A long-time track prediction method based on deep learning, medium and device

Abstract

The application provides a long-time track prediction method based on deep learning, a medium and a device, and the method comprises the following steps: step one, obtaining track data; step two, preprocessing the track data; step three, using the preprocessed track data to construct a track data set; step four, using a long-time track prediction model based on deep learning to train the track data set, stopping training after the loss function converges, and obtaining a weight file of the long-time track prediction model; step five, preprocessing the track data to be predicted through step two to obtain a track sample to be predicted; step six, using the long-time track prediction model based on deep learning and the trained weight file to perform long-time track prediction on the track sample to be predicted, and obtaining a predicted track. The application has high accuracy, a wide application range and good versatility.

Description

Technical Field

[0001] This invention relates to the field of trajectory prediction technology, and more specifically, to a long-term trajectory prediction method, medium, and device based on deep learning. Background Technology

[0002] With the booming development of smart cities, smart transportation, and intelligent navigation, trajectory prediction is playing an increasingly important role. Trajectory prediction is divided into single-step trajectory prediction and multi-step trajectory prediction. Single-step trajectory prediction uses the trajectory state of a target at multiple past sampling moments to predict the trajectory state at the next sampling moment, while multi-step trajectory prediction uses the trajectory state of a target at multiple past sampling moments to predict the trajectory state at multiple future sampling moments.

[0003] Although deep learning-based short-term trajectory prediction has made significant progress, it has not yet been directly applied to long-term trajectory prediction. Existing long-term trajectory prediction methods mainly include one based on clustering and trajectory matching techniques to estimate trajectories at multiple future moments. This type of method is not suitable for long-term prediction of complex, multi-state trajectories. Another method is based on single-moment trajectory prediction to predict the trajectory at the next moment and then using the trajectory at the next moment as input to iteratively predict the trajectory at multiple future moments. Since predictions always have biases, the prediction error will become larger and larger as the number of iterations increases, resulting in poor prediction results for multiple future moments. Summary of the Invention

[0004] The present invention aims to provide a long-term trajectory prediction method, medium, and device based on deep learning, in order to solve the problem that current long-term trajectory prediction methods are not suitable for long-term prediction of complex multi-state trajectories and the prediction results are very poor.

[0005] This invention provides a long-term trajectory prediction method based on deep learning, comprising the following steps:

[0006] Step 1: Obtain track data;

[0007] Step 2: Preprocess track data;

[0008] Step 3: Construct a track dataset using the preprocessed track data;

[0009] Step 4: Train the trajectory dataset using a deep learning-based long-term trajectory prediction model. Stop training after the loss function converges to obtain the weight file of the long-term trajectory prediction model.

[0010] Step 5: Preprocess the trajectory data to be predicted using the methods described in Step 2 to obtain the trajectory sample to be predicted;

[0011] Step 6: Using a deep learning-based long-term trajectory prediction model and a trained weight file, perform long-term trajectory prediction on the trajectory samples to be predicted, and obtain the predicted trajectory.

[0012] Furthermore, in step two, the preprocessing of track data includes:

[0013] Speed ​​and heading are calculated using latitude, longitude, and time.

[0014] Smoothing filters are applied to longitude, latitude, speed, and heading;

[0015] The longitude, latitude, speed, and heading before smoothing filtering, as well as the longitude, latitude, speed, and heading after smoothing filtering, are used to construct track data in eight states.

[0016] Furthermore, in step two, track data that does not meet the duration requirement needs to be removed.

[0017] Furthermore, step three, which involves constructing a track dataset using the preprocessed track data, includes:

[0018] Set the prediction time step N, and use a window width of 2*N to perform single-step sliding sampling on the track data. The longitude, latitude, speed, direction, smoothed longitude, smoothed latitude, smoothed speed and smoothed direction of the first to Nth time in each window are used as historical tracks, and the longitude and latitude of the (N+1)th to 2*Nth time are used as future tracks. The historical track and the future track obtained after each sliding are combined to form a track sample in the track dataset.

[0019] Furthermore, in step four, the long-term trajectory prediction model based on deep learning is as follows:

[0020] The eight inputs to the deep learning-based long-term trajectory prediction model are: original longitude (size 1*24), smoothed longitude, original latitude, smoothed latitude, original speed, smoothed speed, original direction, and smoothed direction.

[0021] The eight inputs are sequentially processed through a convolutional layer, an LSTM layer, another convolutional layer, an LSTM layer, and a fully connected layer to obtain a feature vector of size 1*64.

[0022] The longitude feature vector is obtained by adding the feature vectors of the original longitude and the smoothed longitude respectively; the latitude feature vector is obtained by adding the feature vectors of the original latitude and the smoothed latitude respectively; the velocity feature vector is obtained by adding the feature vectors of the original velocity and the smoothed velocity respectively; and the direction feature vector is obtained by adding the feature vectors of the original direction and the smoothed direction respectively.

[0023] The longitude feature vector, latitude feature vector, velocity feature vector and direction feature vector are concatenated to obtain a 1*96 feature vector. A convolutional layer, an LSTM layer and a fully connected layer are applied to it in sequence to obtain a feature vector of size 1*64. A fully connected layer of length 2*24 is used to predict 24 latitude and longitude points.

[0024] The present invention also provides a computer terminal storage medium storing computer terminal executable instructions, which are used to execute the above-described deep learning-based long-term trajectory prediction method.

[0025] The present invention also provides a computing device, comprising:

[0026] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aforementioned deep learning-based long-term trajectory prediction method.

[0027] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0028] 1. This invention has a high accuracy rate. Traditional long-term trajectory prediction and recognition methods require a series of complex operations and a lot of manual intervention, which cannot guarantee prediction accuracy. In contrast, this invention uses a deep learning-based method to fuse and extract trajectory features, resulting in a high prediction accuracy rate.

[0029] 2. This invention has a wide range of applications. Different targets have different historical flight path durations, and corresponding prediction durations can be set based on different historical flight paths to meet different flight path prediction needs.

[0030] 3. This invention has good versatility. Traditional long-term trajectory prediction methods are designed for specific trajectory data. When the new trajectory to be predicted differs significantly from the original trajectory dataset, the prediction error is large. It is necessary to optimize and modify the original prediction method to adapt to the new trajectory data. The method has poor versatility and makes system maintenance difficult. In contrast, this invention only requires adding the encountered trajectory to the dataset for training, and then replacing the model with the new one. This makes maintenance convenient. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of a long-term trajectory prediction method based on deep learning in an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram of a long-term trajectory prediction model based on deep learning in an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0035] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0036] Example

[0037] like Figure 1 As shown, this embodiment proposes a long-term trajectory prediction method based on deep learning, including the following steps:

[0038] Step 1: Obtain J tracks of flight data (J≥100), in this embodiment J=600;

[0039] Step 2: Preprocess track data:

[0040] (1) Track data that does not meet the duration requirement (less than 8 hours) needs to be removed;

[0041] (2) Calculate the speed and heading using latitude and longitude points and time; perform smoothing filtering on the longitude, latitude, speed and heading; and use the longitude, latitude, speed and heading before smoothing filtering, as well as the longitude, latitude, speed and heading after smoothing filtering, to form track data in 8 states.

[0042] Step 3: Construct a track dataset using the preprocessed track data:

[0043] The prediction time step is set to N = 4 hours. The track data is sampled every 10 minutes using a window width of 8 hours. The longitude, latitude, speed, direction, smoothed longitude, smoothed latitude, smoothed speed, and smoothed direction of the first 10 minutes to the 24th 10 minutes in each window are used as historical tracks. The longitude and latitude of the 25th to the 48th 10 minutes are used as future tracks. The historical tracks and future tracks obtained after each slide constitute one track sample in the track dataset.

[0044] Step 4: Train the trajectory dataset using a deep learning-based long-term trajectory prediction model. Stop training after the loss function converges to obtain the weight file of the long-term trajectory prediction model.

[0045] like Figure 2 As shown, the deep learning-based long-term trajectory prediction model is as follows:

[0046] The eight inputs to the deep learning-based long-term trajectory prediction model are: original longitude (size 1*24), smoothed longitude, original latitude, smoothed latitude, original speed, smoothed speed, original direction, and smoothed direction.

[0047] The eight inputs are sequentially processed through a convolutional layer, an LSTM layer, another convolutional layer, an LSTM layer, and a fully connected layer to obtain a feature vector of size 1*64.

[0048] The longitude feature vector is obtained by adding the feature vectors of the original longitude and the smoothed longitude respectively; the latitude feature vector is obtained by adding the feature vectors of the original latitude and the smoothed latitude respectively; the velocity feature vector is obtained by adding the feature vectors of the original velocity and the smoothed velocity respectively; and the direction feature vector is obtained by adding the feature vectors of the original direction and the smoothed direction respectively.

[0049] The longitude feature vector, latitude feature vector, velocity feature vector and direction feature vector are concatenated to obtain a 1*96 feature vector. A convolutional layer, an LSTM layer and a fully connected layer are applied to it in sequence to obtain a feature vector of size 1*64. A fully connected layer of length 2*24 is used to predict 24 latitude and longitude points.

[0050] Step 5: Preprocess the trajectory data to be predicted using the methods described in Step 2 to obtain the trajectory sample to be predicted;

[0051] Step 6: Using a deep learning-based long-term trajectory prediction model and a trained weight file, perform long-term trajectory prediction on the trajectory samples to be predicted, and obtain the predicted trajectory.

[0052] Furthermore, in some embodiments, a computer terminal storage medium is proposed, storing computer terminal executable instructions for executing the deep learning-based long-term trajectory prediction method as described in the preceding embodiments. Examples of computer storage media include magnetic storage media (e.g., floppy disks, hard disks, etc.), optical recording media (e.g., CD-ROMs, DVDs, etc.), or memory such as memory cards, ROMs, or RAMs. The computer storage medium can also be distributed across a network-connected computer system, for example, as an application store.

[0053] Furthermore, in some embodiments, a computing device is proposed, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the deep learning-based long-term trajectory prediction method as described in the foregoing embodiments. Examples of computing devices include PCs, tablets, smartphones, or PDAs.

[0054] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A long-term trajectory prediction method based on deep learning, characterized in that, Includes the following steps: Step 1: Acquire track data; Step 2: Preprocess track data; Step 3: Construct a track dataset using the preprocessed track data; Step 4: Train the trajectory dataset using a deep learning-based long-term trajectory prediction model. Stop training after the loss function converges to obtain the weight file of the long-term trajectory prediction model. Step 5: Preprocess the trajectory data to be predicted using the methods described in Step 2 to obtain the trajectory sample to be predicted; Step 6: Using a deep learning-based long-term trajectory prediction model and a trained weight file, perform long-term trajectory prediction on the trajectory samples to be predicted, and obtain the predicted trajectory. In step two, the preprocessing of track data includes: Speed ​​and heading are calculated using latitude, longitude, and time. Smoothing filters are applied to longitude, latitude, speed, and heading; The longitude, latitude, speed, and heading before smoothing and filtering, as well as the longitude, latitude, speed, and heading after smoothing and filtering, are used to form track data in 8 states. Step three, constructing the track dataset using the preprocessed track data, includes: Set the prediction time step N, and use a window width of 2*N to perform single-step sliding sampling on the track data. The longitude, latitude, speed, direction, smoothed longitude, smoothed latitude, smoothed speed and smoothed direction of the first to Nth time in each window are used as historical tracks, and the longitude and latitude of the N+1 to 2*Nth time are used as future tracks. The historical track and the future track obtained after each sliding are combined to form a track sample in the track dataset. In step four, the long-term trajectory prediction model based on deep learning is as follows: The eight inputs to the deep learning-based long-term trajectory prediction model are: original longitude (size 1*24), smoothed longitude, original latitude, smoothed latitude, original speed, smoothed speed, original direction, and smoothed direction. The eight inputs are sequentially processed through a convolutional layer, an LSTM layer, another convolutional layer, an LSTM layer, and a fully connected layer to obtain a feature vector of size 1*64. The longitude feature vector is obtained by adding the feature vectors of the original longitude and the smoothed longitude respectively; the latitude feature vector is obtained by adding the feature vectors of the original latitude and the smoothed latitude respectively; the velocity feature vector is obtained by adding the feature vectors of the original velocity and the smoothed velocity respectively; and the direction feature vector is obtained by adding the feature vectors of the original direction and the smoothed direction respectively. The longitude feature vector, latitude feature vector, velocity feature vector and direction feature vector are concatenated to obtain a 1*96 feature vector. A convolutional layer, an LSTM layer and a fully connected layer are applied to it in sequence to obtain a feature vector of size 1*64. A fully connected layer of length 2*24 is used to predict 24 latitude and longitude points.

2. The long-term trajectory prediction method based on deep learning according to claim 1, characterized in that, In step two, track data that does not meet the duration requirement needs to be removed.

3. A computer terminal storage medium storing computer terminal executable instructions, characterized in that, The computer terminal can execute instructions for performing the long-term trajectory prediction method based on deep learning as described in any one of claims 1-2.

4. A computing device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the deep learning-based long-term trajectory prediction method as described in any one of claims 1-2.

Images