Path-dependent trajectory prediction method based on deep neural network, medium and product
Through the path-dependent trajectory prediction method based on deep neural networks, the pre-trained trajectory prediction neural network and probability distribution describe the trajectory prediction results, solving the lack of real-time and accuracy caused by modeling errors and computational complexity in the prior art. The problem is to achieve efficient trajectory prediction and improve the robustness and stability of prediction.
Patent Information
- Application Number
- CN202411420233.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-12
- Publication Date
- 2025-01-10
AI Technical Summary
The existing trajectory prediction technology has shortcomings in the real-time and accuracy of prediction, especially when dealing with complex traffic scenarios and uncertain behaviors, which leads to large modeling errors and high computational complexity, which limits the application of trajectory prediction technology. and development.
Using a path-dependent trajectory prediction method based on deep neural networks, data preprocessing and completion are performed by obtaining the reference path and actual operating trajectory of the object to be predicted, and the trajectory prediction neural network is used to predict, and the trajectory is described through probability distribution. The prediction results are randomly sampled to generate a complete sequence of predicted trajectory data points.
It significantly improves the real-time and accuracy of predictions, improves the robustness and stability of predictions, ensures the reliability of prediction results, and can effectively capture the diversity of trajectories and path dependence.
Smart Images
Figure CN119294673A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a path-dependent trajectory prediction method, medium and product based on a deep neural network, belonging to the technical field of intersection of transportation and artificial intelligence. Background Art
[0002] As the cornerstone of modern intelligent systems, trajectory prediction technology has shown wide application potential in many fields such as autonomous driving, intelligent traffic management, aerospace, and personal mobile devices. Its core lies in the use of advanced technologies such as big data analysis, machine learning, and deep learning to comprehensively process historical trajectory data and real-time sensor information to achieve accurate prediction of the future movement path of objects or individual targets. In the field of transportation, current trajectory prediction technology mainly focuses on improving the ability to recognize and predict target behavior patterns, especially for behavior patterns with high uncertainty and complexity. To achieve this goal, researchers have developed a variety of algorithms and models, such as Markov chain-based prediction models, deep learning networks, and hybrid models, aiming to capture and learn the potential movement patterns of traffic participants, thereby improving the long-distance stability and reliability of trajectory prediction. These technologies not only help optimize traffic flow management, but also significantly improve the safety and efficiency of autonomous driving systems.
[0003] Although existing trajectory prediction technologies have made significant progress in dealing with complex traffic scenarios and uncertain behaviors, there are still obvious technical bottlenecks in accurately predicting trajectories on specific paths. At present, many methods still rely on traditional physical motion models and simulation methods for prediction. Although these models can reflect the basic motion laws of objects to a certain extent, they are often difficult to accurately capture all the details when faced with complex and changeable dynamic factors in actual traffic environments, resulting in large modeling errors. In addition, these traditional methods are usually accompanied by high computational complexity, especially when dealing with large-scale data sets and real-time prediction needs, the consumption of computing resources and the delay in prediction time become issues that cannot be ignored.
[0004] Therefore, finding a better balance between the real-time performance and accuracy of prediction has become a major challenge facing current trajectory prediction technology. The lack of innovative solutions that can effectively combine high-precision prediction with low-latency computing has limited the application and development of trajectory prediction technology in more high-precision scenarios.
[0005] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgement or any form of suggestion that the information constitutes the prior art already known to ordinary technicians in this field. Summary of the invention
[0006] The technical problem to be solved by the present invention is that the prior art has deficiencies in terms of real-time and accuracy of prediction due to modeling errors and high computational complexity.
[0007] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions.
[0008] In a first aspect, the present invention provides a path-dependent trajectory prediction method based on a deep neural network, comprising:
[0009] Obtain a reference path of the object to be predicted;
[0010] The starting path point of the reference path is used as the reference path data point for current alignment, and prediction is performed based on a pre-trained trajectory prediction neural network, and the trajectory prediction result of the object to be predicted at the next moment is output, and random sampling is performed to obtain trajectory data points;
[0011] According to the trajectory data point, the corresponding reference data point on the path midline is determined as the reference path data point for current alignment, and a prediction is performed based on a pre-trained trajectory prediction neural network, and the trajectory prediction result of the object to be predicted at the next moment is output, and random sampling is performed to sequentially generate trajectory data points at subsequent moments to obtain a complete sequence of predicted trajectory data points; wherein the pre-training method of the trajectory prediction neural network includes:
[0012] Obtain the ideal path and actual trajectory of the object to be predicted;
[0013] According to the ideal path, path point data is extracted to obtain a reference path sequence;
[0014] Completing the missing data points in the actual running trajectory to obtain a trajectory input data sequence;
[0015] According to the reference path sequence, determining a reference data point corresponding to a path midline where a data point at a current moment in the trajectory input data sequence is located, to obtain an aligned reference path data sequence;
[0016] The trajectory input data sequence and the aligned reference path data sequence are used as training sets, input into a pre-built trajectory prediction neural network for training, and the training is terminated when the loss value of the loss function reaches a threshold, thereby obtaining a pre-trained trajectory prediction neural network.
[0017] Furthermore, the step of extracting path point data according to the ideal path to obtain a reference path sequence includes:
[0018] According to the electronic map, multiple ideal paths of the object to be predicted are obtained;
[0019] On the path centerline of each ideal path, path point data is extracted at fixed intervals to obtain a path data sequence;
[0020] Data preprocessing is performed on the path data sequence to obtain a reference path sequence.
[0021] Furthermore, the path point data includes the longitude, latitude, curvature radius and slope angle of the path point at the current moment, and the path data sequence is preprocessed to obtain a reference path sequence, including:
[0022] Normalizing each path point data in the path data sequence to obtain a reference path sequence,
[0023] The expression for normalizing the longitude of the path point is:
[0024] (1);
[0025] In the formula, Indicates The normalized value of the longitude of the path point. Indicates The longitude of the location of the waypoint, Indicates the minimum longitude of all path points. Indicates the maximum value of the longitude of all path points, where , represents the total number of path points in the path data sequence;
[0026] The expression for normalizing the latitude of the path point is:
[0027] (2);
[0028] In the formula, Indicates The normalized value of the latitude of the path point. Indicates The latitude of the location of the waypoint, Indicates the minimum latitude of all path points. Indicates the maximum latitude of all path points;
[0029] The expression for normalizing the curvature radius of the path point is:
[0030] (3);
[0031] In the formula, Indicates The normalized value of the curvature radius at the location of each path point. Indicates The radius of curvature at the location of the path point, Indicates the minimum value of the curvature radius of all path points. Indicates the maximum value of the curvature radius of all path points;
[0032] The expression for normalizing the slope angle value at the location of the path point is:
[0033] (4);
[0034] In the formula, Indicates The normalized value of the slope angle at the location of each path point, Indicates The latitude of the location of the waypoint, Indicates the minimum slope angle of all path points. Indicates the maximum value of the slope angles at all path points.
[0035] Furthermore, the missing data points in the actual running trajectory are supplemented to obtain a trajectory input data sequence, including:
[0036] The actual running trajectory is subjected to Gaussian filtering to remove noise interference, thereby obtaining a smoothed actual running trajectory;
[0037] Use polynomial interpolation to complete the missing data points in the actual running trajectory after smoothing to obtain a trajectory data sequence, wherein the data points include the longitude and latitude corresponding to the current position;
[0038] Data preprocessing is performed on the trajectory data sequence to obtain a trajectory input data sequence.
[0039] Furthermore, data preprocessing is performed on the trajectory data sequence to obtain a trajectory input data sequence including:
[0040] Calculate the longitude offset and latitude offset at the current moment according to the longitude and latitude of the data points in the trajectory data sequence at the previous moment and the current moment;
[0041] Normalizing the longitude offset and latitude offset at the current moment respectively to obtain normalized longitude offset and latitude offset;
[0042] The trajectory input data sequence is constructed according to the data point coordinates at the starting time of the trajectory input data sequence and the normalized longitude offset and latitude offset.
[0043] Furthermore, a corresponding reference data point on the path midline where any data point in the trajectory input data sequence is located is determined by a polynomial interpolation method to obtain an aligned reference path data sequence.
[0044] Furthermore, the loss function is expressed as:
[0045] (5);
[0046] In the formula, represents the loss value of the loss function, represents the current moment, where , It indicates the actual end time of the object to be predicted, log indicates the logarithmic function, represents the probability distribution function of the trajectory data in the training set, represents the trajectory input data at the current time t, The variance of the probability distribution of the offset of the latitude and longitude of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude of the data point in the trajectory input data sequence at the time t-1, Represents the mean of the probability distribution of the offset of the latitude and longitude of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude of the data point in the trajectory input data sequence at the time t-1.
[0047] Furthermore, the trajectory prediction neural network is constructed based on the gated recurrent unit; wherein the relationship between the trajectory prediction result output by the trajectory prediction neural network at the current moment and the input data is expressed as:
[0048] (6);
[0049] In the formula, represents the trajectory prediction result output by the trajectory prediction neural network at the current time t, represents the output of the fully connected layer, represents the state update function of the gated recurrent unit in the output layer, represents the state update function of the gated recurrent unit in the input layer, represents the feature embedding function implemented by the fully connected layer and the rectified linear unit, represents the reference path data point aligned at time t-1, represents the state vector of the gated recurrent unit in the input layer at time t-1, Represents the state vector of the gated recurrent unit in the output layer at time t-1.
[0050] In a second aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the method described in the first aspect when the computer program / instruction is executed by a processor.
[0051] In a third aspect, the present invention provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the method described in the first aspect.
[0052] Compared with the prior art, the present invention has the following beneficial effects:
[0053] The present invention significantly improves the real-time and accuracy of prediction through an innovative path-dependent trajectory prediction method based on deep neural networks, and effectively solves the deficiencies in the prior art caused by modeling errors and high computational complexity. First, the present invention obtains the reference path of the object to be predicted, and uses a pre-trained trajectory prediction neural network for prediction, thereby outputting the trajectory prediction results of the object to be predicted at all times. Secondly, the present invention further aligns the reference path data points based on the trajectory data points of the object to be predicted at the previous moment as the prediction input for the next moment, describes the trajectory prediction results at the next moment with a probability distribution, and performs random sampling to obtain the trajectory data points at the next moment, which can not only effectively represent the path dependency, but also help capture the diversity of trajectories and improve prediction robustness.
[0054] In terms of pre-training of the trajectory prediction neural network, the present invention obtains the ideal path and actual running trajectory of the object to be predicted, and processes them to obtain a reference path sequence and a trajectory input data sequence. At the same time, by determining the reference data point corresponding to the midline of the path where the data point at the current moment in the trajectory input data sequence is located, an aligned reference path data sequence is obtained. The above process fully considers the geometric characteristics of the path and the complexity of the actual running trajectory, which helps the deep learning network learn a more accurate trajectory prediction model. The processed data is input as a training set into the pre-constructed trajectory prediction neural network for training. When the loss value of the loss function reaches the threshold, the training is terminated to obtain a pre-trained trajectory prediction neural network, which improves the generalization ability of the network and ensures the stability and reliability of the prediction results.
[0055] In summary, the beneficial effects of the present invention are mainly reflected in the following aspects: first, by introducing reference paths and iterative prediction methods, the accuracy and real-time performance of the prediction are improved; second, by describing the trajectory prediction results through probability distribution, the prediction robustness is improved; third, by using a pre-trained trajectory prediction neural network, the stability and reliability of the prediction results are ensured. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1It is a flowchart of a path-dependent trajectory prediction method based on a deep neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0057] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.
[0058] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.
[0059] Example 1
[0060] like Figure 1 As shown, this embodiment introduces a path-dependent trajectory prediction method based on a deep neural network, which specifically includes the following steps:
[0061] Step 1: Get the ideal path of the object to be predicted.
[0062] According to the ideal path, the path point data is extracted to obtain the reference path sequence, including:
[0063] According to the electronic map, multiple ideal paths of the object to be predicted are obtained;
[0064] On the midline of each ideal path, path point data is extracted at fixed intervals of D meters to obtain a path data sequence containing K points. ,in, D ∈ [ 0 . 5 , 1 ] , Represents the first path point in the path data sequence, Represents the second path point in the path data sequence, Indicates the first Waypoints, , Indicates the total number of path points in the path data sequence.
[0065] The path point data includes the longitude, latitude, curvature radius and slope angle of the path point at the current moment, which is expressed as ,in, Indicates The longitude of the location of the waypoint, Indicates The latitude of the location of the waypoint, Indicates The radius of curvature at the location of the path point, Indicates The latitude of the waypoint's location.
[0066] Performing data preprocessing on the path data sequence to obtain a reference path sequence includes:
[0067] Normalize each path point data in the path data sequence to obtain a reference path sequence ,in, .
[0068] The expression for normalizing the longitude of the path point is:
[0069] (1);
[0070] In the formula, Indicates The normalized value of the longitude of the path point. Indicates the minimum longitude of all path points. Indicates the maximum longitude of all waypoints.
[0071] The expression for normalizing the latitude of the path point is:
[0072] (2);
[0073] In the formula, Indicates The normalized value of the latitude of the path point. Indicates the minimum latitude of all path points. Indicates the maximum latitude of all path points;
[0074] The expression for normalizing the curvature radius of the path point is:
[0075] (3);
[0076] In the formula, Indicates The normalized value of the curvature radius at the location of each path point. Indicates the minimum value of the curvature radius of all path points. Indicates the maximum value of the curvature radius of all path points;
[0077] The expression for normalizing the slope angle value at the location of the path point is:
[0078] (4);
[0079] In the formula, Indicates The normalized value of the slope angle at the location of each path point, Indicates the minimum slope angle of all path points. Indicates the maximum value of the slope angles at all path points.
[0080] Step 2: Get the actual trajectory of the object to be predicted.
[0081] The missing data points in the actual running trajectory are completed to obtain a trajectory input data sequence.
[0082] The missing data points in the actual running trajectory are supplemented to obtain a trajectory input data sequence, including:
[0083] The actual running trajectory is subjected to Gaussian filtering to remove noise interference, thereby obtaining a smoothed actual running trajectory;
[0084] Polynomial interpolation is used to complete the missing data points in the actual running trajectory after smoothing to obtain the trajectory data sequence ,in, The first data point in the trajectory data sequence, The second data point in the trajectory data sequence, The trajectory data sequence data points, Indicates the actual end time of the operation of the object to be predicted.
[0085] Among them, the data points Including the longitude and latitude corresponding to the location of the path point at the current time t, , Indicates the longitude corresponding to the location of the data point at the current time t, Indicates the latitude corresponding to the location of the data point at the current time t.
[0086] The trajectory data sequence is subjected to data preprocessing to obtain a trajectory input data sequence, including:
[0087] Calculate the longitude offset at the current time according to the longitude and latitude of the data points in the trajectory data sequence at the previous time t-1 and the current time t and latitude offset , where x represents longitude and y represents latitude.
[0088] The longitude offset of the current time and latitude offset Normalize to get the normalized longitude offset and latitude offset ;
[0089] According to the trajectory input data sequence starting time data point coordinates and the normalized longitude offset and latitude offset Construct trajectory input data sequence ,in, Indicates the data point coordinates at the next moment after the start moment of the trajectory input data sequence, Represents the data point coordinates at the end of the trajectory input data sequence.
[0090] Step 3: Train the trajectory prediction neural network.
[0091] Firstly, the trajectory prediction neural network is constructed based on the gated recurrent unit.
[0092] According to the reference path sequence , determine the trajectory input data sequence by polynomial interpolation method The data point at the current moment in The reference data point corresponding to the center line of the path , get the aligned reference path data sequence ;
[0093] Enter the trajectory into the data sequence and aligned reference pathway data sequences As a training set, the pre-built trajectory prediction neural network is trained, and the training is terminated when the loss value of the loss function reaches a threshold, thereby obtaining a pre-trained trajectory prediction neural network.
[0094] The input data of the trajectory prediction neural network is the aligned reference path data sequence , the output data is the trajectory prediction result ,in, represents the trajectory prediction result at the starting time, It represents the trajectory prediction result of the next moment after the starting moment. represents the trajectory prediction result at the termination time, , that is, the mean is , the variance is The normal distribution of The variance of the probability distribution of the latitude and longitude offset of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude offset of the data point in the trajectory input data sequence at the time t-1. It represents the mean of the probability distribution of the latitude and longitude offset of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude offset of the data point in the trajectory input data sequence at the time t-1.
[0095] Among them, the trajectory prediction result output by the trajectory prediction neural network at the current time t is The relationship with the input data is expressed as:
[0096] (5);
[0097] In the formula, represents the trajectory prediction result output by the trajectory prediction neural network at the current time t, represents the output of the fully connected layer, represents the state update function of the gated recurrent unit in the output layer, represents the state update function of the gated recurrent unit in the input layer, represents the feature embedding function implemented by the fully connected layer and the rectified linear unit, represents the reference path data point aligned at time t-1, represents the state vector of the gated recurrent unit in the input layer at time t-1, Represents the state vector of the gated recurrent unit in the output layer at time t-1.
[0098] The loss function is expressed as follows:
[0099] (6);
[0100] In the formula, represents the loss value of the loss function, represents the current moment, where , It indicates the actual end time of the object to be predicted, log indicates the logarithmic function, represents the probability distribution function of the trajectory data in the training set, represents the trajectory input data at the current time t, The variance of the probability distribution of the offset of the latitude and longitude of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude of the data point in the trajectory input data sequence at the time t-1, Represents the mean of the probability distribution of the offset of the latitude and longitude of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude of the data point in the trajectory input data sequence at time t-1.
[0101] Step 4: Get the reference path of the object to be predicted.
[0102] Get the reference path of the object to be predicted ,in, represents the first path point in the reference path, represents the first path point in the reference path, represents the last waypoint in the reference path, Represents the total number of all path points in the reference path.
[0103] Step 5: Trajectory prediction.
[0104] The reference path The first waypoint in , input the pre-trained trajectory prediction neural network, make predictions, and output the trajectory prediction results of the object to be predicted at time t=1 ;
[0105] According to the trajectory prediction result of the object to be predicted at time t=1 , perform random sampling to obtain the trajectory data point at time t=1 ;
[0106] For t=2,…,T, according to the predicted trajectory data point at time t-1 , determined by polynomial interpolation method Corresponding reference path data points ,Will Input the pre-trained trajectory prediction neural network, make predictions, and get the prediction results at time t ;
[0107] According to the trajectory prediction result of the object to be predicted at time t , perform random sampling to obtain the trajectory data points at time t .
[0108] Example 2
[0109] A computer-readable storage medium having computer instructions stored thereon, characterized in that the computer instructions implement the steps of the method in the above-mentioned embodiment 1 when executed by a processor.
[0110] Example 3
[0111] The present invention also provides a computer program product, including computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the method in the above-mentioned embodiment 1 are implemented.
[0112] In summary, the beneficial effects of the present invention are mainly reflected in the following aspects: first, by introducing reference paths and iterative prediction methods, the accuracy and real-time performance of the prediction are improved; second, by describing the trajectory prediction results through probability distribution, the prediction robustness is improved; third, by using a pre-trained trajectory prediction neural network, the stability and reliability of the prediction results are ensured.
[0113] The present invention significantly improves the real-time and accuracy of prediction through an innovative path-dependent trajectory prediction method based on deep neural networks, and effectively solves the deficiencies in the prior art caused by modeling errors and high computational complexity. First, the present invention obtains the reference path of the object to be predicted, and uses a pre-trained trajectory prediction neural network for prediction, thereby outputting the trajectory prediction results of the object to be predicted at all times. Secondly, the present invention further aligns the reference path data points based on the trajectory data points of the object to be predicted at the previous moment as the prediction input for the next moment, describes the trajectory prediction results at the next moment with a probability distribution, and performs random sampling to obtain the trajectory data points at the next moment, which can not only effectively represent the path dependency, but also help capture the diversity of trajectories and improve prediction robustness.
[0114] In terms of pre-training of the trajectory prediction neural network, the present invention obtains the ideal path and actual running trajectory of the object to be predicted, and processes them to obtain a reference path sequence and a trajectory input data sequence. At the same time, by determining the reference data point corresponding to the midline of the path where the data point at the current moment in the trajectory input data sequence is located, an aligned reference path data sequence is obtained. The above process fully considers the geometric characteristics of the path and the complexity of the actual running trajectory, which helps the deep learning network learn a more accurate trajectory prediction model. The processed data is input as a training set into the pre-constructed trajectory prediction neural network for training. When the loss value of the loss function reaches the threshold, the training is terminated to obtain a pre-trained trajectory prediction neural network, which improves the generalization ability of the network and ensures the stability and reliability of the prediction results.
[0115] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0116] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0117] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0119] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.
Claims
1. A path-dependent trajectory prediction method based on deep neural network, characterized in that: include: Obtain a reference path of the object to be predicted; The starting path point of the reference path is used as the reference path data point for current alignment, and prediction is performed based on a pre-trained trajectory prediction neural network, and the trajectory prediction result of the object to be predicted at the next moment is output, and random sampling is performed to obtain trajectory data points; According to the trajectory data points, the corresponding reference data points on the path midline are determined as the reference path data points for the current alignment, and prediction is performed based on the pre-trained trajectory prediction neural network, and the trajectory prediction result of the object to be predicted at the next moment is output, and random sampling is performed to generate trajectory data points at subsequent moments in sequence to obtain a complete sequence of predicted trajectory data points; The pre-training method of the trajectory prediction neural network includes: Obtain the ideal path and actual trajectory of the object to be predicted; According to the ideal path, path point data is extracted to obtain a reference path sequence; Completing the missing data points in the actual running trajectory to obtain a trajectory input data sequence; According to the reference path sequence, determining a reference data point corresponding to a path midline where a data point at a current moment in the trajectory input data sequence is located, to obtain an aligned reference path data sequence; The trajectory input data sequence and the aligned reference path data sequence are used as training sets, input into a pre-built trajectory prediction neural network for training, and the training is terminated when the loss value of the loss function reaches a threshold, thereby obtaining a pre-trained trajectory prediction neural network.
2. The path-dependent trajectory prediction method based on deep neural network according to claim 1 is characterized in that: The step of extracting path point data according to the ideal path to obtain a reference path sequence includes: According to the electronic map, multiple ideal paths of the object to be predicted are obtained; On the path centerline of each ideal path, path point data is extracted at fixed intervals to obtain a path data sequence; Data preprocessing is performed on the path data sequence to obtain a reference path sequence.
3. The path-dependent trajectory prediction method based on deep neural network according to claim 2 is characterized in that: The path point data includes the longitude, latitude, curvature radius and slope angle of the path point at the current moment. The path data sequence is preprocessed to obtain a reference path sequence, including: Normalizing each path point data in the path data sequence to obtain a reference path sequence, The expression for normalizing the longitude of the path point is: （1）； In the formula, Indicates The normalized value of the longitude of the path point. Indicates The longitude of the location of the waypoint, Indicates the minimum longitude of all path points. Indicates the maximum value of the longitude of all path points, where , represents the total number of path points in the path data sequence; The expression for normalizing the latitude of the path point is: （2）； In the formula, Indicates The normalized value of the latitude of the path point. Indicates The latitude of the location of the waypoint, Indicates the minimum latitude of all path points. Indicates the maximum latitude of all path points; The expression for normalizing the curvature radius of the path point is: （3）； In the formula, Indicates The normalized value of the curvature radius at the location of each path point. Indicates The radius of curvature at the location of the path point, Indicates the minimum value of the curvature radius of all path points. Indicates the maximum value of the curvature radius of all path points; The expression for normalizing the slope angle value at the location of the path point is: （4）； In the formula, Indicates The normalized value of the slope angle at the location of each path point, Indicates The latitude of the location of the waypoint, Indicates the minimum slope angle of all path points. Indicates the maximum value of the slope angles at all path points.
4. The path-dependent trajectory prediction method based on deep neural network according to claim 1 is characterized in that: The step of completing the missing data points in the actual running trajectory to obtain a trajectory input data sequence includes: The actual running trajectory is subjected to Gaussian filtering to remove noise interference, thereby obtaining a smoothed actual running trajectory; Use polynomial interpolation to complete the missing data points in the actual running trajectory after smoothing to obtain a trajectory data sequence, wherein the data points include the longitude and latitude corresponding to the current position; Data preprocessing is performed on the trajectory data sequence to obtain a trajectory input data sequence.
5. The path-dependent trajectory prediction method based on deep neural network according to claim 4 is characterized in that: The trajectory data sequence is subjected to data preprocessing to obtain a trajectory input data sequence including: Calculate the longitude offset and latitude offset at the current moment according to the longitude and latitude of the data points in the trajectory data sequence at the previous moment and the current moment; Normalizing the longitude offset and latitude offset at the current moment respectively to obtain normalized longitude offset and latitude offset; The trajectory input data sequence is constructed according to the data point coordinates at the starting time of the trajectory input data sequence and the normalized longitude offset and latitude offset.
6. The path-dependent trajectory prediction method based on deep neural network according to claim 1 is characterized in that: The reference data point corresponding to the path midline where any data point in the trajectory input data sequence is located is determined by a polynomial interpolation method to obtain an aligned reference path data sequence.
7. The path-dependent trajectory prediction method based on deep neural network according to claim 5 is characterized in that: The expression of the loss function is expressed as: （5）； In the formula, represents the loss value of the loss function, represents the current moment, where , It indicates the actual end time of the object to be predicted, log indicates the logarithmic function, represents the probability distribution function of the trajectory data in the training set, represents the trajectory input data at the current time t, The variance of the probability distribution of the offset of the latitude and longitude of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude of the data point in the trajectory input data sequence at the time t-1, Represents the mean of the probability distribution of the offset of the latitude and longitude of the data point in the trajectory input data sequence at the current time t relative to the latitude and longitude of the data point in the trajectory input data sequence at the time t-1.
8. The path-dependent trajectory prediction method based on deep neural network according to claim 1 is characterized in that: The trajectory prediction neural network is constructed based on the gated recurrent unit; wherein the relationship between the trajectory prediction result output by the trajectory prediction neural network at the current moment and the input data is expressed as: （6）； In the formula, represents the trajectory prediction result output by the trajectory prediction neural network at the current time t, represents the output of the fully connected layer, represents the state update function of the gated recurrent unit in the output layer, represents the state update function of the gated recurrent unit in the input layer, represents the feature embedding function implemented by the fully connected layer and the rectified linear unit, represents the reference path data point aligned at time t-1, represents the state vector of the gated recurrent unit in the input layer at time t-1, Represents the state vector of the gated recurrent unit in the output layer at time t-1.
9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.
10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.