Ship trajectory prediction method based on conditional recurrent neural network

Abstract

The present invention provides a ship trajectory prediction method, system, storage medium and electronic device based on conditional recurrent neural network, and relates to the field of trajectory prediction technology. In the present invention, firstly, according to the information of the attraction point and repulsion point of the current trajectory, an artificial potential field model is introduced to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge; then the magnitude of the resultant force and the coordinates of the resultant force point are correspondingly used as the first non-time series characteristic condition and the second non-time series characteristic condition, and the conditional recurrent neural network model and the time series characteristic are embedded to form mixed information to obtain the best predicted trajectory sequence of the ship. On the one hand, it solves the problem of insufficient training data in the actual track prediction application scenario; on the other hand, the important prior knowledge for track prediction is added to the time series prediction process, giving full play to the role of time series characteristics and non-time series characteristic conditions with prior knowledge in ship track prediction, and improving the accuracy and efficiency of ship track prediction in the face of complex marine environments.

Description

Technical Field

[0001] The present invention relates to the technical field of trajectory prediction, and in particular to a ship trajectory prediction method, system, storage medium and electronic device based on a conditional recurrent neural network. Background Art

[0002] At present, the research on the moving target trajectory prediction problem has realized a variety of classic trajectory prediction methods, which are mainly divided into: ship trajectory prediction based on statistics and ship dynamics equations, ship trajectory prediction based on machine learning, ship trajectory prediction based on deep learning algorithms, etc.

[0003] Among them, the most common are five machine learning methods, namely Kalman Filtering (KF), support vector machine (SVM), back propagation network (BP), Gaussian process regression (GPR) and random forest (RF), and seven deep learning methods, namely recurrent neural network (RNN), long short-term memory (LSTM), gated recurrent unit (GRU), bidirectional long short-term memory (Bi-LSTM), sequence to sequence (Seq2seq), bidirectional gated recurrent unit (Bi-GRU) and Transformer neural network.

[0004] The above traditional machine learning starts with a specific problem for which there is training data. The problem can usually be formulated as a regression task where the input X must be mapped to the output Y. The training data is generated or collected and then processed by the algorithm in an attempt to approximate the unknown mapping. The main processing flow consists of four main parts, namely the training data, the hypothesis set, the learning algorithm, and the final hypothesis. However, this purely data-driven approach may reach its limits or lead to unsatisfactory results when it comes to medium- and long-term predictions. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the shortcomings of the prior art, the present invention provides a ship trajectory prediction method, system, storage medium and electronic device based on a conditional recurrent neural network, which solves the technical problem that a purely data-driven method may reach its limit or lead to unsatisfactory results in medium- and long-term predictions.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A ship trajectory prediction method based on conditional recurrent neural network, comprising:

[0010] According to the information of the attraction point and repulsion point of the current trajectory, an artificial potential field model is introduced to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge;

[0011] The magnitude of the resultant force and the coordinates of the resultant force point are correspondingly used as the first non-time series feature condition and the second non-time series feature condition, and the conditional recurrent neural network model and the time series features are embedded in the mixed information to obtain the best predicted trajectory sequence of the ship; wherein the time series features are the longitude and latitude features of the ship in a pre-acquired similar trajectory data set.

[0012] Preferably, the step of obtaining the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge comprises:

[0013] The target point is used as the attraction point generated by external information for the current trajectory; and the non-navigable point generated by meteorological and geographical factors and the prohibited navigation area point generated by other factors are used as the repulsion point generated by external information for the current trajectory;

[0014] According to the coordinates of the ship's position and the coordinates of the target point, a gravitational potential field function is constructed, and the negative gradient of the gravitational potential field function is solved as the gravitational force exerted on the ship at the corresponding position;

[0015] According to the ship position coordinates and the obstacle position coordinates, a repulsive potential field function is constructed, and the negative gradient of the repulsive potential field function is solved as the repulsive force exerted on the ship at the corresponding position;

[0016] The resultant force is calculated according to the gravitational force and repulsive force received by the ship at the corresponding position, so as to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge.

[0017] Preferably, the gravitational force exerted on the ship at the corresponding position is expressed as:

[0018]

[0019] in, is the gravitational potential field function; is the proportional gain factor of the gravitational potential field; X is the ship position coordinates; X g is the target point position coordinate; m is the gravitational potential field factor; is the Euclidean distance between the ship and the target point;

[0020] is the gravitational force exerted on the ship at the corresponding position; for The negative gradient of

[0021] The repulsive force on the ship at the corresponding position is expressed as:

[0022]

[0023] in, is the repulsive potential field function; is the proportional gain factor of the repulsive potential field; is the obstacle position coordinate; is the Euclidean distance between the ship and the obstacle;

[0024] is the repulsive force on the ship at the corresponding position; for The negative gradient of

[0025] The resultant force is expressed as:

[0026]

[0027] Among them, the combined Contains the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge.

[0028] Preferably, the conditional recurrent neural network model includes an input node, seven hidden layers and an output node, and the seven hidden layers are respectively a GRU layer, a Dropout layer and five Dense layers.

[0029] Preferably, embedding the first non-time series feature condition and the second non-time series feature condition into the conditional recurrent neural network model and the time series features in the pre-acquired similar trajectory data set to form mixed information means:

[0030] The first non-time series feature condition and the second non-time series feature condition are linearly transformed after being standardized through corresponding Dense layers, and the two output vectors are stacked respectively, and then output as a one-dimensional vector through a Dense layer;

[0031] The one-dimensional vector is input into the GRU layer and assigned to the initial state of the conditional recurrent neural network. At the starting moment, the first non-time series feature condition, the second non-time series feature condition and the time series feature output by the input node form mixed information.

[0032] Preferably, the step of obtaining the best predicted trajectory sequence of the ship comprises:

[0033] The GRU layer outputs the final memory of the current moment based on the mixed information, and sequentially passes through a Dense layer, a Dropout layer, and a Dense layer to predict the ship position at the next moment at the output node;

[0034] The predicted value is used as a known time series feature to obtain the corresponding first non-time series feature condition and the second non-time series feature condition, and cyclic prediction of subsequent moments is performed based on the conditional recurrent neural network to ultimately obtain the optimal predicted trajectory sequence.

[0035] A ship trajectory prediction system based on conditional recurrent neural network, comprising:

[0036] The acquisition module is used to introduce the artificial potential field model according to the information of the attraction point and repulsion point of the current trajectory, and obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge;

[0037] A prediction module is used to correspond the magnitude of the resultant force and the coordinates of the resultant force point as the first non-time series feature condition and the second non-time series feature condition, and embed the conditional recurrent neural network model and the time series feature to form mixed information to obtain the best predicted trajectory sequence of the ship; wherein the time series feature is the longitude and latitude feature of the ship in a pre-acquired similar trajectory data set.

[0038] A storage medium stores a computer program for ship trajectory prediction based on a conditional recurrent neural network, wherein the computer program enables a computer to execute the ship trajectory prediction method as described above.

[0039] An electronic device, comprising:

[0040] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include instructions for executing the ship trajectory prediction method as described above.

[0041] (III) Beneficial effects

[0042] The present invention provides a ship trajectory prediction method, system, storage medium and electronic device based on conditional recurrent neural network. Compared with the prior art, it has the following beneficial effects:

[0043] In the present invention, firstly, according to the information of the attraction point and repulsion point of the current trajectory, an artificial potential field model is introduced to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge; then the magnitude of the resultant force and the coordinates of the resultant force point are used as the first non-time series characteristic condition and the second non-time series characteristic condition, and the conditional recurrent neural network model is embedded with the time series characteristics to form mixed information to obtain the best predicted trajectory sequence of the ship. On the one hand, it solves the problem of insufficient training data in the application scenario of real track prediction; on the other hand, the important prior knowledge for track prediction is added to the time series prediction process, and the important and discoverable knowledge between the data is scientifically extracted, giving full play to the role of time series characteristics and non-time series characteristic conditions with prior knowledge in ship track prediction, and improving the accuracy and efficiency of ship track prediction in the face of complex marine environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A flow chart of a ship trajectory prediction method based on a conditional recurrent neural network provided in an embodiment of the present invention;

[0046] Figure 2 A block diagram of a ship trajectory prediction method based on a conditional recurrent neural network provided by an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of an artificial potential field model effect provided by an embodiment of the present invention;

[0048] Figure 4 A structural diagram of a prediction model of a conditional recurrent neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The embodiments of the present application provide a ship trajectory prediction method, system, storage medium and electronic device based on a conditional recurrent neural network, thereby solving the technical problem that a purely data-driven method may reach its limit or lead to unsatisfactory results in medium- and long-term predictions, thereby ensuring the accuracy of ship trajectory prediction in complex marine environments.

[0051] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0052] Traditional machine learning methods have deficiencies in processing training data. They do not integrate non-temporal feature conditions with prior knowledge into the training process. In many real-world ship trajectory prediction application scenarios, purely data-driven methods may reach their limits or lead to unsatisfactory results in medium- and long-term predictions. The most obvious case is that there is not enough data to train a well-performing and fully generalized model, and the impact of complex external conditions on the ship's navigation trajectory is not considered, so the accuracy of ship trajectory prediction in complex marine environments cannot be guaranteed. In other words, the existing technology does not truly use external non-temporal features with prior knowledge as an independent source or through separate representations, but only adjusts them as needed.

[0053] Different from this, the conditional recurrent neural network (Cond-RNN) model in the embodiment of the present invention is aimed at the existing research on ship trajectory prediction problems. Traditional prediction methods such as linear regression, Kalman filtering and basic LSTM are mainly used for short-term prediction and have insufficient prediction accuracy. Figure 1 As shown: First, an artificial potential field model is introduced, and the resultant force conditions are calculated by the expected control points, including the magnitude of the resultant force and the coordinates of the resultant force points. Then, the non-time series feature conditions with prior knowledge are embedded in the conditional recurrent neural network model and the time series features to form mixed information for learning and training.

[0054] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0055] Example:

[0056] like Figure 2 As shown, an embodiment of the present invention provides a ship trajectory prediction method based on a conditional recurrent neural network, comprising:

[0057] S1. According to the information of the attraction point and repulsion point of the current trajectory, an artificial potential field model is introduced to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge;

[0058] S2. The magnitude of the resultant force and the coordinates of the resultant force point are used as the first non-time series feature condition and the second non-time series feature condition, and the conditional recurrent neural network model is embedded in the time series feature to form mixed information to obtain the optimal predicted trajectory sequence of the ship; wherein the time series feature is the longitude and latitude feature of the ship in a pre-acquired similar trajectory data set.

[0059] On the one hand, the embodiments of the present invention solve the problem of insufficient training data in real track prediction application scenarios; on the other hand, the important prior knowledge for track prediction is added to the time series prediction process, and the important and discoverable knowledge between the data is scientifically extracted, so as to give full play to the role of temporal characteristics and non-temporal characteristic conditions with prior knowledge in ship track prediction, and improve the accuracy and efficiency of ship trajectory prediction in complex marine environments.

[0060] Next, we will combine Figure 3 and 4 The following are the steps of the above solution:

[0061] First, it is explained that in the embodiment of the present invention, the definition is the gravitational force of the destination point on the ship, is the repulsive force of the obstacle on the ship, Gravity and repulsion The resultant force includes the magnitude of the resultant force and the coordinates of the resultant point. Also changes accordingly. For example, Figure 3 As shown, when obstacles 1 and 2 exist, is the repulsive force of obstacle 1 on the ship, is the repulsive force of obstacle 2 on the ship. Repulsion and repulsion The sum of .

[0062] In step S1, according to the information of the attraction point and repulsion point of the current trajectory, an artificial potential field model is introduced to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge.

[0063] The basic idea of ​​unified modeling of external information and domain knowledge in this step is to fuse this information and transform it into a force field that attracts and repels the current set of track points, and to express the correction of the predicted track by external non-temporal characteristic conditions through the comprehensive influence of the force field on the track.

[0064] Specifically, Figure 3 As shown, this step includes:

[0065] S11. Determine modeling elements: take the target point as the attraction point generated by external information for the current trajectory; and take the non-navigable point generated by meteorological and geographical factors and the prohibited navigation area point generated by other factors as the repulsion point generated by external information for the current trajectory.

[0066] S12. Construct a gravitational potential field function according to the ship position coordinates and the target point coordinates, and solve the negative gradient of the gravitational potential field function as the gravitational force exerted on the ship at the corresponding position.

[0067] The gravitational force on the ship at the corresponding position is expressed as:

[0068]

[0069] in, is the gravitational potential field function; is the proportional gain factor of the gravitational potential field; X is the ship position coordinates; X g is the target point position coordinate; m is the gravitational potential field factor; is the Euclidean distance between the ship and the target point;

[0070] is the gravitational force exerted on the ship at the corresponding position; for The negative gradient of .

[0071] S13. Construct a repulsive potential field function according to the ship position coordinates and the obstacle position coordinates, and solve the negative gradient of the repulsive potential field function as the repulsive force exerted on the ship at the corresponding position.

[0072] The repulsive force on the ship at the corresponding position is expressed as:

[0073]

[0074] in, is the repulsive potential field function; is the proportional gain factor of the repulsive potential field; is the obstacle position coordinate; is the Euclidean distance between the ship and the obstacle;

[0075] is the repulsive force on the ship at the corresponding position; for The negative gradient of .

[0076] S14. Calculate the resultant force according to the gravitational force and repulsive force received by the ship at the corresponding position to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge.

[0077] Wherein, the resultant force is expressed as:

[0078]

[0079] Among them, the combined Contains the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge.

[0080] The embodiment of the present invention innovatively introduces the theoretical knowledge of artificial potential fields in the field of robot path planning into the construction of a trajectory prediction model, changes the traditional prediction method of a neural network model that is completely data-driven, and models the external information of the trajectory prediction problem based on the artificial potential field theory. The non-time-series characteristic conditions with prior knowledge of the output of the artificial potential field model are used to facilitate the addition of time-series training data of the conditional recurrent neural network in subsequent steps, and a complete time series feature is formed through hierarchical multi-layer embedding. Parameters such as potential field strength and potential field direction are trained through the learning of the neural network, and the coupling function of external non-time-series information and time series trajectory information is learned through the training process of the conditional recurrent neural network.

[0081] In step S2, the magnitude of the resultant force and the coordinates of the resultant force point are used as the first non-time series feature condition and the second non-time series feature condition, and the conditional recurrent neural network model and the time series feature are embedded to form mixed information to obtain the optimal predicted trajectory sequence of the ship; wherein the time series feature is the longitude and latitude feature of the ship in a pre-acquired similar trajectory data set.

[0082] It should be noted that if Figure 4 As shown, the conditional recurrent neural network provided by the embodiment of the present invention includes an input node, seven hidden layers and an output node, and the seven hidden layers are respectively a GRU layer, a Dropout layer and five Dense layers.

[0083] On this basis, in this step, the first non-time series feature condition and the second non-time series feature condition are embedded in the conditional recurrent neural network model and the time series features in the pre-acquired similar trajectory data set to form mixed information, which means:

[0084] The first non-time series feature condition and the second non-time series feature condition are linearly transformed after being standardized through corresponding Dense layers, and the two output vectors are stacked respectively, and then output as a one-dimensional vector through a Dense layer;

[0085] The one-dimensional vector is input into the GRU layer and assigned to the initial state of the conditional recurrent neural network (this state is 0 by default in a general RNN). At the starting moment, the first non-time series feature condition, the second non-time series feature condition and the time series feature output by the input node constitute mixed information (in the training phase, the mixed information will also be propagated backward according to the calculation rules of the recurrent neural network); wherein at the input node, a similar trajectory data set obtained after data preprocessing is received, assuming that the length of the sequence X is n and the dimension is 2 (i.e., the longitude and latitude coordinates of the ship).

[0086] In addition, since the GRU unit is selected as one of the hidden layers, it is necessary to introduce:

[0087] In the structure of the GRU unit, r and z are reset gate and update gate respectively, h and are the activation information of the current state and the candidate activation information respectively. i GRU units, given an input time series feature sequence x =( x 1 ，x 2 ，…，x T ) ,in x t are the longitude and latitude coordinates of the ship’s trajectory at each moment.

[0088] The hidden state at the current moment is , this hidden state contains the relevant information of the previous moment. According to the hidden state passed down from the previous moment h t-1 and the current input x t , two gating states are obtained by the following two formulas:

[0089]

[0090] in, for the update gate, which determines the extent to which the unit updates the activation information or content; To reset the gate, it determines how much past information needs to be forgotten. When the value of is close to 0, the reset gate causes the unit to forget the previous computational state and work like reading the first component of the input sequence; σ is the sigmoid function; W and U is the weight matrix, respectively x t and h t-1The linear transformation, update gate and reset gate add the two parts of information after linear transformation and input them into the sigmoid function, so that the data is transformed into a value in the range of 0 to 1, thus acting as a gating signal.

[0091] After getting the gate signal, use the reset gate to get the new memory content:

[0092]

[0093] in, is the Hadamard product, which is the multiplication of the corresponding elements in the operation matrix; tanh is the hyperbolic tangent function, which is used as the activation function.

[0094] Use the resulting update gate to calculate the hidden state at the current moment , the calculation formula is as follows:

[0095]

[0096] in, It is the linear interpolation between the hidden state of the previous moment and the new memory content, and is the final memory of the output of the GRU layer. represents the selective forgetting of the hidden state at the previous moment, and Represents the selective memory of the current new information, using an update gate to perform forgetting and selective memory at the same time.

[0097] The GRU layer outputs the final memory of the current moment based on the mixed information, and then passes through a Dense layer, a Dropout layer, and a Dense layer in sequence to reduce the overfitting of the model and the linear transformation of the output results, respectively, and predicts the ship position at the next moment at the output node.

[0098] The predicted value is used as a known time series feature, and the corresponding first non-time series feature condition and the second non-time series feature condition are obtained at the same time. The cyclic prediction of subsequent moments is performed based on the conditional recurrent neural network to finally obtain the optimal predicted trajectory sequence (the longitude and latitude coordinates of the ship in the future period of time).

[0099] For example: assuming that the prediction step is determined to be 10, first start from the 1st position, fuse the first 10 positions (true values) with the non-time series feature conditions corresponding to the 10th position to predict the 11th position; then start from the 2nd position, fuse the 2nd position to the 11th position (predicted value), and fuse the non-time series feature conditions corresponding to the 11th position to predict the 12th position, so as to cyclically predict the values ​​for a longer time in the future, so as to finally obtain the optimal prediction trajectory sequence.

[0100] Exemplarily, the embodiment of the present invention sets the number of neurons in the neural network model to 100 and the probability of Dropout to 0.2. In particular, during the training phase of the conditional recurrent neural network model, the mean square error function (MSE) and the adaptive moment estimation (Adam) algorithm can also be selected for network training.

[0101] The embodiment of the present invention provides a ship trajectory prediction system based on a conditional recurrent neural network, comprising:

[0102] The acquisition module is used to introduce the artificial potential field model according to the information of the attraction point and repulsion point of the current trajectory, and obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge;

[0103] A prediction module is used to correspond the magnitude of the resultant force and the coordinates of the resultant force point as the first non-time series feature condition and the second non-time series feature condition, and embed the conditional recurrent neural network model and the time series feature to form mixed information to obtain the best predicted trajectory sequence of the ship; wherein the time series feature is the longitude and latitude feature of the ship in a pre-acquired similar trajectory data set.

[0104] An embodiment of the present invention provides a storage medium storing a computer program for ship trajectory prediction based on a conditional recurrent neural network, wherein the computer program enables a computer to execute the ship trajectory prediction method as described above.

[0105] An embodiment of the present invention provides an electronic device, including:

[0106] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include instructions for executing the ship trajectory prediction method as described above.

[0107] It can be understood that the ship trajectory prediction system based on conditional recurrent neural network, storage medium and electronic device provided in the embodiments of the present invention correspond to the ship trajectory prediction method based on conditional recurrent neural network provided in the embodiments of the present invention, and the explanations, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the ship trajectory prediction method, which will not be repeated here.

[0108] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0109] 1. In the embodiment of the present invention, firstly, according to the information of the attraction point and repulsion point of the current trajectory, an artificial potential field model is introduced to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge; then the magnitude of the resultant force and the coordinates of the resultant force point are used as the first non-time series characteristic condition and the second non-time series characteristic condition, and the conditional recurrent neural network model and the time series characteristics are embedded to form mixed information to obtain the best predicted trajectory sequence of the ship. On the one hand, it solves the problem of insufficient training data in the actual track prediction application scenario; on the other hand, the important prior knowledge for track prediction is added to the time series prediction process, and the important and discoverable knowledge between the data is scientifically extracted, giving full play to the role of time series characteristics and non-time series characteristic conditions with prior knowledge in ship track prediction, and improving the accuracy and efficiency of ship track prediction in the face of complex marine environments.

[0110] 2. Most existing neural network methods for ship trajectory prediction problems learn the dynamic model of moving targets by building various neural network structures and using a large amount of historical track data to train the neural network model parameters. However, in many actual application scenarios, there is not enough data support. Therefore, the problem of insufficient training data in real track prediction application scenarios is alleviated by introducing external non-time series feature conditions.

[0111] 3. The main technologies for current track prediction include three categories: linear prediction methods, nonlinear state estimation methods and neural network methods. Linear prediction methods mainly include linear regression, Kalman filtering and other methods, which are mainly suitable for linear target motion characteristics and short-term target motion state estimation problems. Nonlinear trajectory methods include a series of improved methods such as extended Kalman filtering. Their basic principles are similar to linear methods and can only fit short-term nonlinear dynamic models. They are usually less effective when faced with medium- and long-term ship trajectory prediction problems. Adding non-time-series characteristic conditions with prior knowledge that are important for track prediction to the time series prediction calculation process can effectively improve the accuracy of prediction in the case of nonlinear target motion in medium- and long-term ship trajectory prediction.

[0112] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0113] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship trajectory prediction method based on conditional recurrent neural network, characterized in that: include: According to the information of the attraction point and repulsion point of the current trajectory, an artificial potential field model is introduced to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge; The magnitude of the resultant force and the coordinates of the resultant force point are used as the first non-time series feature condition and the second non-time series feature condition, and the conditional recurrent neural network model is embedded with the time series feature to form mixed information, so as to obtain the best predicted trajectory sequence of the ship; wherein the time series feature is the latitude and longitude feature of the ship in the pre-acquired similar trajectory data set; The conditional recurrent neural network model includes an input node, seven hidden layers and an output node, and the seven hidden layers are respectively a GRU layer, a Dropout layer and five Dense layers; Embedding the first non-time series feature condition and the second non-time series feature condition into the conditional recurrent neural network model and the time series features in the pre-acquired similar trajectory data set to form mixed information means: The first non-time series feature condition and the second non-time series feature condition are linearly transformed after being standardized through corresponding Dense layers, and the two output vectors are stacked respectively, and then output as a one-dimensional vector through a Dense layer; The one-dimensional vector is input into the GRU layer, and is assigned to the initial state of the conditional recurrent neural network, and at the start time, the first non-time series feature condition, the second non-time series feature condition and the time series feature output by the input node form mixed information; The steps to obtain the best predicted trajectory sequence of the ship include: The GRU layer outputs the final memory of the current moment based on the mixed information, and sequentially passes through a Dense layer, a Dropout layer, and a Dense layer to predict the ship position at the next moment at the output node; The predicted value is used as a known time series feature to obtain the corresponding first non-time series feature condition and the second non-time series feature condition, and cyclic prediction of subsequent moments is performed based on the conditional recurrent neural network to ultimately obtain the optimal predicted trajectory sequence.

2. The ship trajectory prediction method according to claim 1, characterized in that: The steps of obtaining the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge include: The target point is used as the attraction point generated by external information for the current trajectory; and the non-navigable point generated by meteorological and geographical factors and the prohibited navigation area point generated by other factors are used as the repulsion point generated by external information for the current trajectory; According to the coordinates of the ship's position and the coordinates of the target point, a gravitational potential field function is constructed, and the negative gradient of the gravitational potential field function is solved as the gravitational force exerted on the ship at the corresponding position; According to the ship position coordinates and the obstacle position coordinates, a repulsive potential field function is constructed, and the negative gradient of the repulsive potential field function is solved as the repulsive force exerted on the ship at the corresponding position; The resultant force is calculated according to the gravitational force and repulsive force received by the ship at the corresponding position, so as to obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge.

3. The ship trajectory prediction method according to claim 2, characterized in that: The gravitational force on the ship at the corresponding position is expressed as: in, is the gravitational potential field function; is the proportional gain factor of the gravitational potential field; X is the ship position coordinates; X g is the target point position coordinate; m is the gravitational potential field factor; is the Euclidean distance between the ship and the target point; is the gravitational force exerted on the ship at the corresponding position; for The negative gradient of The repulsive force on the ship at the corresponding position is expressed as: in, is the repulsive potential field function; is the proportional gain factor of the repulsive potential field; is the obstacle position coordinate; is the Euclidean distance between the ship and the obstacle; is the repulsive force on the ship at the corresponding position; for The negative gradient of The resultant force is expressed as: Among them, the combined Contains the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge.

4. A ship trajectory prediction system based on conditional recurrent neural network, characterized in that: Used to execute the ship trajectory prediction method according to claim 1, comprising: The acquisition module is used to introduce the artificial potential field model according to the information of the attraction point and repulsion point of the current trajectory, and obtain the magnitude of the resultant force and the coordinates of the resultant force point with prior knowledge; A prediction module is used to correspond the magnitude of the resultant force and the coordinates of the resultant force point as the first non-time series feature condition and the second non-time series feature condition, and embed the conditional recurrent neural network model and the time series feature to form mixed information to obtain the best predicted trajectory sequence of the ship; wherein the time series feature is the longitude and latitude feature of the ship in a pre-acquired similar trajectory data set.

5. A storage medium, characterized in that: It stores a computer program for ship trajectory prediction based on conditional recurrent neural network, wherein the computer program enables a computer to execute the ship trajectory prediction method as described in any one of claims 1 to 3.

6. An electronic device, characterized in that: include: one or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include the ship trajectory prediction method according to any one of claims 1 to 3.

Images