Target trajectory prediction method and system based on bidirectional long and short term memory network

Abstract

The invention relates to a target trajectory prediction method and system based on a bidirectional long-short-term memory network, and the method mainly comprises the following steps: firstly, carrying out the spatial-temporal feature extraction of target trajectory data through a bidirectional long-short-term memory network layer, obtaining the spatial-temporal feature vector of target data, capturing a long-term dependency relationship and a short-term dependency relationship, and carrying out the spatial-temporal feature extraction of the target trajectory data; wherein the long-term dependency describes the time sequence and continuity of motion, and the short-term dependency describes the maneuvering characteristics of the motion; secondly, a prediction model is constructed, and hyper-parameters of the prediction model are optimized, so that the prediction performance and generalization ability of the model are improved; and finally, predicting a future trajectory of the target by using the trained optimization model, outputting a trajectory coordinate sequence of a plurality of time points in the future, and realizing accurate prediction of the target in a period of time in the future. According to the method, the bidirectional long-short-term memory network and hyper-parameter optimization are combined, so that high-precision trajectory prediction of the target is realized, and the adaptability of the prediction model to complex motion characteristics is enhanced.

Description

Technical Field

[0001] The present application belongs to the field of trajectory prediction, and in particular, relates to a target trajectory prediction method and system based on a bidirectional long short-term memory network. Background Art

[0002] With the continuous changes in the environment and the rapid iteration and update of technical equipment, the mobility of targets has become more and more significant, showing characteristics such as high speed, diversity, strong suddenness, and high flexibility. The dynamic characteristics of the target mean that it can quickly adjust and change its position, state, and action intention in a short period of time, including sudden turns, acceleration or deceleration, etc., causing its trajectory to show complex spatiotemporal characteristics. Highly dynamic trajectories often involve the joint influence of multiple complex factors and multiple variables, with significant dynamic flexibility and high uncertainty. These characteristics pose higher challenges to related tasks and prompt them to gradually expand from simple scenarios to highly complex scenarios. Therefore, accurate trajectory prediction is particularly important when dealing with complex environmental changes.

[0003] In recent years, with the rapid development of deep learning technology, the algorithms in the field of trajectory prediction have experienced a gradual shift from being dominated by traditional dynamic models such as Kalman filtering and Gaussian processes to technologies such as recurrent neural networks and attention mechanisms based on deep learning. Deep learning can better simulate and learn the trajectory change patterns of targets by building complex prediction models. However, traditional algorithms usually rely on pre-set target models. When there is a deviation between the model and the actual situation, it is easy to lead to a decrease in prediction accuracy and difficulty in coping with complex dynamic environments.

[0004] Therefore, it is urgent to propose a new high-maneuverability trajectory prediction method and system to more accurately predict the target trajectory and adapt to changeable and complex dynamic scenes. Summary of the invention

[0005] The present application provides a target trajectory prediction method and system based on a bidirectional long short-term memory network, so as to at least solve the problems that the traditional method is too dependent on the target motion model and has insufficient prediction accuracy in complex dynamic environments.

[0006] According to a first aspect, the present application provides a target trajectory prediction method based on a bidirectional long short-term memory network, comprising the following steps:

[0007] Spatiotemporal feature extraction: A bidirectional long short-term memory network layer is used to extract the spatial and temporal features of the target's trajectory data to obtain the spatiotemporal feature vector of the target data; the bidirectional long short-term memory network layer captures the long-term dependency and short-term dependency in the sequence data through its gating mechanism. The long-term dependency describes the timing and continuity of the target's motion, and the short-term dependency describes the mobility of the target's motion.

[0008] Prediction model construction: Build a prediction model and optimize the hyperparameters of the prediction model, including learning rate, batch size, and number of network layers, to improve the prediction performance and generalization ability of the prediction model;

[0009] Target trajectory prediction: Use the trained optimization model to predict the future trajectory of the target, output the trajectory point coordinate sequence of multiple future time points in chronological order, and realize accurate trajectory prediction of the target in the future.

[0010] Furthermore, in the spatiotemporal feature extraction step, the spatiotemporal features of the target motion are further extracted through the self-attention mechanism network:

[0011] The spatiotemporal feature vector obtained by the bidirectional long short-term memory network layer is input into the self-attention mechanism network module to focus on extracting key information, enhance the model's attention to important features, and improve the quality of feature representation.

[0012] Furthermore, the target is a manned or unmanned intelligent sports equipment, and the target has the characteristics of sudden maneuvering actions.

[0013] Furthermore, according to the object's own attributes and the object's motion law, a trajectory database of the object's maneuvering motion is constructed, and the trajectory database includes:

[0014] A set of dynamic motion models of the target;

[0015] A set of dynamic motion model parameters of the target;

[0016] A set of long-term regular trajectories of the target;

[0017] A collection of short-duration burst maneuvers of the target.

[0018] Furthermore, the prediction model includes:

[0019] Input layer: input target historical status data;

[0020] Convolutional neural network layer: extracting features of spatial dimension from the sample data input by the input layer;

[0021] Bidirectional long short-term memory network layer: extracting time series features from the data input by the convolutional neural network layer;

[0022] Self-attention mechanism layer: focuses on information that is more relevant to the current time step and extracts the short-term maneuvering features of the target;

[0023] Flattening layer: reduces the dimension of the output data of the self-attention mechanism layer;

[0024] Fully connected layer: receives the data processed by the flat layer and obtains the location information of the target through calculation.

[0025] Furthermore, the hyperparameters of the prediction model are optimized, including:

[0026] Determine the network structure and hyperparameter combination of the prediction model;

[0027] Calculate the fitness function;

[0028] The prediction model was trained using the training set and validation set, and the four behaviors of dung beetles were used in a ratio of 2:3:3:2 to iteratively optimize the network hyperparameter combination to determine the optimal hyperparameter combination.

[0029] Furthermore, the prediction model is iteratively trained with the fitness function as an indicator, and the iterative training includes:

[0030] Input historical data: Input the historical data of the sample sequence into the input layer;

[0031] Spatial feature extraction: In the convolutional neural network layer, the spatial dimension feature extraction is performed on the sample data input into the input layer;

[0032] Temporal feature extraction: In the bidirectional long short-term memory network layer, temporal feature extraction is performed on the data output by the convolutional neural network layer, wherein the bidirectional long short-term memory network layer is composed of two long short-term memory networks, and its calculation process includes:

[0033] f t =σ(W xf x t +W hf h t-1 +b f )

[0034] i t =σ(W xi x t +W hi h t-1 +b i )

[0035] O t =σ(W xo x t +W ho h t-1 +b o )

[0036] c t =f t ·c t-1 +i t tanh(W xc x t +Wxc h t-1 +b c )

[0037] h t =O t ·tanh(c t )

[0038]

[0039] In the formula, x t is the input sequence; f t For the forget gate; t is the input gate; O t is the output gate; h t is the hidden layer output; σ is the sigmoid activation function; tanh is the tanh activation function; W xf , W xi , W xo , W xc is the same as the input sequence x t The associated weight matrix; W hf , W hi , W ho , W hc The output h is the same as the previous moment t-1 The associated weight matrix; b f 、b i 、b o 、b c represents the bias vector; is the dot pair product;

[0040] Attention mechanism processing: The output of the convolutional neural network and the bidirectional long short-term memory network layer is used as the input of the self-attention mechanism layer. The implementation process includes:

[0041] Q=H t W q ; K=H t W k ; V = H t W v

[0042] α=QK Τ

[0043]

[0044] Attention(Q,K,V)=A·V

[0045] In the formula, H t is the input sequence; W q , W k , W vis the parameter matrix; Q is the query matrix; K is the key matrix; V is the value matrix; α is the attention score matrix; :d k The dimension is k, and its purpose is to scale the dot product result of Q and K;

[0046] Dimensionality reduction: Add a flattening layer after the attention layer to reduce the dimension of the data;

[0047] Output result: The fully connected layer receives the data processed by the flat layer, and obtains the location information of the target at the future moment through nonlinear mapping and weight learning.

[0048] Furthermore, for flight targets, the trajectory database also includes a set of basic maneuvering actions, which includes accelerated and decelerated straight flight, maximum overload turning, maximum overload climbing, and diving, while taking into account the target's tangential overload, normal overload and roll angle; based on the basic maneuvering action set, according to the dynamic motion model of the flight target, trajectory data is constructed using different initial conditions and parameters to cover possible trajectories and maneuvers of the flight target; measurement errors are added to the trajectory data and maneuvering action data and normalized to improve data quality and model training effects.

[0049] Furthermore, the target trajectory prediction method further includes:

[0050] The mean absolute error and weighted mean percentage error are used to evaluate the algorithm performance, specifically:

[0051]

[0052] In the formula, y i is the true value of the target trajectory; is the target trajectory prediction value; n is the number of samples.

[0053] According to a second aspect, the present application provides a target trajectory prediction system based on a bidirectional long short-term memory network, comprising:

[0054] Spatiotemporal feature extraction module: used to extract spatial and temporal features of the target's trajectory data using a bidirectional long short-term memory network layer to obtain the spatiotemporal feature vector of the target data; the bidirectional long short-term memory network layer captures the long-term dependency and short-term dependency in the sequence data through its gating mechanism, the long-term dependency describes the timing and continuity of the target's motion, and the short-term dependency describes the mobility of the target's motion;

[0055] Prediction model building module: used to build a prediction model and optimize the hyperparameters of the prediction model, including learning rate, batch size, and number of network layers, to improve the prediction performance and generalization ability of the prediction model;

[0056] Target trajectory prediction module: It is used to predict the future trajectory of the target using the trained optimization model, output the trajectory point coordinate sequence of multiple future time points in chronological order, and realize accurate trajectory prediction of the target in the future.

[0057] Compared with the prior art, the advantages and positive effects of this application are:

[0058] Compared with traditional algorithms, the present application demonstrates significant advantages in trajectory prediction accuracy, generalization ability, training efficiency, and adaptability; first, by introducing a bidirectional long short-term memory network layer, the model can more comprehensively extract the spatiotemporal characteristics of the target trajectory, accurately capture long-term dependencies and short-term maneuverability characteristics, and thus significantly improve prediction accuracy in complex dynamic scenarios; in addition, the combination of the self-attention mechanism enhances the model's ability to focus on key information, enabling the model to exhibit stronger generalization capabilities in a changing environment; at the same time, the present application incorporates the DBO algorithm to automatically optimize hyperparameters, reduces the complexity of manual parameter adjustment, and improves training efficiency and model stability; the present application is applicable to various types of manned or unmanned intelligent sports equipment, and constructs a large-scale trajectory database covering a variety of characteristics, which enables the target to have higher prediction accuracy under different motion conditions, while effectively reducing dependence on specific dynamic models, laying a solid foundation for wider applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic diagram of a flow chart of a target trajectory prediction method based on a bidirectional long short-term memory network according to an embodiment of the present application;

[0060] Figure 2 A schematic diagram of constructing a sample using a sliding window in one embodiment of the present application;

[0061] Figure 3 This is a schematic diagram of the self-attention mechanism layer of an embodiment of the present application;

[0062] Figure 4 A schematic diagram of the physical structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0063] The present application is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are intended to enable the present application to be better understood. However, those skilled in the art can easily recognize that some of the features can be omitted in different situations, or can be replaced by other elements, materials, and methods. In some cases, some operations related to the present application are not shown or described in the specification, in order to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0064] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0065] The present application proposes a target trajectory prediction method based on a bidirectional long short-term memory network, which mainly includes the following steps:

[0066] Constructing a trajectory database of target maneuvering motion: constructing a trajectory database of target maneuvering motion according to the target's own attributes and the target's motion law, the trajectory database includes: a set of dynamic motion models of the target; a set of dynamic motion model parameters of the target; a set of long-term regular trajectories of the target; and a set of short-term sudden maneuvering actions of the target.

[0067] Spatiotemporal feature extraction: A bidirectional long short-term memory (Bilstm) network layer is used to extract features of the target's trajectory data in spatial and temporal dimensions to obtain the spatiotemporal feature vector of the target data; the bidirectional long short-term memory network layer captures the long-term dependency and short-term dependency in the sequence data through its gating mechanism, the long-term dependency describes the timing and continuity of the target movement, and the short-term dependency describes the maneuverability of the target movement; in the spatiotemporal feature extraction step, the spatiotemporal features of the target movement are further extracted through the self-attention mechanism network: the spatiotemporal feature vector obtained by the bidirectional long short-term memory network layer is input into the self-attention mechanism network module to focus on extracting key information, enhance the model's attention to important features, and improve the quality of feature representation.

[0068] Prediction model construction: Build a prediction model and optimize the hyperparameters of the prediction model, including learning rate, batch size, and number of network layers, to improve the prediction performance and generalization ability of the prediction model.

[0069] Target trajectory prediction: Use the trained optimization model to predict the future trajectory of the target, output the trajectory point coordinate sequence of multiple future time points in chronological order, and realize accurate trajectory prediction of the target in the future.

[0070] The targets described in this application include manned or unmanned intelligent sports equipment, suitable for civil and military vehicles, ships, aircraft, etc., as well as civil and military drones, unmanned vehicles, unmanned ships, etc., and these targets usually have sudden maneuvering characteristics.

[0071] See also Figure 1 , below, taking a drone or an aircraft as an example, a target trajectory prediction method based on a bidirectional long short-term memory network is described, including the following steps:

[0072] S1. Construction of high-maneuverability trajectory database: Based on the seven basic maneuvers, the high-maneuverability trajectory under the influence of multiple variables such as tangential overload, normal overload, and roll angle is considered, and the trajectory database is constructed according to the UAV or aircraft dynamics model;

[0073] Specifically, step S1 includes:

[0074] S101, construct high maneuverability trajectory data using different initial conditions based on the UAV or aircraft dynamics model to cover more flight trajectories;

[0075] S102, adding a measurement error with a mean of 0 and a variance of 0.1, and performing data processing in a normalized manner.

[0076] The above seven basic maneuvers generally refer to some common maneuvers performed by drones or aircraft in the air, including:

[0077] Accelerated straight flight: The drone or aircraft flies in a straight line under acceleration.

[0078] Slow down and fly straight: The drone or aircraft flies in a straight line while slowing down.

[0079] Maximum G-load turn: The drone or aircraft performs a turn with the maximum G-load it can withstand.

[0080] Maximum overload climb: The drone or aircraft climbs upward at the maximum overload it can withstand.

[0081] Dive: A drone or aircraft dives downward, usually with an increase in speed.

[0082] Roll: The drone or aircraft rolls (rotates) around its longitudinal axis.

[0083] Hovering or evasive maneuvers: Drones or aircraft perform specific maneuvers to avoid collision or tracking.

[0084] S2. Sample construction and data set division: The sliding window method is used to construct samples, and the data set is divided into training set, validation set and test set, see Figure 2 ;

[0085] Specifically, step S2 includes:

[0086] S201, constructing samples by using a sliding window method;

[0087] S202, divide the training set, validation set and test set into a ratio of 6:3:1.

[0088] S3. Construction of prediction model integrating Dung Beetle Optimization (DBO): Build a prediction model and integrate DBO into the prediction model to optimize the neural network hyperparameters;

[0089] Specifically, step S3 includes:

[0090] S301, construct AT-CNN-Bilstm prediction network model, including:

[0091] S3011, input layer;

[0092] S3012, convolutional network layer: extracting features of spatial dimension from the sample data input by the input layer;

[0093] S3013, bidirectional long short-term memory network layer: extracting time series features from the data input by the convolutional neural network layer;

[0094] S3014, attention mechanism layer: focusing on information that is more relevant to the current time step;

[0095] S3015, flattening layer: reduce the dimension of output data;

[0096] S3016, fully connected layer: receiving the data processed by the flat layer, and obtaining target position information through calculation;

[0097] S302, DBO optimization model, including:

[0098] S3021. Construct model hyperparameter space;

[0099] S3022. Determine the fitness function in the optimization process.

[0100] S4, model training and hyperparameter optimization: train the prediction model to execute the optimization algorithm and find the optimal hyperparameter combination;

[0101] Specifically, step S4 includes:

[0102] S401, determining the prediction model network structure according to the hyperparameter combination;

[0103] S402, calculating fitness function;

[0104] S403, using the training set and the validation set to train the prediction model, and using the four behaviors of the dung beetle in a ratio of 2:3:3:2 to iteratively optimize the network hyperparameter combination to determine the optimal hyperparameter combination.

[0105] The AT-CNN-Bilstm model is iteratively trained using the fitness function as an indicator:

[0106] The input layer is the historical data of the sample sequence.

[0107] The convolutional network layer is used to extract features of the spatial dimension of the sample data input by the input layer.

[0108] The bidirectional long short-term memory network layer is used to extract time series features from the data input by the convolutional neural network layer.

[0109] The bidirectional long short-term memory network layer (Bilstm) consists of two long short-term memory networks (Long and Short-Term Memory Networks, LSTM), and its calculation process is as follows:

[0110] f t =σ(W xf x t +W hf h t-1 +b f )

[0111] i t =σ(W xi x t +W hi h t-1 +b i )

[0112] O t =σ(W xo x t +W ho h t-1 +b o )

[0113] c t =f t ·ct-1 +i t tanh(W xc x t +W xc h t-1 +b c )

[0114] h t =O t ·tanh(c t )

[0115]

[0116] In the formula, x t is the input sequence; f t For the forget gate; t is the input gate; O t is the output gate; h t is the hidden layer output; σ is the sigmoid activation function; tanh is the tanh activation function; W xf , W xi , W xo , W xc is the same as the input sequence x t The associated weight matrix; W hf , W hi , W ho , W hc The output h is the same as the previous moment t-1 The associated weight matrix; b f 、b i 、b o 、b c represents the bias vector; · is the dot product.

[0117] The self-attention mechanism layer focuses on time steps that are more relevant to the current maneuver, thereby improving prediction accuracy.

[0118] After the data is extracted from the sequential spatiotemporal features by CNN-Bilstm, an attention mechanism layer is established to further extract the important features of the data. The output of CNN-Bilstm will be used as the input of the attention mechanism, so that the computational model focuses on the time step with higher relevance to the current maneuver and reduces the attention to other non-critical factors.

[0119] The principle diagram of the self-attention mechanism layer is as follows Figure 3 As shown, the implementation process of the self-attention mechanism layer is as follows:

[0120] Q=H t W q ; K=H t W k ; V = H t Wv

[0121] α=QK Τ

[0122]

[0123] Attention(Q,K,V)=A·V

[0124] In the formula, H t is the input sequence; W q , W k , W v is the parameter matrix; Q is the query matrix; K is the key matrix; V is the value matrix; α is the attention score matrix; :d k The dimension is k, and its purpose is to scale the dot product result of Q and K.

[0125] Flatten is used to reduce the dimension of the data.

[0126] The fully connected layer receives the data processed by the flat layer and obtains the location information of the target at future time points through nonlinear mapping and weight learning.

[0127] S5: Predicting the trajectory points of the high-maneuverability target at multiple future time points in chronological order to obtain a trajectory point coordinate sequence, thereby realizing trajectory prediction of the high-maneuverability target within a period of time in the future.

[0128] In order to verify the prediction accuracy of the model, this application uses the mean absolute error (MAE) and weighted average percentage error (WMAPE) to evaluate the algorithm performance. The calculation formula is:

[0129]

[0130] In the formula, y i is the true value of the target trajectory; is the target trajectory prediction value; n is the number of samples.

[0131] This embodiment has significant advantages in trajectory prediction accuracy, generalization ability, algorithm optimization and model flexibility, and is suitable for high-dynamic target trajectory prediction in a variety of complex dynamic environments. Specifically:

[0132] First, improve the accuracy of trajectory prediction: By integrating the self-attention mechanism with the BiLSTM network model, this application can more effectively capture the spatiotemporal characteristics of the target, improve the ability to capture high-dynamic trajectories, and significantly improve the trajectory prediction accuracy in complex environments.

[0133] Second, enhance the generalization ability of the model: By introducing the self-attention mechanism, the model can focus on time steps that are more relevant to the current maneuver and reduce dependence on non-critical information, thereby improving the generalization ability of the model in a variety of dynamic scenarios and adapting to different task environments.

[0134] Third, DBO algorithm optimizes hyperparameters: This application optimizes the hyperparameters of the neural network by introducing the DBO (Dung Beetle Optimization) algorithm, effectively avoiding the complexity of traditional manual parameter adjustment and further improving the training efficiency and prediction effect of the model.

[0135] Fourth, compatibility with multi-variable dynamic input: During the prediction process, this method comprehensively considers a variety of maneuvering variables (such as tangential overload, normal overload, roll angle, etc.), so that the prediction model can adapt to multi-variable and highly flexible target trajectory changes, ensuring the accuracy of the prediction results.

[0136] Fifth, reduce model dependence: Compared with the traditional method that relies on a specific dynamic model, this application adopts a data-driven approach and trains through a large-scale trajectory database, so that the model can better adapt to different flight conditions and maneuvers, reducing dependence on the target motion model.

[0137] In addition, an embodiment of the present application also provides a target trajectory prediction system based on a bidirectional long short-term memory network, the system including the following main modules: a module for constructing a trajectory database for target maneuvering motion: used to construct a trajectory database for the target maneuvering motion according to the target's own attributes and the target's motion law, the trajectory database including: a set of dynamic motion models of the target; a set of dynamic motion model parameters of the target; a set of long-term regular trajectories of the target; and a set of short-term sudden maneuvering actions of the target. Spatiotemporal feature extraction module: used to extract the spatial and temporal features of the target's trajectory data using a bidirectional long short-term memory network layer (Bidirectional Long Short-Term Memory, Bilstm) to obtain the spatiotemporal feature vector of the target data; the bidirectional long short-term memory network layer captures the long-term dependency and short-term dependency in the sequence data through its gating mechanism, the long-term dependency describes the timing and continuity of the target's movement, and the short-term dependency describes the mobility of the target's movement; in the spatiotemporal feature extraction step, the spatiotemporal features of the target's movement are further extracted through the self-attention mechanism network: the spatiotemporal feature vector obtained by the bidirectional long short-term memory network layer is input into the self-attention mechanism network module, and the key information is focused and extracted, the model's attention to important features is enhanced, and the quality of feature representation is improved. Prediction model construction module: used to construct a prediction model and optimize the hyperparameters of the prediction model, the hyperparameters include learning rate, batch size, and number of network layers, etc., to improve the prediction performance and generalization ability of the prediction model. Target trajectory prediction module: It is used to predict the future trajectory of the target using the trained optimization model, output the trajectory point coordinate sequence of multiple future time points in chronological order, and realize accurate trajectory prediction of the target in the future.

[0138] In addition, the embodiment of the present application also provides an electronic device for target trajectory prediction based on a bidirectional long short-term memory network. Figure 4 The physical structure diagram of the electronic device provided in the embodiment of the present application, the electronic device may include: a processor (processor) 301, a communication interface (Communications Interface) 302, a memory (memory) 303 and a bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the bus 304. The processor 301 can call a computer program stored in the memory 303 and can be run on the processor 301 to execute the target trajectory prediction method based on the bidirectional long short-term memory network provided in the above embodiment.

[0139] In addition, the logic instructions in the above-mentioned memory 303 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk.

[0140] The above specific examples are used to illustrate the present application, which is only used to help understand the present application and is not intended to limit the present application. For technicians in the technical field to which the present application belongs, they can also make some simple deductions, deformations or substitutions based on the ideas of the present application.

Claims

1. A target trajectory prediction method based on a bidirectional long short-term memory network, characterized in that: include: Spatiotemporal feature extraction: A bidirectional long short-term memory network layer is used to extract the spatial and temporal features of the target's trajectory data to obtain the spatiotemporal feature vector of the target data; the bidirectional long short-term memory network layer captures the long-term dependency and short-term dependency in the sequence data through its gating mechanism. The long-term dependency describes the timing and continuity of the target's motion, and the short-term dependency describes the mobility of the target's motion. Prediction model construction: Build a prediction model and optimize the hyperparameters of the prediction model, including learning rate, batch size, and number of network layers, to improve the prediction performance and generalization ability of the prediction model; Target trajectory prediction: Use the trained optimization model to predict the future trajectory of the target, output the trajectory point coordinate sequence of multiple future time points in chronological order, and realize accurate trajectory prediction of the target in the future.

2. According to claim 1, the target trajectory prediction method based on the bidirectional long short-term memory network is characterized in that: In the spatiotemporal feature extraction step, the spatiotemporal features of the target motion are further extracted through the self-attention mechanism network: The spatiotemporal feature vector obtained by the bidirectional long short-term memory network layer is input into the self-attention mechanism network module to focus on extracting key information, enhance the model's attention to important features, and improve the quality of feature representation.

3. According to claim 1, the target trajectory prediction method based on bidirectional long short-term memory network is characterized in that: The target is a manned or unmanned intelligent sports equipment, and the target has the characteristic of sudden maneuvering action.

4. According to claim 3, the target trajectory prediction method based on bidirectional long short-term memory network is characterized in that: According to the properties of the target and the motion law of the target, a trajectory database of the target's maneuvering motion is constructed, and the trajectory database includes: A set of dynamic motion models of the target; A set of dynamic motion model parameters of the target; A set of long-term regular trajectories of the target; A collection of short-duration burst maneuvers of the target.

5. The target trajectory prediction method based on bidirectional long short-term memory network according to claim 1 is characterized in that: The prediction model comprises: Input layer: input target historical status data; Convolutional neural network layer: extracting features of spatial dimension from the sample data input by the input layer; Bidirectional long short-term memory network layer: extracting time series features from the data input by the convolutional neural network layer; Self-attention mechanism layer: focuses on information that is more relevant to the current time step and extracts the short-term maneuvering features of the target; Flattening layer: reduces the dimension of the output data of the self-attention mechanism layer; Fully connected layer: receives the data processed by the flat layer and obtains the location information of the target through calculation.

6. The target trajectory prediction method based on bidirectional long short-term memory network according to claim 1 is characterized in that: Optimizing the hyperparameters of the prediction model includes: Determine the network structure and hyperparameter combination of the prediction model; Calculate the fitness function; The prediction model was trained using the training set and validation set, and the four behaviors of dung beetles were used in a ratio of 2:3:3:2 to iteratively optimize the network hyperparameter combination to determine the optimal hyperparameter combination.

7. The target trajectory prediction method based on bidirectional long short-term memory network according to claim 6 is characterized in that: The prediction model is iteratively trained using the fitness function as an indicator, and the iterative training includes: Input historical data: Input the historical data of the sample sequence into the input layer; Spatial feature extraction: In the convolutional neural network layer, the spatial dimension feature extraction is performed on the sample data input into the input layer; Temporal feature extraction: In the bidirectional long short-term memory network layer, the temporal feature extraction is performed on the data output by the convolutional neural network layer, wherein the bidirectional long short-term memory network layer is composed of two long short-term memory networks, and the calculation process includes: f t =σ(W xf x t +W hf h t-1 +b f ) i t =σ(W xi x t +W hi h t-1 +b i ) O t =σ(W xo x t +W ho h t-1 +b o ) c t =f t ·c t-1 +i t ·tanh(W xc x t +W xc h t-1 +b c ) h t =O t ·tanh(c t ) In the formula, x t is the input sequence; f t For the forget gate; t is the input gate; O t is the output gate; h t is the hidden layer output; σ is the sigmoid activation function; tanh is the tanh activation function; W xf , W xi , W xo , W xc is the same as the input sequence x t The associated weight matrix; W hf , W hi , W ho , W hc The output h is the same as the previous moment t-1 The associated weight matrix; b f , b i , b o , b c represents the bias vector; is the dot pair product; Attention mechanism processing: The output of the convolutional neural network and the bidirectional long short-term memory network layer is used as the input of the self-attention mechanism layer. The implementation process includes: Q=H t W q ;K=H t W k ;V=H t W v α=QK Τ Attention(Q,K,V)=A·V In the formula, H t is the input sequence; W q , W k , W v is the parameter matrix; Q is the query matrix; K is the key matrix; V is the value matrix; α is the attention score matrix; :d k The dimension is k, and its purpose is to scale the dot product result of Q and K; Dimensionality reduction: Add a flattening layer after the attention layer to reduce the dimension of the data; Output result: The fully connected layer receives the data processed by the flat layer, and obtains the location information of the target at the future moment through nonlinear mapping and weight learning.

8. The target trajectory prediction method based on bidirectional long short-term memory network according to claim 4 is characterized in that: For flight targets, the trajectory database also includes a set of basic maneuvering actions, which includes accelerated and decelerated straight flight, maximum overload turning, maximum overload climbing, and diving, while taking into account the target's tangential overload, normal overload and roll angle; based on the basic maneuvering action set, according to the dynamic motion model of the flight target, trajectory data is constructed using different initial conditions and parameters to cover possible trajectories and maneuvers of the flight target; measurement errors are added to the trajectory data and maneuvering action data and normalized to improve data quality and model training effects.

9. The target trajectory prediction method based on bidirectional long short-term memory network according to claim 1 is characterized in that: The target trajectory prediction method also includes: The mean absolute error and weighted mean percentage error are used to evaluate the algorithm performance, specifically: In the formula, y i is the true value of the target trajectory; is the target trajectory prediction value; n is the number of samples.

10. A target trajectory prediction system based on a bidirectional long short-term memory network, characterized in that: include: Spatiotemporal feature extraction module: used to extract spatial and temporal features of the target's trajectory data using a bidirectional long short-term memory network layer to obtain the spatiotemporal feature vector of the target data; the bidirectional long short-term memory network layer captures the long-term dependency and short-term dependency in the sequence data through its gating mechanism, the long-term dependency describes the timing and continuity of the target's motion, and the short-term dependency describes the mobility of the target's motion; Prediction model building module: used to build a prediction model and optimize the hyperparameters of the prediction model, including learning rate, batch size, and number of network layers, to improve the prediction performance and generalization ability of the prediction model; Target trajectory prediction module: It is used to predict the future trajectory of the target using the trained optimization model, output the trajectory point coordinate sequence of multiple future time points in chronological order, and realize accurate trajectory prediction of the target in the future.

Images