] A5, sequence-to-sequence ship trajectory prediction model;

[0107] The sequence-to-sequence ship trajectory prediction model is designed according to the encoder-decoder architecture and consists of two multi-layer LSTM networks. The first LSTM (encoder) maps x to a state vector c, where c summarizes the entire source sequence; the second LSTM (decoder) uses c and x to T Generate target sequences. The encoder extracts the ship’s behavior features based on its historical trajectory positions, and the decoder predicts the corresponding future movement trajectory of the ship.

[0108] The elements in the target sequence are generated sequentially over several time steps. When the time step is t, the target element y tAccording to the previously generated element y t-1 and the hidden element h t-1 (Decoder) is generated according to the following formula:

[0109] h t =f(h t-1 ,y t-1 );

[0110] Because the first element of the target sequence has no precedent and hidden state, it is calculated differently. In the standard encoder-decoder approach, h1 is calculated based on c and a virtual string start element ( <bos>), in this model, the value will be calculated from the position of the current historical trajectory, so the condition of h1 is the source sequence x T The last element in and c are replaced by <bos>, so the hidden state h1 is initialized as:

[0111] h1=f(c,x T ).

[0112] like Figure 4 As shown in the figure, each LSTM in the sequence-to-sequence ship trajectory prediction model has three hidden layers (hollow and horizontal striped rectangles). The diagonal striped rectangles represent the hidden state of the network at each time step. The end of the source and target sequences are marked with two special elements: a dot character (.) and a sequence end character (terminator). <eos>By inputting the latest position of the ship (E5, F5, F6, (.), <eos>) sequence, the first LSTM calculates the representation vector, and then uses the vector to generate (G7, H7, G8, I9, G10, arrival, (.), <eos>)sequence.

[0113] Through this sequence, the probability distribution vector of the grid position where the ship is expected to arrive at a specified time can be obtained, and the probability distribution map of the ship's position can be obtained by outputting the constrained probability vector.

[0114] In summary, the present invention uses ship trajectory grid mapping coding to uniformly geocode multi-source data, supporting long-time series and large-scale spatial ship prediction. It fully utilizes the complementarity of multi-source data such as AIS, remote sensing, public opinion, radar, and radio in ship target monitoring, while taking into account the confidence of multi-source means, and fuses multi-source data based on Kalman filtering to improve prediction accuracy. According to the encoder-decoder architecture, a sequence-to-sequence ship trajectory prediction model composed of two deep neural network models LSTM is designed to achieve long-time series and large-scale spatial ship trajectory prediction.

[0115] Compared with the existing technology, the technical solution proposed in the present invention fully utilizes the complementarity of multi-source data such as AIS, remote sensing, public opinion, radar, and radio in ship target monitoring, while taking into account the confidence of multi-source means. By mapping and encoding the ship trajectory grid and adopting a sequence-to-sequence ship trajectory prediction model, long-term series and large-scale space ship trajectory prediction can be achieved.

[0116] Through the description of the specific implementation methods, a deeper and more specific understanding of the technical means and effects adopted by the present invention to achieve the intended purpose should be obtained. However, the accompanying drawings are only for reference and illustration purposes and are not intended to limit the present invention.< / eos> < / eos> < / eos> < / bos> < / bos>

Claims

1. A method for predicting maritime target trajectories based on multi-source data, characterized in that: include: In the case of incomplete observation data information, AIS, remote sensing, public opinion, radar, and radio are used to obtain multi-source perception data of maritime targets. Each of the perception data includes the position, speed, confidence, acquisition time, and data source of the maritime target. Based on multi-source perception data, performing trajectory grid mapping encoding on the maritime target; Data fusion is performed on multi-source sensing data within a preset time period in the trajectory grid, and the speed, position, and confidence information of the unique geocoding are generated after data fusion; Obtain input sequence based on the perception data after data fusion and input it into the prediction model; An output sequence of maritime target trajectory prediction is obtained by calculating the prediction model, the prediction model including a first LSTM network and a second LSTM network, the first LSTM network being used to map the perception data after data fusion to a state vector, wherein the state vector summarizes the entire input sequence, the second LSTM network being used to generate an output sequence using the state vector and the input sequence, and the first LSTM network and the second LSTM network both including a diagonally striped rectangular hidden layer representing a network hidden state at each time step; Obtaining a grid position probability distribution vector of the maritime target expected to arrive at a specified time according to the output sequence; According to the constrained grid position probability distribution vector, the predicted position probability distribution map of the sea target is obtained; The data fusion of multi-source sensing data within a preset time period in the trajectory grid includes: Grid the target area according to the specified longitude and latitude, and perform unique geocoding based on the row and column information of each grid; Arrange all the trajectory geocodes of the maritime targets into a text sequence in order, where each code in the text sequence contains speed, position, and confidence information; Add the departure and destination at the beginning and end of the text sequence, and add a colon to the text sequence to generate a complete trajectory encoding text; When mapping grids, if there is a sensed data obtained from multiple sources within a preset time period within a grid, a multi-source data fusion method based on Kalman filtering is used to generate unique geocoded speed, position, and confidence information. The fused data is expressed by weighted averaging as: Where, represents the fusion estimation result of the Cartesian coordinate position of the ship in time period t; represents the fused estimation result of the ship speed in the reference system during the time period t; A represents the fusion estimation result of ship confidence within time period t; i (t) represents the fusion weight matrix; It represents the Cartesian coordinate estimation result of the ship position in time period t when the Kalman filter is used to perform the i-th (1≤i≤l) multi-source data observation; It represents the estimated result of the ship speed in the reference system during the time period t when the Kalman filter is used to perform the i-th (1≤i≤l) multi-source data observation; It represents the built-in confidence estimation result in time period t when using Kalman filtering for the i-th (1≤i≤l) multi-source data observation; The multi-source data fusion weight matrix is ​​determined by the following formula: Where, P jj (t), P ii (t) are all matrices P ij The elements on the main diagonal of (t), P ij (t) represents the cross-covariance matrix of the filtering errors between the i-th and j-th multi-source observation data; The cross-covariance matrix of the filter errors is determined by: P ij (t)=[I5-K i (t)·H]·[Φ·P ij (t-1)·Φ T ]·[I5-K j (t)·H] T ; Where I5 ​​represents the 5×5 unit matrix; K i (t), K j (t) represents the Kalman filter gain matrix subsystem (1≤i≤l, 1≤j≤l) of the i-th and j-th sensors in the specified time period t, respectively; H, Φ represent the constant matrix of the system:

2. The method for predicting maritime target trajectories based on multi-source data according to claim 1, characterized in that: The method further comprises: The acquired multi-source perception data is annotated and stored.

3. A maritime target trajectory prediction device based on multi-source data, characterized in that: include: The data acquisition module is used to acquire multi-source perception data of maritime targets using AIS, remote sensing, public opinion, radar, and radio in the absence of complete observation data information. Each of the perception data includes the position, speed, confidence level, acquisition time, and data source of the maritime target. A grid mapping module is used to perform trajectory grid mapping encoding on the maritime target based on multi-source perception data; The data fusion module is used to fuse the multi-source sensing data within a preset time period in the trajectory grid, and generate unique geocoded speed, position, and confidence information after data fusion; a computing module, configured to obtain an input sequence based on the data-fused perception data and input the input sequence into a prediction model, and to calculate an output sequence for maritime target trajectory prediction using the prediction model, wherein the prediction model includes a first LSTM network and a second LSTM network, wherein the first LSTM network is configured to map the data-fused perception data to a state vector, wherein the state vector summarizes the entire input sequence, and the second LSTM network is configured to generate an output sequence using the state vector and the input sequence, and wherein both the first LSTM network and the second LSTM network include a hidden layer of oblique stripes and rectangles representing a network hidden state at each time step; A vector acquisition module, configured to acquire a grid position probability distribution vector of the maritime target expected to arrive at a specified time according to the output sequence; A prediction module is used to obtain a predicted position probability distribution map of the maritime target based on the constrained grid position probability distribution vector; The data fusion module is specifically used to: Grid the target area according to the specified longitude and latitude, and perform unique geocoding based on the row and column information of each grid; Arrange all the trajectory geocodes of the maritime targets into a text sequence in order, where each code in the text sequence contains speed, position, and confidence information; Add the departure and destination at the beginning and end of the text sequence, and add a colon to the text sequence to generate a complete trajectory encoding text; When mapping grids, if there is a sensed data obtained from multiple sources within a preset time period within a grid, a multi-source data fusion method based on Kalman filtering is used to generate unique geocoded speed, position, and confidence information. The fused data is expressed by weighted averaging as: Where, represents the fusion estimation result of the Cartesian coordinate position of the ship in time period t; represents the fused estimation result of the ship speed in the reference system during the time period t; A represents the fusion estimation result of ship confidence within time period t; i (t) represents the fusion weight matrix; It represents the Cartesian coordinate estimation result of the ship position in time period t when the Kalman filter is used to perform the i-th (1≤i≤l) multi-source data observation; It represents the estimated result of the ship speed in the reference system during the time period t when the Kalman filter is used to perform the i-th (1≤i≤l) multi-source data observation; It represents the built-in confidence estimation result in time period t when using Kalman filtering for the i-th (1≤i≤l) multi-source data observation; The multi-source data fusion weight matrix is ​​determined by the following formula: Where, P jj (t), P ii (t) are all matrices P ij The elements on the main diagonal of (t), P ij (t) represents the cross-covariance matrix of the filtering errors between the i-th and j-th multi-source observation data; The cross-covariance matrix of the filter errors is determined by: P ij (t)=[I5-K i (t)·H]·[Φ·P ij (t-1)·Φ T ]·[I5-K j (t)·H] T ; Where I5 ​​represents the 5×5 unit matrix; K i (t), K j (t) represents the Kalman filter gain matrix subsystem (1≤i≤l, 1≤j≤l) of the i-th and j-th sensors in the specified time period t, respectively; H, Φ represent the constant matrix of the system:

4. The marine target trajectory prediction device based on multi-source data according to claim 3, characterized in that: The device further comprises: The standardization processing module is used to annotate and store the acquired multi-source perception data.

Images