CN116629115A - Bidirectional data driving ship track prediction method and system based on attention mechanism - Google Patents

Abstract

The invention discloses a bidirectional data driving ship track prediction method and a bidirectional data driving ship track prediction system based on an attention mechanism, relates to the technical field of track prediction, and mainly obtains an observation sequence with a forward length of g from an AIS data setThe sequence will be observedInput to a first machine learning model to obtain an intermediate pre-run of length lSequencingAt the same time, the observation sequence with the backward length of g is obtained from the AIS data setThe sequence will be observedInputting the intermediate prediction sequence into a second machine learning model to obtain an intermediate prediction sequence with the length of lIntermediate prediction sequenceIntermediate prediction sequencesAnd splicing to form composite training data serving as the input of a third machine learning model to obtain a prediction result. The method solves the problems of limitation and low prediction precision existing in the prior art for predicting the future track in one direction.

Description

Two-way data-driven ship trajectory prediction method and system based on attention mechanism

technical field

The invention relates to the technical field of trajectory prediction, in particular to a two-way data-driven ship trajectory prediction method and system based on an attention mechanism.

Background technique

AIS system is an important technology to ensure the safety of marine transportation. The static state of the ship is recorded by the AIS system, such as the Maritime Mobile Service Identity (MMSI), and dynamic information such as the ship's position and speed over the ground. AIS provides expert services in support of maritime navigation decision-making, information broadcasting, maritime avoidance and environmental protection. However, ships at sea are vulnerable to problems such as bad weather or improper route planning that affect the efficiency of navigation. Sometimes there are also problems such as ship collision or jamming problems due to improper driving and poor maneuverability of the captain. Therefore, it is necessary to explore the great potential of AIS data in ship risk warning and route optimization.

The current method for predicting ship trajectories mainly uses one-way historical trajectory data to train the model to predict future trajectories, which does not consider two-way historical trajectory data, that is, lacks consideration of the historical trajectory data after the target predicted trajectory. In the traffic scene reproduction prediction tasks in some maritime areas, the existing methods have deficiencies in the data mining of historical multi-directional trajectories of ships, resulting in room for improvement in prediction accuracy.

Linear models work well for trajectory prediction of ships traveling in a straight line. A typical linear model is the Constant Velocity Model (CVM). Representative prediction methods mainly include kinematics-based models, constant-velocity models, Ornstein-Uhlenbeck models, and Kalman filter variants. However, linear models perform weakly when ships need to change course or speed and turn to sail under certain conditions. Therefore, in order to overcome the weakness of linear models, some researchers have studied nonlinear models to improve prediction accuracy. Typical nonlinear trajectory prediction models are based on machine learning and deep learning methods. Common machine learning methods include trajectory clustering, support vector machines, Gaussian models, etc. However, the accuracy of machine learning-based methods relies on labels and physical knowledge for recognition. Methods based on deep learning can capture more features of the input data to improve trajectory prediction accuracy. Common deep learning prediction methods such as artificial neural networks, recurrent neural networks and their variants. However, existing methods often only consider the effectiveness of training and learning on one-way historical trajectory data to achieve target prediction trajectory, and there is a lack of research on bidirectional historical trajectory data. In reconstructing traffic scenarios in historical sea areas, how to use the characteristics of bidirectional historical trajectory data to improve the accuracy of trajectory prediction is still a difficult point.

Contents of the invention

One of the purposes of the present invention is to provide a two-way data-driven ship trajectory prediction method based on attention mechanism to solve the problems of insufficient feature extraction and low prediction accuracy of two-way historical trajectory data in the prior art.

In order to solve the above technical problems, the present invention adopts the following technical solutions: a two-way data-driven ship track prediction method based on attention mechanism, comprising the following steps:

S1. Obtain an observation sequence with forward length g from the AIS dataset This observation sequence is input to the first machine learning model to obtain an intermediate prediction sequence of length l /> At the same time, obtain the observation sequence with backward length g from the AIS dataset /> Input the observation sequence to the second machine learning model to obtain an intermediate prediction sequence of length l />

S2, the intermediate prediction sequence for the first machine learning model and the intermediate prediction sequence of the second machine learning model /> Splicing to form a composite training data;

S3. Using the composite training data as an input of the third machine learning model to obtain a prediction result.

Preferably, the acquisition process of the first machine learning model, the second machine learning model and the third machine learning model includes: collecting ship time series trajectory data as an AIS data set, and dividing the AIS data set into a training set and a test set; Using the training set as the input of the forward sub-block and the backward sub-block model, training the forward sub-block and the backward sub-block model to obtain the first machine learning model and the second machine learning model; during the training process The obtained intermediate outputs of the optimal forward sub-block and backward sub-block models are combined into a composite fusion training set, and the composite fusion training set is input into the fusion prediction block, and the fusion prediction block is trained to obtain the third machine learning Model.

More preferably, the first machine learning model is composed of a GRU neural network and an attention mechanism; the second machine learning model is composed of a BiGRU neural network and an attention mechanism; the third machine learning model is a multi-layer perceptron network.

More preferably, in step S1, the first machine learning model inputs the forward observation sequence Mapped to an output sequence, an intermediate prediction sequence of length l /> The acquisition process includes:

The sequence of forward observations in the dataset The elements of are sequentially input into the first machine learning model, where t represents the position in the current forward trajectory sequence;

Update the hidden sequence of the first machine learning model according to in per element /> Represents the feature extracted from the t-th time sample point in the intermediate feature state with a length of g in the forward trajectory sequence;

Represents the forward attention distribution coefficient of the tth sample point in the forward trajectory sequence, calculated as follows: Represents the attention weight scoring value of the tth sample point in the forward trajectory sequence, calculated as follows: /> is the network parameter of the additive calculation process, /> is the spatio-temporal feature code extracted by the GRU network from the input sample point at the tth moment in the input time series length g trajectory;

element Represents the spatio-temporal feature code extracted from the input sample point at the tth moment in the input time series length of g trajectories through the GRU network, /> each /> Represents the spatio-temporal features extracted from the last sample point at the tth moment in the input time series length g trajectory, /> Indicates the trajectory point of the sequence input to the GRU neural network at the current moment, that is, the input sample point of the first machine learning model, and θ _gru indicates the parameter value in the process of mapping each input to output;

θ _gru represents the parameter set in the GRU network mapping process in the first machine learning model: Indicates the i-th timing point in the forward trajectory sequence of length g input to the first machine learning model, L represents the quantization error, N represents the total number of training samples, /> Indicates the jth timing point in the mapping output sample with length l, and θ represents the parameter value of each input to output mapping process;

M _g,l represents the forward input sequence at a given length g Next, predict the output sequence Y _l of length l, so as to maximize the conditional probability: M _g,l = argmax _Y p(Y _l |X _g ), p(Y _l |X _g ) means that given g observation sequences The probability that X _g is mapped to the future l predicted trajectory sequence Y _l ;

Feed a vector consisting of the outputs of all hidden layers into a fully connected layer to obtain an intermediate prediction sequence of length l

In step S1, the second machine learning model takes the input backward observation sequence Mapped to an output sequence, an intermediate prediction sequence of length l /> The acquisition process includes:

The sequence of backward observations in the dataset The elements of are sequentially input into the second machine learning model, where t represents the position in the current backward trajectory sequence;

Update the hidden sequence of the second machine learning model according to in per element /> Represents the feature extracted from the t-th time sample point in the intermediate feature states with a backward input sequence length of g;

Represents the forward attention distribution coefficient of the t-th sample point in the backward trajectory sequence, calculated as follows: Represents the attention weight scoring value of the tth sample point in the backward trajectory sequence, calculated as follows: /> v, W, U are the network parameters of the additive calculation process, /> The spatio-temporal feature code extracted by the BiGRU network from the input sample point at the tth moment in the input time series length g trajectory;

For the fusion of the hidden states of the forward layer and the reverse layer by the BiGRU network, the calculation formula is as follows: Where W is the weight matrix in the network, and b is the bias value;

in Represents the trajectory point of the backward sequence input at the current t moment, that is, the input sample point of the second machine learning model;

Indicates the parameter set in the forward layer mapping process in the second machine learning model: Indicates the i-th timing point in the backward trajectory sequence of length g input to the forward layer of the second machine learning model. /> Represents the parameter set during the reverse layer mapping process in the second machine learning model: Indicates the i-th timing point in the backward trajectory sequence of length g input to the reverse layer, L represents the quantization error, N represents the total number of training samples, /> Indicates the jth timing point in the mapping output sample with length l, and θ represents the parameter value of each input to output mapping process;

Feed a vector consisting of the outputs of all hidden layers into a fully connected layer to obtain an intermediate prediction sequence of length l

More preferably, in step S3, the third machine learning model uses the input splicing intermediate prediction sequence Mapped to an output sequence, predicted output of length l /> The acquisition process includes:

will splice the intermediate prediction sequence The elements of are sequentially input into the third machine learning model, where t represents the t-th intermediate prediction pair in the current intermediate prediction sequence;

Obtain the hidden sequence of the third machine learning model according to the following formula in Indicates the intermediate prediction splicing value of the input data at the current time t, that is, the input data of the third machine learning model, /> Indicates the output hidden state value obtained at the previous moment, and θ indicates the parameter value in the process of mapping each pair of input concatenated values to the output;

The output of the jth hidden layer according to the following formula Among them, the hidden state h ₁ of the first hidden layer is /> σ is the activation function of the 1st hidden layer, l is the total number of inputs, b _t is the bias of the 1st hidden layer, w _t is the weight of the connection layer, σ _j is the non-linear activation of the jth hidden layer with a learnable parameter θ function, b _j is the bias of the jth hidden layer, w _j is the weight of the jth connection layer;

Input the vector consisting of the outputs of all hidden layers into the fully connected layer to obtain a predicted output sequence of length l

In addition, the present invention also provides a two-way data-driven ship trajectory prediction system based on attention mechanism, which includes:

The data acquisition module is used to acquire the observation sequence whose forward length is g from the AIS data set and a backward sequence of observations of length g />

A data processing module for converting the observation sequence Input to the first machine learning model to obtain an intermediate prediction sequence of length l /> and used to watch the sequence /> Input to the second machine learning model to obtain an intermediate prediction sequence of length l />

Data splicing module for the intermediate prediction sequence of the first machine learning model and the intermediate prediction sequence of the second machine learning model /> Splicing to obtain a composite training data;

The result prediction module is used to use the composite training data as an input of the third machine learning model to obtain a prediction result.

The system operates using the attention mechanism-based two-way data-driven ship trajectory prediction method described above.

Compared with the existing technology, the present invention can not only extract the features of bidirectional historical time series trajectory data, but also introduce the attention mechanism to assign weights to the output features of the neural network to enhance the model's ability to learn and perceive the features of the data, and skillfully use multi-layer perception The machine network performs fusion learning of the intermediate output features of different sub-models, which greatly improves the feature learning ability of the model for time-series trajectory data and improves the prediction accuracy of the model.

Description of drawings

Fig. 1 is the principle schematic diagram of the prediction method of the embodiment of the present invention;

Fig. 2 is a schematic flow chart of a prediction method according to an embodiment of the present invention;

Fig. 3 is the GRU network structural diagram in the embodiment of the present invention;

Fig. 4 is the BiGRU network structural diagram in the embodiment of the present invention;

Fig. 5 is a multi-layer perceptron (MLP) network structure diagram in the embodiment of the present invention;

FIG. 6 is a schematic diagram of results of searching for optimal input and output step sizes in terms of RMSE performance in an embodiment of the present invention.

Detailed ways

In order to facilitate the understanding of those skilled in the art, the present invention will be further described below in conjunction with the embodiments and accompanying drawings, and the contents mentioned in the embodiments are not intended to limit the present invention.

The machine learning model training process adopted by the embodiment of the present invention is as follows: In this embodiment, the existing historical track data in the AIS data set is used, and it is divided into two parts: a training set and a test set. Those skilled in the art should understand that when this method is used in practice, the test set should be obtained from the time series data set to be predicted by the ship trajectory. Since this embodiment is only used to illustrate and verify the method, it can be The time series data in the existing dataset is used as the test set.

According to the download of 164 ship data from the official website, a total of 940943 ship time series trajectory data collected by the AIS receiver equipment are used as the AIS data set, and the data set is divided into a training set and a test set according to the ratio of 7:3.

In the training set, after preprocessing the trajectory segment corresponding to each ship in the data set, it is divided into a forward input trajectory sequence and a backward input trajectory sequence according to the front and rear relationship of the target preprocessing real label, and then input to the first machine learning model respectively and the second machine learning model to extract the deep features of the time series data, use the tanh function to predict at the output layer, use the root mean square error function to calculate the error between the predicted trajectory label and the real trajectory label, and calculate the neural network through the back propagation algorithm. The weights and biases of each layer of the network, iteratively train the neural network in the first machine learning model and the second machine learning model until the loss function converges, and obtain the optimal training of the first machine learning model and the optimal training of the second machine learning model, and at the same time obtain the intermediate prediction sequence of the training data of the training set.

Splicing the intermediate prediction sequence of the optimally trained first machine learning model and the second machine learning model, and then inputting the spliced intermediate prediction sequence vector into the fusion prediction block, using the fusion prediction block multi-layer perceptron network to splicing state network Carry out learning and deep feature extraction, use the tanh function in the output layer to predict, use the root mean square error function to calculate the error between the predicted trajectory label and the real trajectory label, and calculate the weight and bias of each layer of the neural network through the back propagation algorithm The neural network in the fusion prediction block is continuously iteratively trained until the loss function converges, and the model of the optimal training fusion prediction block is obtained, and finally the overall first, second and third machine learning models are saved as a whole.

Example 1

The embodiment of the present invention utilizes the GRU neural network and the attention mechanism network obtained after training, the BiGUR neural network and the attention mechanism network, and the multi-layer perceptron network to realize the feature extraction and analysis of the forward and backward historical trajectory data by the prediction model. better predictive performance.

Figure 2 shows the structural overview of the proposed two-way data-driven ship trajectory prediction method, where the forward sub-block, backward sub-block and fusion prediction block form an overall prediction framework for trajectory prediction. The key features of the two-way data-driven forecasting method of this embodiment are as follows:

First, the two-way data-driven prediction method of this embodiment inputs the observed g existing trajectory points, and outputs a future prediction of l trajectory points after learning and fusion of three key blocks. Wherein, the values of g and l can be adjusted according to actual service requirements.

Secondly, the two-way data-driven prediction method of this embodiment constructs prediction sub-blocks based on the same data set, and uses different neural network structures in the prediction sub-blocks to perform feature learning and prediction on information in the trajectory in the same data set. For example, the forward sub-block and the backward sub-block are used to learn the forward and backward trajectory sequence segments and the intermediate prediction sequence in the data set, so that the feature learning and extraction of the comprehensive trajectory can be performed on the data set using the bidirectional data trajectory, and Using the multi-layer perceptron network to splice and fuse the intermediate prediction sequence to obtain the final prediction effect. The two-way data-driven trajectory prediction process based on the attention mechanism includes the following steps, as shown in Figure 2:

Step (1): First, the forward sub-block receives the forward sequence of length g observed from the data set, and inputs the observed sequence into the GRU network and the attention mechanism network for learning, wherein the data included in each sequence point in the data set The types are: time, latitude, longitude, course and speed.

Step (2): Secondly, the backward sub-block receives the backward sequence of length g observed from the data set, and inputs the observed sequence into the BiGRU network and the attention mechanism network for learning, wherein each sequence point in the data set includes The data types are: time, latitude, longitude, course and speed.

Step (3): Then, after the GRU network and the attention mechanism network structure training and learning, the state sequence composed of the output of the hidden layer is obtained and input to the fully connected layer to obtain an intermediate prediction sequence of length l

Step (4): Similarly, after the training and learning of the BiGRU network and the attention mechanism network structure, the state sequence composed of the output of the hidden layer is input to the fully connected layer, and an intermediate feature state output sequence of length l is obtained

Step (5): Then, splicing the intermediate prediction sequences of the forward sub-block and the backward sub-block to form a new training data This training data will serve as input to the fused prediction block.

Step (6): The new training data Input to the fusion prediction block, the fusion prediction block uses a multi-layer perceptron network for learning, and finally outputs the overall prediction model result />

The two-way data-driven ship trajectory prediction method based on the attention mechanism designed in this embodiment can extract the two-way historical trajectory data to enhance the feature learning and perception ability of the two-way historical trajectory data, and introduce the attention mechanism to realize the behavior in the trajectory data. The weight assignment of the feature state enhances the relationship between the model and different data dimensions in the time series data, enhances the relationship between the model and the different dimensions of the time series data and improves the accuracy of the prediction model.

The forward sub-block adopts GRU neural network and attention mechanism network. The Gated Recurrent Neural Network (GRU) is a variant of the Recurrent Neural Network (RNN). As shown in Figure 3, the GRU network contains two structures: a reset gate and an update gate. Among them, the reset gate is used to reduce the irrelevant information in the previous unit, and the update gate is used to determine how much information from the previous unit needs to be passed to the next unit.

This example uses the forward sequence of observations defining the dataset Represents a general sequence of length, and t represents the position in the current trajectory sequence. input sequence /> Sequentially calculate the hidden vector sequence through the GRU network The specific GRU model is controlled by the following formulas (1)-(4)

Where σ represents the sigmoid activation function, tanh is the hyperbolic tangent function, r _t , z _t represent the output of the reset gate and update gate, Indicates the candidate output and the actual output, /> Represents element-wise multiplication, Us and Ws are weight matrices, and bs is a bias term.

for the output vector Do a fully connected layer calculation and output a feature state output sequence of length l where each element /> Indicates the spatio-temporal feature data extracted from the trajectory sequence of the input GRU neural network for encoding the tth component of the sequence. /> is the output set of the hidden state fully connected layer connected to the GRU neural network. The hidden layer state dimension is q, and q is the same as the dimension of the model input data.

The output vector to the GRU network Use the attention mechanism network to learn the weight assignment and assignment, and get the prediction output of the forward sub-block /> The specific attention mechanism network is calculated by the following formulas (5)-(6)

Among them, v, W, U are the network parameters of the additive calculation process, is the spatio-temporal feature code extracted by the GRU network from the input sample point at the tth moment in the t-th trajectories in the forward trajectory sequence, /> Indicates the attention weight scoring value of the tth sample point in the forward trajectory sequence, /> Indicates the forward attention distribution coefficient of the t-th sample point in the forward trajectory sequence.

for the output vector Do another fully connected layer calculation and output an intermediate prediction sequence of length l/>

The backward sub-block adopts BiGRU neural network and attention mechanism network. As shown in Figure 4, the BiGRU model has an additional set of GRU models propagated in the opposite direction than the unidirectional GRU model, which enables BiGRU to explore the past and future information in the observed backward sequence, thus providing more effective prediction results. BiGRU will input the backward sequence Mapped to two output sequences, the forward hidden sequence /> backward hidden sequence /> And operate through the following formulas (8)-(10):

Each of the GRU functions is a recurrent network of formulas (1)-(4), and the GRU function performs a nonlinear conversion of the input time-series trajectory vector into the corresponding GRU hidden state, Indicates the trajectory point of the backward sequence input at the current time t, Indicates the parameter set in the forward layer mapping process in the BiGRU network, /> Represents the parameter set in the reverse layer mapping process in the BiGRU network, W is the weight matrix in the network, and b is the bias value. The hidden state of the forward layer and the hidden layer of the reverse layer calculated by the two unidirectional GRU networks through formulas (8)-(10) are spliced into a compact bidirectional representation, and finally the feature state output vector of the BiGRU block is obtained.>

for the output vector Do a fully connected layer calculation and output a feature state output sequence of length l /> where each element /> Represents the spatiotemporal feature data extracted from the trajectory sequence whose t-th component of the sequence is input into the BiGRU neural network. /> is the output set of the hidden state fully connected layer connected to the BiGRU neural network. The hidden layer state dimension is q, and q is the same as the dimension of the model input data.

Output vector to BiGRU neural network Use the attention mechanism network for weight assignment and assignment learning, and get the predicted output of the backward sub-block/> The specific attention mechanism network is calculated by the following formulas (11)-(13)

Among them, v, W, U are the network parameters of the additive calculation process, is the spatio-temporal feature code extracted by the BiGRU network from the input sample point at the tth moment in the backward trajectory sequence with a length of g trajectories, /> Indicates the attention weight scoring value of the tth sample point in the backward trajectory sequence, /> Indicates the backward attention distribution coefficient of the t-th sample point in the backward trajectory sequence.

for the output vector Do another fully connected layer calculation and output an intermediate prediction sequence of length l/>

In this embodiment, the intermediate prediction sequence of the forward sub-block and the backward sub-block and /> splicing to form the input sequence of the final fused prediction block />

The fused prediction block employs a multi-layer perceptron network. A multilayer perceptron network (MLP) is a neural network with a supervised learning technique using the backpropagation method. The multilayer perceptron network sequentially reads the intermediate prediction sequence consisting of the output of the forward block and the backward block Each pair of concatenated elements in and updates the internal hidden state according to the following formula (14):

where σ is the activation function of the first hidden layer, Indicates the splicing state at the tth moment in the input variable, l is the total number of inputs, b _t is the bias of this layer, and w _t is the weight of the connection layer. Then, the following hidden layers update the internal hidden state by the following formula:

where _σj is the non-linear activation function of the jth hidden layer with θ learnable parameters, _bj is the bias of the jth hidden layer, w _tj is the weight of the jth layer connection, Represents the output hidden state value obtained at the previous moment, and θ represents the parameter value in the process of mapping each pair of input concatenated values to the output. Finally, an additional output layer accepts the hidden state in Equation (15) > As input, to make predictions in sequence.

In the fusion prediction block, the multi-layer perceptron (MLP) network is shown in Figure 5. The training process of the MLP network structure is to input the sequence is mapped to an output sequence, the hidden sequence /> Calculated by formula (16):

Among them, MLP represents the operation process of formula (14) and formula (15), using output vector to the hidden layer Do another fully connected layer operation and output a sequence of length l. Finally, the predicted output sequence of the fused prediction block is calculated by formula (17)

where W _y and b _y are the mapping of the MLP output to the next predicted position The trainable parameters of the neural network. /> Indicates the output value of the tth trajectory sequence point in the output sequence of length l.

The following describes the prediction scenario and application model of the two-way data-driven prediction method based on the attention mechanism of this embodiment, and then analyzes the effectiveness of the method in this scenario.

Prediction model:

(1) It is assumed that the user knows the trajectory observation point of the ship at sea for a certain period of time, and the trajectory observation sequence comes from the AIS (Automatic Identification System) data set.

(2) The user can input the known observation forward and backward sequence into the two-way data-driven prediction method based on the attention mechanism, which will pass the forward and backward observation sequence through the forward sub-block, backward The sub-blocks are fused with the prediction block to obtain the final sequence of future predicted trajectories.

(3) In the two-way data-driven prediction method based on the attention mechanism of this embodiment, the trajectory points in the input observation sequence are not allowed to be less than 2, and the trajectory points in the output prediction sequence are not allowed to be less than 1.

Effectiveness Analysis:

This section analyzes the effectiveness of the method provided in this embodiment by taking the ship trajectory sequence in the west coast of the United States as an example.

(1) Compared with existing work. This example compares this method with a naive LSTM, a GRU network, and three state-of-the-art existing works. Compared with the existing research, the RMSE performance of the work in this embodiment is better than 85.42% on average. Among them, the RMSE performance of this embodiment can be improved by at least 30.77% compared with the existing research, and the RMSE performance of this embodiment can be improved by up to 99.15% compared with the existing research. Therefore, the two-way data-driven prediction method based on the attention mechanism proposed in this embodiment has good prediction accuracy. In addition, this embodiment explores that in the experiment of predicting target trajectory sequences of different lengths, the method proposed in this embodiment is superior to existing research in short-term, medium-term and long-term target trajectory prediction tasks.

(2) Study the optimal input sequence step size and prediction step size parameters, and further explore the optimal parameters of the proposed method. In this embodiment, each record in the data set represents the track point of the ship on the sea area. In this embodiment, the number of records input into the model in the data set is used as the input step size, and the number of records predicted by the model in this embodiment is output as the output. step size. This example explores the optimal input step size and output step size relationship. As shown in FIG. 6 , in the RMSE performance of this embodiment, it is found that when the input step size of the method of this embodiment is 4, 5, 8, 9, the prediction effect of this embodiment reaches a better level. Therefore, the present invention can set a reasonable number of input step size and output step size in actual work to achieve the best working effect.

Example 2

This embodiment relates to a two-way data-driven ship trajectory prediction device based on an attention mechanism. The ship prediction device includes a processor and a memory, and a computer program is stored on the memory. When the computer program is executed through processing, it is used to implement the embodiment 1. An Attention Mechanism-Based Bidirectional Data-Driven Ship Trajectory Prediction Method.

Specifically, the processor can use Intel(R) Core(TM) i7-1165G7@2.80GHz processor, 16GB memory, and use python3.6 to perform software programming on the Keras framework.

The purpose of the two-way data-driven ship trajectory prediction device based on the attention mechanism provided in this embodiment is to implement the two-way data-driven ship trajectory prediction method based on the attention mechanism mentioned in Embodiment 1. Therefore, the technical effects of Embodiment 1, the two-way data-driven ship trajectory prediction device based on the attention mechanism provided by this embodiment, will not be repeated here.

In order to allow those of ordinary skill in the art to more easily understand the improvement of the present invention compared to the prior art, some drawings and descriptions of the present invention have been simplified, and the above-mentioned embodiments are preferred implementation solutions of the present invention, in addition In addition, the present invention can also be realized in other ways, and any obvious replacements are within the protection scope of the present invention without departing from the concept of the technical solution.

Claims (8)

1. The bidirectional data driving ship track prediction method based on the attention mechanism is characterized by comprising the following steps of:

s1, acquiring an observation sequence with a forward length of g from an AIS data setInputting the observation sequence into the firstA machine learning model, obtaining the intermediate prediction sequence +.>At the same time, the observation sequence with the backward length g is acquired from the AIS data set>Inputting the observation sequence into a second machine learning model to obtain an intermediate prediction sequence with the length of l

S2, middle prediction sequence of first machine learning modelAnd an intermediate prediction sequence of a second machine learning modelSplicing to form composite training data;

s3, taking the composite training data as the input of a third machine learning model to obtain a prediction result.

2. The attention mechanism based bi-directional data driven vessel trajectory prediction method of claim 1, wherein the acquiring process of the first, second and third machine learning models comprises:

the acquired ship time sequence track data are used as AIS data sets, and the AIS data sets are divided into training sets and testing sets;

training the forward sub-block and the backward sub-block model by taking the training set as the input of the forward sub-block and the backward sub-block model to obtain a first machine learning model and a second machine learning model;

and combining the intermediate outputs of the optimal forward sub-block and the backward sub-block model obtained in the training process into a composite fusion training set, inputting the composite fusion training set into a fusion prediction block, and training the fusion prediction block to obtain a third machine learning model.

3. The attention-based bi-directional data driven vessel trajectory prediction method according to claim 1 or 2, wherein: the first machine learning model is composed of a GRU neural network and an attention mechanism; the second machine learning model is composed of a BiGRU neural network and an attention mechanism; the third machine learning model is a multi-layer perceptron network.

4. A bi-directional data driven vessel trajectory prediction method based on an attention mechanism as claimed in claim 3, wherein: in step S1, the first machine learning model observes an input forward observation sequenceMapping to an output sequence, intermediate prediction sequence of length l +.>The acquisition process of (1) comprises:

forward observation sequence in datasetSequentially inputting elements of (a) into a first machine learning model, wherein t represents a position in a current forward track sequence;

updating hidden sequences of a first machine learning model according toWherein the method comprises the steps ofEach element->Representing features extracted from sample points at t time in g middle feature states in length in a forward track sequence;

the forward attention distribution coefficient representing the t-th sample point in the forward trajectory sequence is calculated as follows: the attention weighting score representing the t-th sample point in the forward trajectory sequence is calculated as follows: />v, W, U are network parameters of the additive calculation process, < >>Space-time feature codes extracted from input sample points at the t-th moment in g tracks of input time sequence length for the GRU network;

element(s)Representing a spatiotemporal feature code extracted from an input sample point at time t in a g-track of input timing length over a GRU network,/>Each->Representing a spatiotemporal feature extracted from the last sample point of the g tracks of length of input timing, t>Track points representing sequences input to the GRU neural network at the current moment, namely input sample points of the first machine learning model, theta _gru Representing the parameter values in each input-to-output mapping process;

θ _gru representing a set of parameters in a GRU network mapping process in a first machine learning model: representing the ith time sequence point in the forward track sequence with length g input into the first machine learning model, L representing quantization error, N representing total number of training samples, +.>A j-th time sequence point in a mapping output sample with the length of l is represented, and theta represents a parameter value in each input-to-output mapping process;

M _g,l representing a forward input sequence of g at a given lengthNext, the output sequence Y of length l is predicted _l Thereby maximizing the conditional probability: m is M _g,l ＝argmax _Y p(Y _l |X _g )，p(Y _l |X _g ) Representing a given g observation sequences X _g Mapping to future l predicted track sequences Y _l Probability of (2);

inputting vectors composed of the outputs of all hidden layers into a full-connection layer to obtain an intermediate prediction sequence with length of l

5. The attention-based mechanism of claim 3The bidirectional data driving ship track prediction method is characterized by comprising the following steps of: in step S1, the second machine learning model observes the sequence of input backward observationsMapping to an output sequence, intermediate prediction sequence of length l +.>The acquisition process of (1) comprises:

backward observation sequence in datasetSequentially inputting elements of (a) into a second machine learning model, wherein t represents a position in a current backward track sequence;

updating hidden sequences of a second machine learning model according toWherein the method comprises the steps ofEach element->Representing features extracted from a t-th time sample point in a backward input time sequence state with g intermediate feature states;

the forward attention distribution coefficient representing the t-th sample point in the backward trajectory sequence is calculated as follows: the attention weighting score representing the t-th sample point in the backward trajectory sequence is calculated as follows: />v, W, U are network parameters of the additive calculation process, < >>Space-time feature codes extracted from input sample points at the t-th moment in g tracks of the input time sequence length for the BiGRU network;

for the fusion of hidden states of a forward layer and a reverse layer by a BiGRU network, the calculation formula is as follows:wherein W is a weight matrix in the network, and b is a bias value;

wherein the method comprises the steps of Track points of a backward sequence input at the current t moment are represented, namely input sample points of a second machine learning model;

representing a set of parameters in a forward layer mapping process in a second machine learning model: representing input to a second machineAnd learning an ith time sequence point in a backward track sequence with the length g in a forward layer of the model. />Representing a set of parameters in a reverse layer mapping process in a second machine learning model: represents the ith time sequence point in the backward track sequence with length g input into the backward layer, L represents quantization error, N represents total number of training samples, +.>A j-th time sequence point in a mapping output sample with the length of l is represented, and theta represents a parameter value in each input-to-output mapping process;

inputting vectors composed of the outputs of all hidden layers into a full-connection layer to obtain an intermediate prediction sequence with length of l

6. A bi-directional data driven vessel trajectory prediction method based on an attention mechanism as claimed in claim 3, wherein: in step S3, the third machine learning model concatenates the input intermediate prediction sequenceMapping to an output sequence, predicted output +.>The acquisition process of (1) comprises:

splice intermediate predicted sequencesSequentially inputting elements of (a) into a third machine learning model, wherein t represents a t-th intermediate prediction pair in the current intermediate prediction sequence;

obtaining a hidden sequence of the third machine learning model according toWherein the method comprises the steps of Intermediate predictive splice value representing the current t-moment input data, i.e. the input data of the third machine learning model,/->Representing the output hiding state value obtained at the last moment, wherein θ represents the parameter value in the process of mapping each pair of input splicing values to the output;

output according to the following jth hidden layer Wherein, the hidden state h of the 1 st hidden layer ₁ Is->Sigma is the activation function of the 1 st hidden layer, l is the total number of inputs, b _t Is the bias of the 1 st hidden layer, w _t Is the weight of the connection layer, sigma _j Is a nonlinear activation function with a leachable parameter theta for the j-th hidden layer, b _j Is the bias of the j-th hidden layer, w _j Is the weight of the j-th connecting layer;

will consist of the output of all hidden layersIs input into the full-connection layer to obtain a predicted output sequence with the length of l

7. A bi-directional data-driven marine vessel trajectory prediction system based on an attention mechanism, comprising:

a data acquisition module for acquiring an observation sequence with a forward length of g from the AIS data setAnd a viewing sequence having a backward length g +.>A data processing module for processing the observation sequence +.>Inputting into a first machine learning model to obtain an intermediate prediction sequence with length of l>And is used to observe the sequence->Inputting into a second machine learning model to obtain an intermediate prediction sequence with length of l>A data stitching module for mid-prediction sequence of the first machine learning model>And intermediate prediction sequence of the second machine learning model +.>Splicing to obtain composite training data;

and the result prediction module is used for taking the composite training data as the input of a third machine learning model to obtain a predicted result.

8. The attention-based bi-directional data driven vessel trajectory prediction system of claim 7, wherein: a bi-directional data driven vessel trajectory prediction method based on an attention mechanism as claimed in any one of claims 1 to 6.