Travel mode sensing trajectory generation method and device based on autoregressive diffusion model, storage medium and equipment

Abstract

The invention provides a travel mode sensing trajectory generation method and device based on an autoregressive diffusion model, a storage medium and equipment. The method comprises the following steps: extracting an individual travel mode from historical trajectory data, training an autoregression time sequence planning model by using the individual travel mode as a training sample, and training a travel mode condition guide diffusion model by using the historical trajectory data and the individual travel mode as training samples together; extracting individual space anchor point data from the residential population distribution data, inputting the individual space anchor point data into the trained autoregression time sequence planning model to obtain a future time mode sequence, and inputting the individual space anchor point data and the future time mode sequence into a trained travel mode condition guide diffusion model to obtain a travel mode condition guide diffusion model; and obtaining a future travel trajectory. According to the method, the individual travel mode is fully considered, so that the generated trajectory conforms to the individual historical behavior in time, the periodic movement mode of the individual can be accurately reproduced, and the spatial and temporal distribution of the trajectory is more real.

Description

Technical Field

[0001] This application belongs to the technical field of geographic information systems. Specifically, it relates to a travel pattern perception trajectory generation method, device, storage medium, and equipment based on an autoregressive diffusion model. Background Art

[0002] With the advancement of urbanization and the popularization of smart devices, human mobility data (such as location trajectory data) has become a valuable resource in the fields of studying individual behavior, urban management, traffic planning, epidemic control, etc. Trajectory Generation (TG) technology is convenient for simulating and predicting individual movement behaviors and has important application value. However, when simulating individual movement behaviors, traditional trajectory generation methods usually ignore individual-specific travel patterns, such as frequently visited locations, movement ranges, time regularities, etc., which are key factors affecting individual mobility behaviors. Therefore, existing methods have significant limitations in capturing complex individual mobility behaviors and ensuring the spatio-temporal consistency of generated trajectories.

[0003] In trajectory generation, as an emerging generation method, the unconditional diffusion model has improved the deficiencies of traditional methods to some extent, but it still has some problems. The unconditional diffusion model usually does not consider the travel pattern characteristics of individuals, resulting in generated trajectories that fail to reflect the behavioral patterns of individuals at different times and locations. This defect makes the generated trajectories lack spatio-temporal consistency and cannot effectively capture complex mobility behaviors.

[0004] Currently, research on trajectory generation mainly focuses on the following technical solutions:

[0005] (1) Mechanics mechanism-driven method: The mechanism-driven model simulates individual mobility behaviors through a set of clear assumptions and parameters. For example, the EPR model simulates the movement of individuals in space by finding a balance between exploration and revisit [1]. However, this type of method is overly dependent on fixed model assumptions and thus has certain limitations in modeling complex human mobility behaviors. These methods usually cannot handle complex and non-linear human behaviors and cannot consider individual-specific time and space regularities.

[0006] (2) Deep learning-based trajectory generation method: With the development of deep learning technology, trajectory generation methods based on deep neural networks have gradually been applied. Typical deep learning methods include Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). For example, methods like Movesim and VOLUNTEER generate trajectories by learning the data distribution. However, existing deep learning methods, especially unconditional generation models, usually ignore individual travel patterns and their spatio-temporal structure. The generated trajectories lack spatio-temporal consistency and cannot effectively capture individual behavior patterns. In recent years, diffusion models, as a new type of generation method, such as Difftraj and TrajGDM, have been applied to the field of trajectory generation. Diffusion models simulate a random process through an iterative process and generate samples from noise. However, existing unconditional diffusion models fail to effectively incorporate individual travel pattern information and can only generate random trajectories unrelated to individual movement behaviors. Therefore, although the generated trajectories have a certain spatial structure, they often lack time series patterns that conform to individual specific behavior patterns. Summary of the Invention

[0007] The technical problem solved by this application is: How to fully consider individual travel pattern characteristics during the travel pattern-aware trajectory generation process to enable the model to more accurately capture individual flow behaviors and improve spatio-temporal consistency.

[0008] This application provides a travel pattern-aware trajectory generation method based on an autoregressive diffusion model. The travel pattern-aware trajectory generation method includes:

[0009] Model training stage: Extract individual travel patterns from historical trajectory data, use the individual travel patterns as training samples to train an autoregressive time series planning model, and use the historical trajectory data and individual travel patterns together as training samples to train a travel pattern-conditioned guided diffusion model;

[0010] Trajectory generation stage: Extract individual spatial anchor point data from the residential population distribution data, input the individual spatial anchor point data into the trained autoregressive time series planning model to obtain a future time pattern sequence, and input the individual spatial anchor point data and the future time pattern sequence into the trained travel pattern-conditioned guided diffusion model to obtain a future travel trajectory.

[0011] Optionally, the individual travel pattern includes spatial anchor point data and a historical time pattern sequence.

[0012] Optionally, the method for extracting spatial anchor point data from historical trajectory data includes:

[0013] Determine the home location data of each individual from the historical trajectory data, and use the home location data as the spatial anchor point data.

[0014] Optionally, the method for extracting the historical time pattern sequence from the historical trajectory data includes:

[0015] Recode each stop location in the historical trajectory data in chronological order to generate a historical time pattern sequence.

[0016] Optionally, the method of using the historical trajectory data and the individual travel pattern as training samples to train the travel pattern conditional guided diffusion model includes:

[0017] In the forward diffusion process, convert the historical trajectory data into continuous latent space vectors and gradually add noise to obtain completely noisy trajectory data;

[0018] In the reverse diffusion process, use the spatial anchor point data and the historical time pattern sequence as constraint conditions to denoise the completely noisy trajectory data, recover the trajectory data, and adjust the model parameters of the travel pattern conditional guided diffusion model according to the recovery result to complete one training.

[0019] Optionally, the noise gradually added to the continuous latent space vector is Gaussian noise.

[0020] Optionally, the method further includes: performing a MeanClamp operation during the denoising process of the completely noisy trajectory data to average the repeated positions in the trajectory.

[0021] The present application also provides a travel pattern perception trajectory generation device based on an autoregressive diffusion model. The travel pattern perception trajectory generation device includes:

[0022] A model training module, which is used to extract the individual travel pattern from the historical trajectory data, use the individual travel pattern as a training sample to train an autoregressive time series planning model, and use the historical trajectory data and the individual travel pattern as training samples to train a travel pattern conditional guided diffusion model;

[0023] A trajectory generation module, which is used to extract individual spatial anchor point data from the residential population distribution data, input the individual spatial anchor point data into the trained autoregressive time series planning model to obtain a future time pattern sequence, and input the individual spatial anchor point data and the future time pattern sequence into the trained travel pattern conditional guided diffusion model to obtain a future travel trajectory.

[0024] The present application also provides a computer-readable storage medium, which stores a travel pattern perception trajectory generation program based on an autoregressive diffusion model. When the travel pattern perception trajectory generation program based on the autoregressive diffusion model is executed by a processor, it implements the above-mentioned travel pattern perception trajectory generation method based on an autoregressive diffusion model.

[0025] This application also provides a computer device, which includes a computer-readable storage medium, a processor, and a travel pattern perception trajectory generation program based on an autoregressive diffusion model stored in the computer-readable storage medium. When the travel pattern perception trajectory generation program based on the autoregressive diffusion model is executed by the processor, it implements the above-mentioned travel pattern perception trajectory generation method based on the autoregressive diffusion model.

[0026] A travel pattern perception trajectory generation method, device, storage medium, and device based on an autoregressive diffusion model provided by this application have the following technical effects:

[0027] Combining an autoregressive time-series planning model and a travel pattern conditional guidance diffusion model, fully considering individual travel patterns, makes the generated trajectory conform to individual historical behaviors in time, can accurately reproduce the periodic movement patterns of individuals, and introduces individual travel patterns as guidance in the diffusion process, making the spatio-temporal distribution of the trajectory more realistic and avoiding spatial drift caused by random generation. In addition, generating based on time patterns and spatial anchor points makes each generated trajectory a completely new synthetic trajectory, avoiding directly copying real data and reducing the risk of privacy leakage.

[0028] At the same time, this method can be used in a variety of application scenarios: Smart city management: Provide high-quality synthetic trajectory data for traffic optimization and urban planning. Epidemic prevention and control: Simulate the movement patterns of the crowd to support epidemic prevention strategies. Privacy-protected trajectory generation: Used to generate synthetic data similar to real data but without revealing individual privacy, suitable for data sharing and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a main step flowchart of a travel pattern perception trajectory generation method based on an autoregressive diffusion model according to one or more embodiments;

[0030] Figure 2 It is a detailed process schematic diagram of a travel pattern perception trajectory generation method based on an autoregressive diffusion model according to one or more embodiments;

[0031] Figure 3 It is an experimental result diagram of spatio-temporal consistency verification of trajectory generation according to one or more embodiments;

[0032] Figure 4 It is an experimental verification result diagram of the privacy protection of generated trajectories according to one or more embodiments;

[0033] Figure 5 It is an experimental result diagram of OD flow distribution analysis of generated trajectories according to one or more embodiments;

[0034] Figure 6Schematic block diagram of a travel pattern-aware trajectory generation device based on an autoregressive diffusion model according to one or more embodiments;

[0035] Figure 7 Schematic diagram of a computer device according to one or more embodiments. Detailed implementation manners

[0036] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0037] Before describing each embodiment of the present application in detail, the technical concept of the present application will be briefly described first: In current trajectory generation methods, individual travel patterns are usually ignored, making it difficult to accurately capture individual mobility behaviors and resulting in poor spatio-temporal consistency. For this reason, the travel pattern-aware trajectory generation method based on an autoregressive diffusion model provided by the present application combines an autoregressive time-series planning model and a travel pattern-conditioned guided diffusion model, fully considering individual travel patterns, making the generated trajectory conform to the individual's historical behavior in time, being able to accurately reproduce the individual's periodic movement pattern, and introducing the individual travel pattern as a guide in the diffusion process, making the spatio-temporal distribution of the trajectory more realistic and avoiding spatial drift caused by random generation. The specific principle of the travel pattern-aware trajectory generation method based on an autoregressive diffusion model of the present application will be described below with more embodiments.

[0038] Specifically, as Figure 1 shown, the travel pattern-aware trajectory generation method based on an autoregressive diffusion model in Embodiment 1 includes the following steps:

[0039] Step S10, model training stage: Extract individual travel patterns from historical trajectory data, use the individual travel patterns as training samples to train an autoregressive time-series planning model, and use the historical trajectory data and individual travel patterns together as training samples to train a travel pattern-conditioned guided diffusion model;

[0040] Step S20, trajectory generation stage: Extract individual spatial anchor point data from the residential population distribution data, input the individual spatial anchor point data into the trained autoregressive time-series planning model to obtain a future time pattern sequence, and input the individual spatial anchor point data and the future time pattern sequence into the trained travel pattern-conditioned guided diffusion model to obtain a future travel trajectory.

[0041] In one or more embodiments, an individual travel pattern includes spatial anchor data and a historical time pattern sequence. Exemplarily, the method for extracting spatial anchor data from historical trajectory data includes: determining the home location data of each individual from the historical trajectory data, and using the home location data as the spatial anchor data. The home location is determined based on the location where each individual stays the longest at night. Using the home location data of the individual as the spatial anchor data is used to constrain the spatial range of the individual. Exemplarily, the method for extracting the historical time pattern sequence from historical trajectory data includes: re-coding each stop location in the historical trajectory data in chronological order to generate a historical time pattern sequence. The historical time pattern sequence reflects the order in which an individual visits different locations over a period of time. The extraction process of the individual travel pattern combines the spatio-temporal dependencies in the trajectory data, ensuring an accurate modeling of the individual's movement behavior.

[0042] Furthermore, the autoregressive time-series planning model (AR-TempPlan) is a key component in this embodiment for generating the time pattern sequence. In the process of extracting the travel pattern of an individual, we obtain the spatial anchor (home location) and time pattern related to the individual. The goal of the autoregressive model is to generate a future time pattern sequence that conforms to the individual's historical behavior based on this information, so as to more accurately simulate the individual's future movement behavior.

[0043] Exemplarily, as Figure 2 shown, the input part of the autoregressive time-series planning model includes the home location and the time pattern sequence of the historical trajectory. The home location, which is the home of the individual, serves as a spatial anchor, providing spatial constraints for the generated time pattern and ensuring that the trajectory does not deviate from the real life range during the generation process. The time pattern sequence extracted from the historical data, which includes each stop point of the individual and its visiting order, reflects the individual's time behavior pattern over a period of time. The output part of the autoregressive time-series planning model is a new time pattern sequence, representing the possible visiting behavior of the individual in the future period of time. The generated time pattern sequence will consider the individual's time regularity, that is, the individual may revisit the same location at certain specific time periods or follow a specific movement rule.

[0044] Specifically, the processing process of the autoregressive time-series planning model includes the following steps: Step 1, time pattern encoding: Convert the stop points and access order in the historical trajectory data into time pattern encoding to ensure that the access behavior at each time point can be accurately reflected. Step 2, sequence prediction: Use the autoregressive model to iteratively predict the time pattern, and generate possible future time behaviors based on the home location and historical time pattern. Step 3, output sequence generation: The finally generated time pattern sequence not only conforms to the individual's historical behavior, but also provides temporal guidance for subsequent trajectory generation. Through this technology, the model can effectively capture the individual's temporal regularity, such as regular trips or regular activities during certain periods. Among them, the autoregressive time-series planning model uses cross-entropy loss as the loss function during the training process and iteratively updates the model parameters.

[0045] In one or more embodiments, the training process of the travel mode conditional guided diffusion model mainly includes two parts, namely the forward diffusion process and the reverse diffusion process. In the forward diffusion process, the historical trajectory data is converted into a continuous latent space vector and noise is gradually added to obtain completely noisy trajectory data; in the reverse diffusion process, the spatial anchor data and the historical time pattern sequence are used as constraint conditions to denoise the completely noisy trajectory data, restore the trajectory data, and adjust the model parameters of the travel mode conditional guided diffusion model according to the restoration result to complete one training. Among them, the noise added in the forward diffusion process introduces randomness into the generation process, ensuring the diversity of the generated trajectories and avoiding only generating paths highly similar to the historical trajectories. Exemplarily, the noise gradually added to the continuous latent space vector is Gaussian noise. When denoising in the reverse diffusion process, the noise signal is gradually converted into a spatial trajectory that conforms to spatio-temporal laws. The output of the travel mode conditional guided diffusion model is a new trajectory sequence, representing the moving positions of an individual at different time points. The generation of these positions not only considers the individual's spatial anchor (home location), but also strictly follows the temporal constraints of the time pattern sequence generated by the autoregressive model.

[0046] Specifically, in the forward diffusion process, the model converts the original trajectory data into continuous latent space vectors using a pre-trained embedding matrix and gradually adds noise until the data becomes completely random. This process helps the model learn how to recover the trajectory from the noise. In the reverse diffusion process, the model starts from a completely noisy trajectory, gradually removes the noise, and generates a real trajectory that conforms to the individual's movement behavior through spatial anchor points and historical time pattern sequences. This process is completed through iterative optimization, and each optimization depends on the previously generated trajectory and the noise information at the current moment. Among them, in the generation process, the diffusion model ensures the spatio-temporal consistency of the trajectory through the following technical details. One is spatial anchor guidance: the home location serves as a spatial anchor, ensuring that the generated trajectory always revolves around the individual's home in space and conforms to the individual's spatial behavior pattern. The second is time pattern guidance: according to the generation of the time pattern sequence, it ensures that the position changes in the trajectory are consistent with the individual's time pattern, that is, the individual will visit certain locations during specific time periods. The third is denoising operation: gradually eliminate the noise through the denoising process, and at the same time adjust the trajectory position through the feedback of the model to ensure that the spatio-temporal rules are accurately followed.

[0047] In one or more embodiments, in order to reduce the rounding error that may occur when generating a trajectory during the diffusion process, a MeanClamp operation is performed during the denoising process of the completely noisy trajectory data, and the repeated positions in the trajectory are averaged to ensure the spatio-temporal consistency of the trajectory.

[0048] Furthermore, in the trajectory generation stage, individual spatial anchor point data is extracted from the residential population distribution data, and the individual spatial anchor point data is input into the trained autoregressive time-series planning model to obtain the future time pattern sequence. The individual spatial anchor point data is input into the trained travel pattern conditional guidance diffusion model, and the future time pattern sequence is used as a constraint condition to obtain the future travel trajectory. Among them, after the travel pattern conditional guidance diffusion model outputs the trajectory data, a continuous-to-discrete mapping is performed to obtain the synthetic trajectory, avoiding directly copying real data and reducing the risk of privacy leakage.

[0049] This embodiment is verified through experiments, simulations, and real data, demonstrating the superiority of the travel pattern-aware trajectory generation method based on the autoregressive diffusion model of this embodiment in terms of spatio-temporal consistency of trajectory generation, individual behavior pattern capture, privacy protection, and computational efficiency. The experiments were conducted on real trajectory datasets (Beijing, Shenzhen) and were compared and analyzed with the existing state-of-the-art methods (GAN, RNN, traditional diffusion models). The experimental results show that this method can effectively generate high-fidelity trajectory data and maintain privacy security.

[0050] (1) Verification of spatio-temporal consistency of trajectory generation

[0051] Experimental objective: To evaluate whether the generated trajectories can truly reflect an individual's travel patterns, including spatial distribution (whether it conforms to the real geographical range) and temporal regularity (whether it conforms to historical behaviors).

[0052] Experimental method: Train on real urban trajectory data of Beijing and Shenzhen (from Tencent and China Unicom data), and use the generation method of this embodiment to generate individual trajectories.

[0053] Adopt multiple evaluation metrics, including: Displacement: Measure whether the moving distance of trajectory points matches the real trajectory. Radius: Measure whether the generated trajectory conforms to an individual's spatial activity range. Duration: Evaluate whether an individual's stay pattern is reasonable. LocNum: Evaluate whether the diversity of the generated trajectory matches the real trajectory. G-rank / I-rank: Evaluate the temporal regularity of the trajectory.

[0054] Experimental results: As Figure 3 shown, for short-term trajectory generation (24 hours): Compared with GAN / RNN methods (such as SeqGAN, FC-LSTM), the trajectory generation error (JS divergence) of this method is reduced by 40%, and it can better match the real trajectory. The Displacement and Radius errors of this method are reduced by 35% and 50% respectively, that is, the spatial distribution of the trajectory is more reasonable. For long-term trajectory generation (168 hours): In the long-term trajectory generation task, the LocNum and G-rank metrics of this method are improved by 20%-30% compared with traditional methods, indicating that the generated trajectory can better retain an individual's long-term behavior patterns.

[0055] (2) Privacy protection of generated trajectories

[0056] Experimental objective: To evaluate whether the generated trajectories can prevent privacy leakage while maintaining data authenticity, and avoid the generated trajectories being overly similar to real trajectories.

[0057] Experimental method: Similarity test (Levenshtein distance): Calculate the similarity between the generated trajectory and the real trajectory to ensure that it does not overly match a certain real individual. Membership Inference Attack (MIA): Use models such as Random Forest (RF), Multilayer Perceptron (MLP) and Logistic Regression (LR) to test whether an attacker can identify whether the generated trajectory comes from the training data.

[0058] Experimental results: As Figure 4As shown, the similarity test results: The Levenshtein distances between the generated trajectories and the real data mainly distribute between 0.3 and 0.8, indicating that the data not only conforms to the real trajectory pattern but also has sufficient differences. The member inference attack results: The privacy protection mechanism of this method effectively reduces the attack success rate, and the attack success rate is always lower than 60%.

[0059] (3) Analysis of the OD flow distribution of the generated trajectories

[0060] Experimental objectives: Evaluate whether the OD (Origin-Destination) distribution of the trajectories generated by this method conforms to the real urban travel patterns, and ensure the rationality and spatial consistency of the generated trajectories. Compare the OD distributions of this method with those of GAN (SeqGAN) and the traditional diffusion model (DiffTraj) to verify the advantages of this method.

[0061] Experimental results: As Figure 5 shown, to more intuitively display the OD (Origin-Destination) distribution of the trajectories generated by this method and its comparison with the real data, compare the OD flow results of the generated data (Traveller, SeqGAN, and DiffTraj) with the real data. This method is significantly better than other methods and has better results in the CPC (Common Part of Commuters) metric.

[0062] As Figure 6 shown, in one or more embodiments, the travel pattern-aware trajectory generation device based on the autoregressive diffusion model includes a model training module 100 and a trajectory generation module 200. The model training module 100 is used to extract individual travel patterns from historical trajectory data, use the individual travel patterns as training samples to train the autoregressive time series planning model, and use the historical trajectory data and the individual travel patterns together as training samples to train the travel pattern-conditioned guided diffusion model. The trajectory generation module 200 is used to extract individual spatial anchor data from the residential population distribution data, input the individual spatial anchor data into the trained autoregressive time series planning model to obtain the future time pattern sequence, and input the individual spatial anchor data and the future time pattern sequence into the trained travel pattern-conditioned guided diffusion model to obtain the future travel trajectory. The more detailed working processes of the model training module 100 and the trajectory generation module 200 can refer to the relevant descriptions of the embodiments of the travel pattern-aware trajectory generation method based on the autoregressive diffusion model, which will not be elaborated here.

[0063] In one or more embodiments, a computer-readable storage medium stores a travel pattern perception trajectory generation program based on an autoregressive diffusion model. When the travel pattern perception trajectory generation program based on the autoregressive diffusion model is executed by a processor, the travel pattern perception trajectory generation method based on the autoregressive diffusion model in the above embodiments is implemented.

[0064] In one or more embodiments, a computer device includes a computer-readable storage medium, a processor, and a travel pattern perception trajectory generation program based on an autoregressive diffusion model stored in the computer-readable storage medium. When the travel pattern perception trajectory generation program based on the autoregressive diffusion model is executed by the processor, the travel pattern perception trajectory generation method based on the autoregressive diffusion model described above is implemented. At the hardware level, as Figure 7 shown, the computer device includes a processor 12, an internal bus 13, a network interface 14, and a computer-readable storage medium 11. The processor 12 reads the corresponding computer program from the computer-readable storage medium and then runs it, forming a request processing device at the logical level. Of course, in addition to the software implementation, one or more embodiments of this specification do not exclude other implementation methods, such as logic devices or the combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logical unit, but can also be hardware or logic devices.

[0065] Exemplarily, a computer-readable storage medium includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette tapes, disk storage, quantum memory, graphene-based storage media, or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device.

[0066] The specific embodiments of the present application have been described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments can be modified and perfected without departing from the principles and spirit of the present application defined by the claims and their equivalents, and these modifications and improvements should also be within the protection scope of the present application.

Claims

1. A travel mode perception trajectory generation method based on an autoregressive diffusion model, characterized in that: The travel mode perception trajectory generation method includes: Model training phase: extract individual travel patterns from historical trajectory data, use individual travel patterns as training samples to train the autoregressive time series planning model, and use historical trajectory data and individual travel patterns as training samples to train the travel pattern conditional guided diffusion model; Trajectory generation stage: Individual spatial anchor point data are extracted from the residential population distribution data, and the individual spatial anchor point data are input into the trained autoregressive time series planning model to obtain the future time pattern sequence. The individual spatial anchor point data and the future time pattern sequence are input into the trained travel mode conditional guided diffusion model to obtain the future travel trajectory.

2. The method for generating travel mode perception trajectories based on the autoregressive diffusion model according to claim 1 is characterized in that: The individual travel pattern includes spatial anchor point data and historical time pattern sequences.

3. The method for generating travel mode perception trajectories based on the autoregressive diffusion model according to claim 2 is characterized in that: Methods for extracting spatial anchor data from historical trajectory data include: The home location data of each individual is determined from the historical trajectory data, and the home location data is used as the spatial anchor data.

4. The method for generating travel mode perception trajectories based on the autoregressive diffusion model according to claim 2 is characterized in that: Methods for extracting historical time pattern sequences from historical trajectory data include: Each stop in the historical trajectory data is recoded in chronological order to generate a historical time pattern sequence.

5. The method for generating travel mode perception trajectories based on the autoregressive diffusion model according to claim 2, characterized in that: The method of using historical trajectory data and individual travel patterns as training samples to train a travel pattern conditional guided diffusion model includes: In the forward diffusion process, the historical trajectory data is converted into a continuous latent space vector and noise is gradually added to obtain completely noisy trajectory data; In the reverse diffusion process, the spatial anchor point data and the historical time pattern sequence are used as constraints to denoise the completely noisy trajectory data, restore the trajectory data, and adjust the model parameters of the travel mode condition guided diffusion model based on the restored results to complete a training cycle.

6. The method for generating travel mode perception trajectories based on the autoregressive diffusion model according to claim 5 is characterized in that: The noise added step-by-step to the continuous latent space vectors is Gaussian noise.

7. The method for generating travel mode perception trajectories based on the autoregressive diffusion model according to claim 5, characterized in that: The method further includes: performing a MeanClamp operation on the completely noisy trajectory data during the denoising process, and performing an averaging process on repeated positions in the trajectory.

8. A travel mode perception trajectory generation device based on an autoregressive diffusion model, characterized in that: The travel mode perception trajectory generating device includes: A model training module is used to extract individual travel patterns from historical trajectory data, use the individual travel patterns as training samples to train an autoregressive temporal planning model, and use both historical trajectory data and individual travel patterns as training samples to train a travel pattern conditional guided diffusion model; A trajectory generation module is used to extract individual spatial anchor point data from the residential population distribution data, input the individual spatial anchor point data into a trained autoregressive time series planning model to obtain a future time pattern sequence, and input the individual spatial anchor point data and the future time pattern sequence into a trained travel mode conditional guided diffusion model to obtain future travel trajectories.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a travel mode perception trajectory generation program based on an autoregressive diffusion model. When the travel mode perception trajectory generation program based on an autoregressive diffusion model is executed by a processor, the travel mode perception trajectory generation method based on an autoregressive diffusion model according to any one of claims 1 to 7 is implemented.

10. A computer device, characterized in that: The computer device includes a computer-readable storage medium, a processor, and a travel mode perception trajectory generation program based on an autoregressive diffusion model stored in the computer-readable storage medium. When the travel mode perception trajectory generation program based on the autoregressive diffusion model is executed by the processor, the travel mode perception trajectory generation method based on the autoregressive diffusion model according to any one of claims 1 to 7 is implemented.

Images