Individual travel location prediction method and system

Abstract

This invention proposes a method and system for predicting individual travel locations, relating to the technical field of information processing. The method comprises first acquiring an individual's historical travel trajectory; then representing the individual's historical travel trajectory as knowledge of trip destination distribution under different spatiotemporal conditions; and then using this knowledge as input to a pre-set travel knowledge aggregation and selection network model. The model aggregates and selects the knowledge of trip destination distribution under different spatiotemporal conditions, outputting individual travel location prediction results. This invention can effectively improve the accuracy of travel location predictions and reduce computational costs.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing, and in particular to a method and system for predicting individual travel locations. Background Art

[0002] The task of predicting individual travel locations aims to predict an individual's future travel locations based on their historical travel times and locations. This helps provide efficient travel services for individual travelers and provides traffic managers with information on future traffic trends, supporting the rational allocation of road resources. Therefore, individual travel location prediction is of great significance to location-based services, traffic planning, and traffic management.

[0003] There are three main types of existing travel destination prediction methods: those based on trajectory matching, those based on probabilistic graphical models, and those based on deep learning models. Trajectory matching-based travel destination prediction methods primarily calculate the similarity between an individual's recent travel trajectory and historical travel trajectory, selecting the travel destination with the greatest similarity as the prediction result. Probabilistic graphical model-based travel destination prediction methods primarily rely on Markov methods, which typically select the next travel destination with the highest frequency in historical travel based on the previous travel location. Both of these methods are non-deep learning methods, using a unified representation of individual travel trajectories and unified prediction rules during the prediction process. Therefore, these two methods can be directly applied to travel destination prediction for new individuals and new cities without adjustment. However, the prediction rules of these two methods are relatively simple, making it difficult to fully utilize and mine effective individual travel information, nor can they consider the travel preferences of different individuals, resulting in low accuracy in travel destination prediction. Travel location prediction methods based on deep learning models are mainly based on neural networks. They use a large number of labeled individual travel samples to train the neural network, and then make predictions based on the trained neural network. However, when using this method for new individuals, new regions, or new cities, the neural network usually needs to be retrained, which results in a large amount of training work, is time-consuming and labor-intensive, and has high training costs. In addition, some scholars have proposed a travel location prediction method based on a large language model. This method mainly uses a unified natural language to represent the travel trajectories of different individuals, thereby mapping the travel trajectories of different individuals in different cities into a unified natural language representation space. Based on the large language model, a common individual travel pattern is learned, thereby achieving zero-sample migration of the travel prediction model. This method does not require model adjustment when using new individuals, new regions, or new cities, and can be directly applied. However, the large language model has too many parameters, high training costs, and high hardware requirements. Therefore, the computational cost is high, making it difficult to promote and apply. Summary of the Invention

[0004] In order to solve the problems of low travel location prediction accuracy and high computational cost in the above-mentioned prior art, the present invention proposes an individual travel location prediction method and system, which can effectively improve the travel location prediction accuracy and reduce computational cost.

[0005] In order to achieve the above technical effects, the technical solutions of the present invention are as follows:

[0006] A method for predicting an individual's travel location comprises the following steps:

[0007] S1. Obtain individual historical travel trajectories;

[0008] S2. Representing the individual's historical travel trajectory as travel destination distribution knowledge under different spatiotemporal conditions;

[0009] S3. Use the travel destination distribution knowledge as the input of a preset travel knowledge aggregation and selection network model, aggregate and select the travel destination distribution knowledge under different time and space conditions through the travel knowledge aggregation and selection network model, and output the individual travel location prediction result.

[0010] Preferably, the characterization of the individual's historical travel trajectory as travel destination distribution knowledge under different spatiotemporal conditions includes: setting K kinds of spatiotemporal conditions, the maximum number of individual travel locations is M, and the maximum number of spatiotemporal sub-conditions is N, and organizing the terminal frequency distribution of the individual's historical travel trajectory in (1×M) dimensions under different sub-spatiotemporal conditions to obtain the individual's (K×N×M)-dimensional travel destination distribution knowledge under K kinds of spatiotemporal conditions.

[0011] Preferably, the travel knowledge aggregation and selection network model includes a knowledge aggregation module, a knowledge selection module and a prediction module, the output end of the knowledge aggregation module is connected to the input end of the knowledge selection module, and the output end of the knowledge selection module is connected to the input end of the prediction module.

[0012] Preferably, the aggregating and selecting the trip destination distribution knowledge under different spatiotemporal conditions by the travel knowledge aggregation and selection network model includes:

[0013] S31. Using the knowledge aggregation module to perform weighted aggregation on the travel destination distribution knowledge under different spatiotemporal conditions to obtain comprehensive travel destination distribution knowledge under different spatiotemporal conditions;

[0014] S32. Using the knowledge selection module to perform weighted selection on the comprehensive travel destination distribution knowledge to obtain integrated travel destination distribution knowledge;

[0015] S33. Input the integrated travel destination distribution knowledge into the prediction module, and the prediction module outputs the individual travel location prediction result.

[0016] Preferably, the aggregation process of weighted aggregation of the travel destination distribution knowledge under different spatiotemporal conditions is as follows:

[0017] w i =F i (D i )

[0018]

[0019] Among them, w i represents the knowledge weight vector of the travel destination distribution under different sub-spatial conditions under the i-th spatiotemporal condition, F i Denotes the neural network for the i-th spatiotemporal condition, D i represents the knowledge of the travel destination distribution under the i-th spatiotemporal condition, represents the aggregated comprehensive travel destination distribution knowledge under the i-th spatiotemporal condition, w i,j Represents the knowledge weight vector w of the trip destination distribution i The jth vector in D i,j represents the endpoint distribution under the jth sub-condition of the i-th spatiotemporal condition, N represents the maximum number of spatiotemporal sub-conditions, K represents the number of spatiotemporal conditions, express A collection of .

[0020] Preferably, the specific process of weighted selection of the comprehensive travel destination distribution knowledge is as follows:

[0021]

[0022] Among them, G represents the neural network that calculates the comprehensive travel destination distribution knowledge weight under different time and space conditions, k represents the comprehensive travel destination distribution knowledge weight vector, k i represents the i-th vector in the comprehensive travel destination distribution knowledge weight vector k, Represents the travel destination distribution knowledge after weighted fusion.

[0023] Preferably, the calculation expression of the individual travel location prediction result is as follows:

[0024]

[0025] Among them, P represents the individual travel location prediction result, E represents the fully connected network, t represents the historical travel time, and d represents the historical travel location.

[0026] The present invention also proposes an individual travel location prediction system, comprising:

[0027] Acquisition module, used to obtain individual historical travel trajectories;

[0028] A trip destination distribution knowledge representation module is used to represent the individual's historical travel trajectory as trip destination distribution knowledge under different spatiotemporal conditions;

[0029] The travel knowledge aggregation and selection network model prediction module is used to use the travel destination distribution knowledge as the input of the preset travel knowledge aggregation and selection network model, aggregate and select the travel destination distribution knowledge under different time and space conditions through the travel knowledge aggregation and selection network model, and output the individual travel location prediction result.

[0030] Preferably, the travel knowledge aggregation and selection network model includes a knowledge aggregation module, a knowledge selection module and a prediction module, the output end of the knowledge aggregation module is connected to the input end of the knowledge selection module, and the output end of the knowledge selection module is connected to the input end of the prediction module.

[0031] The present invention also provides a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0032] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform the operations of the individual travel location prediction method.

[0033] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0034] The present invention proposes a method and system for predicting individual travel locations. First, the individual's historical travel trajectory is represented as travel destination distribution knowledge under different spatiotemporal conditions. Then, the travel destination distribution knowledge under different spatiotemporal conditions is aggregated and selected in combination with a travel knowledge aggregation and selection network model, thereby fully considering the spatiotemporal preferences of each individual for each trip and improving the prediction accuracy of the travel location. Then, due to the use of a lightweight travel knowledge aggregation and selection network model to aggregate, select and predict the travel destination distribution knowledge, the travel knowledge aggregation and selection network model has a small number of parameters, a fast running speed, reduced computing costs, and is easier to promote and apply. Further, by predicting individual travel locations through the travel knowledge aggregation and selection network model, it can be directly applied to the travel location prediction of different cities, different regions, and different individuals without adjusting the travel knowledge aggregation and selection network model, thereby saving a lot of computing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flowchart showing a method for predicting individual travel locations proposed in an embodiment of the present invention;

[0036] Figure 2 Another flowchart of a method for predicting an individual's travel location proposed in an embodiment of the present invention is shown;

[0037] Figure 3 An example diagram showing the distribution of travel destinations of individuals under different spatiotemporal conditions proposed in an embodiment of the present invention;

[0038] Figure 4 Another example diagram showing the distribution knowledge of individual travel destinations under different spatiotemporal conditions proposed in an embodiment of the present invention;

[0039] Figure 5 A diagram showing the structure of the travel knowledge aggregation and selection network model proposed in an embodiment of the present invention;

[0040] Figure 6 A structural block diagram of an individual travel location prediction system proposed in an embodiment of the present invention is shown;

[0041] Figure 7 A structural block diagram showing a computer device proposed in an embodiment of the present invention;

[0042] 701. Processor; 702. Memory; 703. Communication interface; 704. Communication bus; 705. Executable instructions. DETAILED DESCRIPTION

[0043] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0044] It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings;

[0045] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0046] Example 1

[0047] like Figure 1 and Figure 2 As shown, this embodiment proposes a method for predicting an individual's travel location, comprising the following steps:

[0048] S1. Obtain individual historical travel trajectories;

[0049] S2. Representing the individual's historical travel trajectory as travel destination distribution knowledge under different spatiotemporal conditions;

[0050] In S2, in order to achieve zero-sample migration of the travel location prediction model across cities, a unified representation of individual travel trajectories is an important prerequisite. To this end, step S2 constructs the distribution knowledge of travel destinations under different spatiotemporal conditions based on the individual's historical travel trajectories. For more details, see Figure 3, S2 describes the representation of the individual historical travel trajectory as travel destination distribution knowledge under different spatiotemporal conditions, including: setting K kinds of spatiotemporal conditions, the maximum number of individual travel locations is M, the maximum number of spatiotemporal sub-conditions is N, and by organizing the terminal frequency distribution of the individual historical travel trajectory in (1×M) dimensions under different sub-spatiotemporal conditions, the individual's (K×N×M)-dimensional travel destination distribution knowledge under K kinds of spatiotemporal conditions is obtained.

[0051] Since the number of historical travel locations of each individual is far lower than the total number of locations in the city, step S2 can effectively reduce the representation dimension of the individual travel trajectory and ensure that different individuals in different cities can use this method to represent their individual travel trajectories. At the same time, the individual travel location prediction method proposed in this embodiment can adapt to the individual's future new travel locations through S2, and supports the representation of travel trajectories of any city, any region, any individual, and any length.

[0052] S3. Use the travel destination distribution knowledge as the input of a preset travel knowledge aggregation and selection network model, aggregate and select the travel destination distribution knowledge under different time and space conditions through the travel knowledge aggregation and selection network model, and output the individual travel location prediction result.

[0053] In S3, the travel knowledge aggregation and selection network model includes a knowledge aggregation module, a knowledge selection module and a prediction module. The output end of the knowledge aggregation module is connected to the input end of the knowledge selection module, and the output end of the knowledge selection module is connected to the input end of the prediction module.

[0054] The aggregating and selecting the travel destination distribution knowledge under different spatiotemporal conditions by the travel knowledge aggregation and selection network model includes:

[0055] S31. Using the knowledge aggregation module to perform weighted aggregation on the travel destination distribution knowledge under different spatiotemporal conditions to obtain comprehensive travel destination distribution knowledge under different spatiotemporal conditions;

[0056] Within each travel time and space condition, the distribution of travel destinations under different sub-time and space conditions is usually similar and can be learned from each other. Therefore, the knowledge aggregation module is based on a neural network and weights the distribution of travel destinations under different sub-time and space conditions within each travel time and space condition respectively, and aggregates to obtain stable comprehensive travel destination distribution knowledge under each time and space condition; the individual's travel destination distribution knowledge under different time and space conditions is: D = {D i,j,z |i∈[1,K],j∈[1,N],z∈[1,M]}, the aggregation process of weighted aggregation of travel destination distribution knowledge under different spatiotemporal conditions is as follows:

[0057] w i =Fi (D i )

[0058]

[0059] Among them, w i represents the knowledge weight vector of the travel destination distribution under different sub-spatial conditions under the i-th spatiotemporal condition, F i Denotes the neural network for the i-th spatiotemporal condition, D i represents the knowledge of the travel destination distribution under the i-th spatiotemporal condition, represents the aggregated comprehensive travel destination distribution knowledge under the i-th spatiotemporal condition, w i,j Represents the knowledge weight vector w of the trip destination distribution i The jth vector in D i,j represents the endpoint distribution under the jth sub-condition of the i-th spatiotemporal condition, N represents the maximum number of spatiotemporal sub-conditions, K represents the number of spatiotemporal conditions, express A collection of .

[0060] S32. Using the knowledge selection module to perform weighted selection on the comprehensive travel destination distribution knowledge to obtain integrated travel destination distribution knowledge;

[0061] Subsequently, based on the comprehensive travel destination distribution knowledge under different spatiotemporal conditions, a neural network is used to perform weighted fusion of the comprehensive travel destination distribution knowledge under each spatiotemporal condition, prompting the neural network to select the appropriate travel destination distribution knowledge for each individual trip to predict the travel location. The specific process of weighted selection of the comprehensive travel destination distribution knowledge is as follows:

[0062]

[0063] Among them, G represents the neural network that calculates the comprehensive travel destination distribution knowledge weight under different time and space conditions, k represents the comprehensive travel destination distribution knowledge weight vector, k i represents the i-th vector in the comprehensive travel destination distribution knowledge weight vector K, Represents the travel destination distribution knowledge after weighted fusion.

[0064] S33. Input the integrated travel destination distribution knowledge into the prediction module, and the prediction module outputs the individual travel location prediction result.

[0065] The prediction module is provided with a fully connected network. Based on the fused travel destination distribution knowledge, the prediction module combines the historical travel time t and the historical travel location d, and predicts the individual travel location prediction result through the fully connected network mapping. The calculation expression of the individual travel location prediction result is as follows:

[0066]

[0067] Among them, P represents the individual travel location prediction result, the location where its maximum value is located is the predicted location number, E represents the fully connected network, t is the historical travel time, and d is the historical travel location.

[0068] It is important to note that this invention proposes and implements a lightweight, efficient, and cross-city, zero-shot migration method for predicting individual travel locations. (Compared to existing trajectory matching, probabilistic graphical models, and large language models, this method fully considers the temporal and spatial travel preferences of different individuals, has a low parameter count, high operational efficiency, and is more easily applicable.)

[0069] The present invention uses the knowledge of the distribution of individual travel destinations under different spatiotemporal conditions to characterize the individual's historical travel trajectory. This method can effectively unify the travel trajectory representations of different cities and different individuals. (Differences from existing travel trajectory representation methods: There are three main existing methods for representing individual historical travel trajectories: The first is to encode the spatiotemporal information of each individual's trip into a vector through a neural network, and use the spatiotemporal information encoding vectors of multiple trips to represent the individual's historical travel trajectory. This method can only encode trained locations. When new locations appear or when migrating to a new city, retraining is required, so it does not support zero-sample migration across cities. The second is to use natural language to describe the individual's historical travel trajectory. This method can be migrated to new locations, new individuals, and new cities, but it usually needs to be combined with a large language model for travel prediction. Therefore, the model parameters are huge, the hardware requirements and application costs are high, and it is difficult to promote and apply. The third is to characterize the individual's travel location and time each time based on one-hot encoding, and use the one-hot encoding vector of the locations and times of multiple trips to represent the individual's historical travel trajectory. The dimension of the one-hot encoding vector is related to the total number of locations in the city, so this method is not suitable for use in cities with a large number of locations.)

[0070] The present invention first uses a neural network to weight the travel destination distribution knowledge within different spatiotemporal conditions for an individual, aggregating it to obtain comprehensive travel destination distribution knowledge under different spatiotemporal conditions. This comprehensive travel destination distribution knowledge under different spatiotemporal conditions is then weighted using a neural network to obtain fused travel destination distribution knowledge. This method adaptively aggregates travel knowledge within different spatiotemporal conditions and adaptively assigns spatiotemporal knowledge weights to each individual's trip, fully accounting for the travel spatiotemporal preferences of different individuals. This process is universal across both the individual and spatial levels, so once trained, it can be transferred to new cities and new individuals without adjustment.

[0071] Existing travel destination prediction methods that can be transferred across cities with zero-shot accuracy primarily include those based on trajectory matching, probabilistic graphical models, and large language models. However, the first two methods are non-deep learning methods and primarily rely on single rules for travel destination prediction. They fail to consider the travel time and space preferences of different individuals and therefore have limited accuracy. Large language model-based travel destination prediction methods primarily use natural language to uniformly represent the travel trajectories of different individuals and utilize large language models to learn the common travel time and space dependencies between individuals. However, large language models have a large number of parameters, long training and prediction times, and high hardware requirements, making them difficult to promote and apply. In this regard, the present invention proposes a method for predicting individual travel locations. First, the individual's historical travel trajectory is represented as travel destination distribution knowledge under different spatiotemporal conditions. Then, the travel destination distribution knowledge under different spatiotemporal conditions is aggregated and selected in combination with the travel knowledge aggregation and selection network model, thereby fully considering the spatiotemporal preferences of each individual for each trip and improving the prediction accuracy of the travel location. Then, due to the use of a lightweight travel knowledge aggregation and selection network model to aggregate, select and predict the travel destination distribution knowledge, the travel knowledge aggregation and selection network model has a small number of parameters, a fast running speed, reduced computing costs, and is easier to promote and apply. Further, by predicting individual travel locations through the travel knowledge aggregation and selection network model, it is possible to directly apply it to the travel location prediction of different cities and different individuals without adjusting the travel knowledge aggregation and selection network model, thereby saving a lot of computing costs.

[0072] Example 2

[0073] This embodiment is based on the Foursquare public dataset and takes the individual travel location prediction of City A and City B as an example to illustrate the implementation steps of an individual travel location prediction method proposed in the present invention. The basic information of the individual travel datasets of City A and City B is shown in Table 1.

[0074] Table 1 Individual travel data sets of City A and City B

[0075]

[0076] 1) First, the individual historical travel trajectory is uniformly represented as the travel destination distribution knowledge under different spatiotemporal conditions.

[0077] According to Table 1, the maximum number of individual travel locations in City A is 88, and the maximum number of individual travel locations in City B is 123. Therefore, the (1×123)-dimensional travel location frequency distribution vector can represent the location frequency distribution of all individuals in the two cities. However, considering the individual's future new travel locations, the location frequency distribution vector dimension is set larger and set to (1×210) dimensions. Considering the three spatiotemporal conditions of individual travel starting point, last travel time, and last travel time-travel starting point, taking an individual's historical travel trajectory as an example: ((2012-06-2010:00:00,6), (2012-06-20 17:30:00,3), (2012-06-21 10:05:00,6), (2012-06-2117:40:00,3), (2012-06-22 10:10:00,6), (2012-06-22 17:40:08,3)), the spatiotemporal distribution knowledge of its travel destination can be expressed as Figure 4 .

[0078] 2) Construct a travel knowledge aggregation and selection network model.

[0079] The travel knowledge aggregation and selection network model includes a knowledge aggregation module, a knowledge selection module, and a prediction module. The output of the knowledge aggregation module is connected to the input of the knowledge selection module, and the output of the knowledge selection module is connected to the input of the prediction module. Considering the three spatiotemporal conditions of individual travel starting point, last travel time, and last travel time-travel starting point, see Figure 5 Here, we use convolutional neural networks and Transformer neural networks to construct the knowledge aggregation module, which denoises and aggregates trip destination distribution knowledge under different spatiotemporal conditions to obtain stable, comprehensive trip destination distribution knowledge. Subsequently, we use a gated neural network to construct the knowledge selection module, which adaptively selects the comprehensive trip destination distribution knowledge under different spatiotemporal conditions to obtain fused trip destination distribution knowledge. Finally, we use a fully connected neural network to construct the prediction module, which maps trip locations and outputs individual trip location prediction results.

[0080] 3) Training of travel knowledge aggregation and selection network models, as well as the cross-city zero-sample migration process of travel location prediction models.

[0081] Here, a travel knowledge aggregation and selection network model was trained based on the individual trip training set for City A. After 20 rounds of training, the best model from this round was selected as the final individual trip location prediction model. This model was then directly applied, without adjustment, to individual trip predictions for City B. The prediction results for the City B test set are shown in Table 2. As can be seen, the proposed method achieves superior zero-shot trip location prediction accuracy compared to classic probabilistic graphical models and large language models. It also significantly outperforms the large language model method in terms of parameter count and runtime speed, demonstrating that the proposed method is a lightweight, efficient, and cross-city zero-shot travel location prediction method.

[0082] Table 2 Zero-sample transfer effects of different methods in City B

[0083]

[0084] Example 3

[0085] See also Figure 6 This embodiment also proposes an individual travel location prediction system, including:

[0086] Acquisition module, used to obtain individual historical travel trajectories;

[0087] A trip destination distribution knowledge representation module is used to represent the individual's historical travel trajectory as trip destination distribution knowledge under different spatiotemporal conditions;

[0088] The travel knowledge aggregation and selection network model prediction module is used to use the travel destination distribution knowledge as the input of the preset travel knowledge aggregation and selection network model, aggregate and select the travel destination distribution knowledge under different time and space conditions through the travel knowledge aggregation and selection network model, and output the individual travel location prediction result.

[0089] The travel knowledge aggregation and selection network model includes a knowledge aggregation module, a knowledge selection module and a prediction module. The output end of the knowledge aggregation module is connected to the input end of the knowledge selection module, and the output end of the knowledge selection module is connected to the input end of the prediction module.

[0090] In this embodiment, the individual's historical travel trajectory is first represented as travel destination distribution knowledge under different spatiotemporal conditions, and then the travel destination distribution knowledge under different spatiotemporal conditions is aggregated and selected in combination with the travel knowledge aggregation and selection network model, thereby fully considering the spatiotemporal preferences of each individual for each trip and improving the prediction accuracy of the travel location; then, due to the use of a lightweight travel knowledge aggregation and selection network model to aggregate, select and predict the travel destination distribution knowledge, the travel knowledge aggregation and selection network model has a small number of parameters, a fast running speed, reduced computing costs, and is easier to promote and apply; further, the individual travel location is predicted by the travel knowledge aggregation and selection network model, which can be directly applied to the travel location prediction of different cities and different individuals without adjusting the travel knowledge aggregation and selection network model, thereby saving a lot of computing costs.

[0091] Example 4

[0092] This embodiment also proposes a computer device, see Figure 7 , comprising: a processor 701, a memory 702, a communication interface 703 and a communication bus 704, wherein the processor 701, the memory 702 and the communication interface 703 communicate with each other via the communication bus 704;

[0093] Wherein: the processor 701, the memory 702 and the communication interface 703 communicate with each other through the communication bus 704. The communication interface 703 is used for network communication with other devices such as a client or other servers. The processor 701 is used to execute executable instructions 705, and specifically can execute the operations of the individual travel location prediction method. Specifically, the executable instructions 705 may include program code. The processor 701 may be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement an embodiment of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or different types of processors, such as one or more CPUs and one or more ASICs.

[0094] The memory 702 is used to store executable instructions 705. The memory 702 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0095] The executable instructions 705 may be specifically invoked by the processor 701 to cause the computer device to perform the following operations:

[0096] S1. Obtain individual historical travel trajectories;

[0097] S2. Representing the individual's historical travel trajectory as travel destination distribution knowledge under different spatiotemporal conditions;

[0098] S3. Use the travel destination distribution knowledge as the input of a preset travel knowledge aggregation and selection network model, aggregate and select the travel destination distribution knowledge under different time and space conditions through the travel knowledge aggregation and selection network model, and output the individual travel location prediction result.

[0099] In this embodiment, the individual's historical travel trajectory is first represented as travel destination distribution knowledge under different spatiotemporal conditions, and then the travel destination distribution knowledge under different spatiotemporal conditions is aggregated and selected in combination with the travel knowledge aggregation and selection network model, thereby fully considering the spatiotemporal preferences of each individual for each trip and improving the prediction accuracy of the travel location; then, due to the use of a lightweight travel knowledge aggregation and selection network model to aggregate, select and predict the travel destination distribution knowledge, the travel knowledge aggregation and selection network model has a small number of parameters, a fast running speed, reduced computing costs, and is easier to promote and apply; further, the individual travel location is predicted by the travel knowledge aggregation and selection network model, which can be directly applied to the travel location prediction of different cities and different individuals without adjusting the travel knowledge aggregation and selection network model, thereby saving a lot of computing costs.

[0100] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. A person skilled in the art would be able to make other variations or modifications based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting individual travel locations, characterized in that: The following steps are involved: S1. Obtain individual historical travel trajectories; S2. Representing the individual's historical travel trajectory as travel destination distribution knowledge under different spatiotemporal conditions; S3. Using the trip destination distribution knowledge as input to a preset travel knowledge aggregation and selection network model, the travel knowledge aggregation and selection network model aggregates and selects the trip destination distribution knowledge under different spatiotemporal conditions, and outputs individual travel location prediction results; The travel knowledge aggregation and selection network model includes a knowledge aggregation module, a knowledge selection module and a prediction module. The output end of the knowledge aggregation module is connected to the input end of the knowledge selection module, and the output end of the knowledge selection module is connected to the input end of the prediction module. The aggregation process of weighted aggregation of trip destination distribution knowledge under different spatiotemporal conditions is as follows: in, Indicates in i The travel destination distribution knowledge weight vector of different sub-spatiotemporal conditions under the spatiotemporal conditions is: Indicates the direction i A spatiotemporal neural network. Indicates in i The knowledge of the distribution of travel destinations under various time and space conditions, Indicates in i The aggregated comprehensive trip destination distribution knowledge under various spatiotemporal conditions, Represents the knowledge weight vector of trip destination distribution The j vectors, Indicates the i The endpoint distribution under the jth sub-condition of the spatiotemporal condition, represents the maximum space-time subcondition number, K represents the space-time condition number, express A collection of .

2. The method for predicting individual travel locations according to claim 1, wherein: The characterization of the individual historical travel trajectory as the travel destination distribution knowledge under different spatiotemporal conditions includes: setting K spatiotemporal conditions, the maximum number of individual travel locations is , the maximum spacetime subcondition number is , by using different sub-spacetime conditions The terminal frequency distribution of the individual historical travel trajectory of the dimensional Under the conditions of time and space ( dimensional knowledge of the distribution of travel destinations.

3. The method for predicting individual travel locations according to claim 1, wherein: The aggregating and selecting the travel destination distribution knowledge under different spatiotemporal conditions by the travel knowledge aggregation and selection network model includes: S31. Using the knowledge aggregation module to perform weighted aggregation on the travel destination distribution knowledge under different spatiotemporal conditions to obtain comprehensive travel destination distribution knowledge under different spatiotemporal conditions; S32. Using the knowledge selection module to perform weighted selection on the comprehensive travel destination distribution knowledge to obtain integrated travel destination distribution knowledge; S33. Input the integrated travel destination distribution knowledge into the prediction module, and the prediction module outputs the individual travel location prediction result.

4. The method for predicting individual travel locations according to claim 1, wherein: The specific process of weighted selection of the comprehensive travel destination distribution knowledge is as follows: in, represents a neural network that calculates the knowledge weights of the comprehensive travel destination distribution under different spatiotemporal conditions, represents the comprehensive travel destination distribution knowledge weight vector, Represents the comprehensive travel destination distribution knowledge weight vector The i vectors, Represents the travel destination distribution knowledge after weighted fusion.

5. The method for predicting individual travel locations according to claim 4, characterized in that: The calculation expression of the individual travel location prediction result is as follows: in, represents the prediction result of individual travel location, represents a fully connected network, t represents the historical travel time, and d represents the historical travel location.

6. An individual travel location prediction system, characterized in that: include: Acquisition module, used to obtain individual historical travel trajectories; A trip destination distribution knowledge representation module is used to represent the individual's historical travel trajectory as trip destination distribution knowledge under different spatiotemporal conditions; a travel knowledge aggregation and selection network model prediction module, configured to use the travel destination distribution knowledge as input to a preset travel knowledge aggregation and selection network model, aggregate and select the travel destination distribution knowledge under different spatiotemporal conditions through the travel knowledge aggregation and selection network model, and output individual travel location prediction results; The travel knowledge aggregation and selection network model includes a knowledge aggregation module, a knowledge selection module and a prediction module. The output end of the knowledge aggregation module is connected to the input end of the knowledge selection module, and the output end of the knowledge selection module is connected to the input end of the prediction module. The aggregation process of weighted aggregation of trip destination distribution knowledge under different spatiotemporal conditions is as follows: in, Indicates in i The travel destination distribution knowledge weight vector of different sub-spatiotemporal conditions under the spatiotemporal conditions is: Indicates the i A spatiotemporal neural network. Indicates in i The knowledge of the distribution of travel destinations under various time and space conditions, Indicates in i The aggregated comprehensive trip destination distribution knowledge under various spatiotemporal conditions, Represents the knowledge weight vector of trip destination distribution The j vectors, Indicates the i The endpoint distribution under the jth sub-condition of the spatiotemporal condition, represents the maximum space-time subcondition number, K represents the space-time condition number, express A collection of .

7. A computer device, characterized in that: include: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform the operation of the individual travel location prediction method according to any one of claims 1 to 5.

Images