A Deep Learning-Based Method and System for Modeling Urban Residents' Travel Distribution

Abstract

This invention discloses a method and system for modeling urban residents' travel distribution based on deep learning. The method includes: obtaining a first OD matrix for urban residents' travel based on resident travel survey data, and obtaining a second OD matrix for urban residents' travel based on mobile phone signaling data; expanding the matrices using census data to obtain expanded first and second travel matrices; performing multi-dimensional verification on the expanded first and second travel matrices to obtain a third OD matrix; extracting the required features for modeling urban residents' travel distribution based on the third OD matrix and influencing factors of urban residents' travel; and modeling the distribution by combining traditional gravity models and deep learning-based travel distribution models using these required features, and finally establishing the model by considering a comprehensive evaluation index system. This invention improves prediction accuracy and provides a more reasonable decision-making basis for modeling urban residents' travel distribution.

Description

Technical Field

[0001] This invention relates to the technical field of urban residents' travel distribution modeling and analysis, and in particular to a method and system for modeling urban residents' travel distribution based on deep learning. Background Technology

[0002] With the rapid development of urbanization, the complexity of urban traffic conditions and the increasing travel demand have gradually become a serious challenge facing society, especially for large cities. Against this backdrop, refined traffic management is becoming increasingly important, and accurate prediction of traffic demand has become an indispensable part of urban traffic planning. Currently, the most widely used traffic demand model is the "four-stage" model, in which trip distribution, as the core step, plays a crucial role in connecting the preceding and following stages. Its output serves as a reference for urban travel demand and is also a core input for subsequent steps in mode selection and traffic allocation, as well as a key element ensuring the overall accuracy of the model. Therefore, accurate trip distribution modeling methods can provide refined elements for traffic planning and provide a basis for decision-making in future urban traffic planning, management, and control.

[0003] Existing research often uses traditional gravity models to model trip distribution. These models can be further categorized into unconstrained, single-constrained, and double-constrained gravity models based on whether they satisfy constraints on the total amount of traffic generated and attracted. Gravity models have advantages such as simple and easy-to-understand form and consideration of road network changes. However, their oversimplification of trip influencing factors and linearity leads to limited predictive performance. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a deep learning-based method and system for modeling urban residents' travel distribution to address the problems of existing travel distribution modeling methods that are too simplistic in considering travel influencing factors and too simplistic in linear models, resulting in limited predictive performance, and the inability to accurately estimate urban residents' travel distribution in years without the latest resident travel survey data.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide a method for modeling urban residents’ travel distribution based on deep learning, comprising: obtaining a first OD matrix of urban residents’ travel based on residents’ travel survey data, obtaining a second OD matrix of urban residents’ travel based on mobile phone signaling data, and expanding the matrix by combining population census data to obtain a first travel matrix expansion and a second travel matrix expansion;

[0009] The first and second travel matrix expansions are validated in multiple dimensions to obtain the third OD matrix.

[0010] Based on the third OD matrix and the influencing factors of urban residents' travel, the required features for modeling the distribution of urban residents' travel are extracted;

[0011] Based on the required characteristics for modeling the travel distribution of urban residents, a model is constructed by combining a traditional gravity model and a deep learning-based travel distribution model, and a comprehensive evaluation index system is considered to achieve the final establishment of the model.

[0012] As a preferred embodiment of the deep learning-based urban resident travel distribution modeling method described in this invention, the step of obtaining the first OD matrix of urban resident travel based on resident travel survey data includes: resident travel survey data preprocessing, origin-destination mapping, and expansion of the first travel matrix based on population census data.

[0013] The process of obtaining the second OD matrix for urban residents' travel based on mobile phone signaling data includes mobile phone signaling data preprocessing, stop point identification, travel chain extraction, resident user identification, and second travel matrix expansion based on population census data expansion.

[0014] As a preferred embodiment of the deep learning-based urban resident travel distribution modeling method described in this invention, the first travel matrix and the second travel matrix are expanded and validated in multiple dimensions, including: spatial distribution similarity verification and average daily travel frequency verification per person.

[0015] The average daily number of trips per person was calculated using the expanded first and second OD matrices and compared with the average daily number of trips per person in the existing travel survey data.

[0016] If the number of differences is less than the first threshold, the verification is successful;

[0017] If the number of differences exceeds the first threshold, the verification fails.

[0018] The spatial distribution of travel volume is calculated using the expanded first and second OD matrices.

[0019] If the similarity of the spatial distribution of travel volume calculated from the expanded first OD matrix and the expanded second OD matrix is ​​greater than the second threshold, then the verification is successful;

[0020] If the similarity of the spatial distribution of travel volume calculated from the expanded first OD matrix and the expanded second OD matrix is ​​less than the second threshold, the verification fails.

[0021] The third OD matrix is ​​obtained based on the successful spatial distribution similarity and the average number of trips per person per day.

[0022] As a preferred embodiment of the deep learning-based urban resident travel distribution modeling method described in this invention, the required features for urban resident travel distribution modeling include: travel generation / attraction volume, urban land use properties, traffic accessibility, socio-economic attributes, and travel impedance.

[0023] As a preferred embodiment of the deep learning-based urban resident travel distribution modeling method described in this invention, the modeling process combines a traditional gravity model and a deep learning-based travel distribution model, including: travel distribution model network design, hyperparameter setting, and model optimization.

[0024] The model network design includes an input layer, intermediate hidden layers, and an output layer;

[0025] The input layer uses the Embedding layer to perform low-dimensional vector space transformation on the spatial numbers of the origin and destination, and represents the spatial relationship between the origin and destination regions. These features are then input into the travel distribution model along with the required feature variables for modeling the travel distribution of the city's residents.

[0026] The intermediate hidden layers are mainly fully connected layers, and non-linearity of feature variables is introduced through activation functions. Overfitting is controlled by Dropout layers. The output layer uses the ReLU activation function to ensure that the output result is non-negative.

[0027] Hyperparameter settings should include at least the learning rate, batch size, number of iterations, and optimizer.

[0028] The learning rate is set using an exponential decay method, decreasing to 30% of the original learning rate after every 5 iterations. The batch size is set based on the performance of the computing device and the model prediction accuracy, and the batch size is set to not exceed the device's graphics card memory and the loss function converges. The number of iterations is determined based on the amount of model data, and the number of iterations is the same as the number of times the loss function converges. The Adam optimizer is used.

[0029] As a preferred embodiment of the deep learning-based urban resident travel distribution modeling method described in this invention, the comprehensive evaluation index system includes: Commuter Index (CPC), Root Mean Square Error (RMSE), Normalized Root Mean Square Error (NRMSE), and Jensen-Shannon Divergence (JSD).

[0030] As a preferred embodiment of the deep learning-based urban resident travel distribution modeling method described in this invention, it further includes:

[0031] The Commuter Index (CPC) is expressed as follows:

[0032]

[0033] Among them, y g (l i ,l j ) is O i D j The correct model predicts the generated traffic, y r (l i ,l j ) is O i D j For the actual observed flow rate, l i Representing regions i, l j Indicates region j;

[0034] The root mean square error (RMSE) is expressed as:

[0035]

[0036] Where N is O i The total number of all origin-destination pairs (OD pairs) that departed;

[0037] The normalized root mean square error (NRMSE) is expressed as:

[0038]

[0039] Among them, y r (l i ,l j ) max It is O i D j The maximum true observed flow rate, y r (l i ,l j ) min It is O i D j The minimum true observed flow rate;

[0040] The Jensen-Shannon divergence JSD is expressed as:

[0041]

[0042]

[0043]

[0044] Where P is the probability distribution of the first sample data, Q is the probability distribution of the second sample data, M represents half of the conditional distribution of P under distribution Q, and KL(P|M) is the relative entropy between P and Q.

[0045] Secondly, this invention provides a deep learning-based urban resident travel distribution modeling system, comprising:

[0046] The matrix acquisition and expansion module is used to acquire the first OD matrix of urban residents' travel based on resident travel survey data, and to acquire the second OD matrix of urban residents' travel based on mobile phone signaling data. The matrix is ​​then expanded by combining population census data to obtain the first travel matrix expansion and the second travel matrix expansion.

[0047] The dimension verification module is used to perform multi-dimensional verification on the expansion of the first travel matrix and the expansion of the second travel matrix to obtain the third OD matrix.

[0048] The feature extraction module is used to extract the features required for modeling the urban residents' travel distribution based on the third OD matrix and the influencing factors of urban residents' travel.

[0049] The model building module is used to build a model based on the required features for modeling the travel distribution of urban residents, combining a traditional gravity model and a deep learning-based travel distribution model, and considering a comprehensive evaluation index system to finally establish the model.

[0050] Thirdly, the present invention provides a computing device, comprising:

[0051] Memory and processor;

[0052] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the deep learning-based urban resident travel distribution modeling method.

[0053] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the deep learning-based urban resident travel distribution modeling method.

[0054] Compared with existing technologies, the beneficial effects of this invention are as follows: By utilizing deep learning algorithms combined with various travel influencing factors, this invention models the travel distribution of urban residents, resulting in a more scientific and accurate travel distribution model. This improves prediction accuracy, enables a more realistic and effective perception of the travel needs of residents in different urban traffic areas, and provides a more reasonable basis for decision-making in urban traffic planning, management, and control. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0056] Figure 1 This is a flowchart of a deep learning-based urban resident travel distribution modeling method and system according to an embodiment of the present invention;

[0057] Figure 2 This is a diagram illustrating the structure of a deep learning-based travel distribution model for a method and system for modeling urban residents' travel distribution, as described in one embodiment of the present invention.

[0058] Figure 3 This is a comparison image of the travel origin-destination (OD) matrix at the street scale of a deep learning-based urban resident travel distribution modeling method and system according to an embodiment of the present invention. Figure 3 (a) is a schematic diagram of the travel origin-destination (OD) matrix based on resident travel survey data at the street scale. Figure 3 (b) A schematic diagram of the travel origin-destination matrix based on mobile signaling data at the street scale;

[0059] Figure 4 This is a schematic diagram of the original OD matrix of a deep learning-based urban resident travel distribution modeling method and system according to an embodiment of the present invention;

[0060] Figure 5 This is a result diagram of a traditional gravity model of a deep learning-based urban resident travel distribution modeling method and system according to an embodiment of the present invention;

[0061] Figure 6 This is a result diagram of a deep learning-based travel distribution model of a method and system for modeling urban residents' travel distribution according to an embodiment of the present invention. Detailed Implementation

[0062] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0063] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0064] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0065] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0066] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0067] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0068] Example 1

[0069] Reference Figures 1-2 As one embodiment of the present invention, this embodiment provides a method for modeling urban resident travel distribution based on deep learning, including:

[0070] S1: Obtain the first OD matrix of urban residents' travel based on the resident travel survey data, and obtain the second OD matrix of urban residents' travel based on the mobile phone signaling data. Combine the population census data to expand the matrix to obtain the first travel matrix expansion and the second travel matrix expansion.

[0071] Specifically, obtaining the first OD matrix of urban residents' travel based on resident travel survey data includes resident travel survey data preprocessing, origin-destination mapping, and expansion of the first travel matrix based on population census data expansion.

[0072] Furthermore, data preprocessing includes removing data that does not meet the travel definition and has missing origin and destination points; origin and destination points are mapped to custom study areas based on latitude and longitude.

[0073] Specifically, obtaining the second OD matrix for urban residents' travel based on mobile signaling data includes mobile signaling data preprocessing, stop point identification, travel chain extraction, resident user identification, and second travel matrix expansion based on population census data expansion.

[0074] Furthermore, data preprocessing includes removing signaling data outside the study area; stop point identification is based on a time-space dual-constraint algorithm to identify valid stop points in the signaling data; travel chain extraction is based on converting mobile signaling data into user travel chains according to valid stop points; and resident user identification is determined by identifying the place of residence and workplace.

[0075] S2: Expand the first and second travel matrices and perform multi-dimensional verification to obtain the third OD matrix;

[0076] Furthermore, the expansion of the first and second travel matrices will be validated in multiple dimensions, including: spatial distribution similarity verification and average daily travel frequency verification per person.

[0077] The average daily number of trips per person was calculated using the expanded first and second OD matrices and compared with the average daily number of trips per person in the existing travel survey data.

[0078] If the number of differences is less than the first threshold, the verification is successful;

[0079] If the number of differences exceeds the first threshold, the verification fails.

[0080] The spatial distribution of travel volume is calculated using the expanded first and second OD matrices.

[0081] If the similarity of the spatial distribution of travel volume calculated from the expanded first OD matrix and the expanded second OD matrix is ​​greater than the second threshold, then the verification is successful;

[0082] If the similarity of the spatial distribution of travel volume calculated from the expanded first OD matrix and the expanded second OD matrix is ​​less than the second threshold, the verification fails.

[0083] The third OD matrix is ​​obtained based on the successful spatial distribution similarity and the average number of trips per person per day.

[0084] It should be noted that, in the embodiments of the present invention, the first threshold for verifying the average daily number of trips per person is 20% of the total number of trips per person per day, and the second threshold for verifying the spatial distribution similarity is 80%.

[0085] S3: Based on the third OD matrix and combined with the influencing factors of urban residents' travel, extract the features required for modeling the distribution of urban residents' travel;

[0086] Furthermore, the features required for modeling urban residents' travel distribution include: travel generation / attraction, urban land use, transportation accessibility, socioeconomic attributes, and travel resistance.

[0087] It should be noted that, based on previous research on the impact on urban residents' travel, five categories of characteristic variables were selected for modeling. Among them, travel generation / attraction is the total amount of travel generation and attraction between each OD pair and the origin-destination traffic area; land use properties include the density of various POIs and the total building area; traffic accessibility includes bus stop density, distance to the nearest subway station, road network density, and distance to the nearest ramp; socioeconomic attributes include population density and nighttime light intensity; and travel impedance data is represented by the Manhattan distance between OD pairs.

[0088] S4: Based on the required characteristics for modeling urban residents' travel distribution, a model is built by combining traditional gravity models and deep learning-based travel distribution models, and a comprehensive evaluation index system is considered to achieve the final establishment of the model;

[0089] Furthermore, modeling is carried out by combining traditional gravity models and deep learning-based travel distribution models, including: travel distribution model network design, hyperparameter setting and model optimization;

[0090] The model network design includes an input layer, intermediate hidden layers, and an output layer;

[0091] The input layer uses the Embedding layer to perform low-dimensional vector space transformation on the spatial numbers of the origin and destination, and represents the spatial relationship between the origin and destination regions, which are then input into the travel distribution model along with the feature variables required for modeling the travel distribution of urban residents.

[0092] The intermediate hidden layers are mainly fully connected layers, and non-linearity of feature variables is introduced through activation functions. Overfitting is controlled by Dropout layers. The output layer uses the ReLU activation function to ensure that the output result is non-negative.

[0093] Hyperparameter settings should include at least the learning rate, batch size, number of iterations, and optimizer.

[0094] The learning rate is set using an exponential decay method, decreasing to 30% of the original learning rate after every 5 iterations. The batch size is set based on the performance of the computing device and the model prediction accuracy, and the batch size is set to not exceed the device's graphics card memory and the loss function converges. The number of iterations is determined based on the amount of model data, and the number of iterations is the same as the number of times the loss function converges. The Adam optimizer is used.

[0095] Furthermore, the comprehensive evaluation index system includes: Common Part of Commuters (CPC), Root Mean Square Error (RMSE), Normalized Root Mean Square Error (NRMSE), and Jansen-Shannon Divergence (JSD).

[0096] Furthermore, it also includes: the Commuter Index (CPC), expressed as:

[0097]

[0098] Among them, y g (l i ,l j ) is O i D j The correct model predicts the generated traffic, y r (l i ,l j ) is O i D j For the actual observed flow rate, l i Representing regions i, l j Indicates region j;

[0099] The root mean square error (RMSE) is expressed as:

[0100]

[0101] Where N is O i The total number of all origin-destination pairs (OD pairs) that departed;

[0102] The normalized root mean square error (NRMSE) is expressed as:

[0103]

[0104] Among them, y r (l i ,l j ) max It is O i D j The maximum true observed flow rate, yr (l i ,l j ) min It is O i D j The minimum true observed flow rate;

[0105] The Jensen-Shannon divergence (JSD) is expressed as:

[0106]

[0107]

[0108]

[0109] Where P is the probability distribution of the first sample data, Q is the probability distribution of the second sample data, M represents half of the conditional distribution of P under distribution Q, and KL(P|M) is the relative entropy between P and Q.

[0110] The above is an illustrative scheme of a deep learning-based urban resident travel distribution modeling method according to this embodiment. It should be noted that the technical solution of this deep learning-based urban resident travel distribution modeling system belongs to the same concept as the technical solution of the deep learning-based urban resident travel distribution modeling method described above. Details not described in detail in this embodiment of the deep learning-based urban resident travel distribution modeling system can be found in the description of the deep learning-based urban resident travel distribution modeling method described above.

[0111] This embodiment presents a deep learning-based urban resident travel distribution modeling system, comprising:

[0112] The matrix acquisition and expansion module is used to obtain the first OD matrix of urban residents' travel based on resident travel survey data, and to obtain the second OD matrix of urban residents' travel based on mobile phone signaling data. The matrix is ​​then expanded by combining population census data to obtain the first travel matrix expansion and the second travel matrix expansion.

[0113] The dimension verification module is used to perform multi-dimensional verification on the expansion of the first travel matrix and the expansion of the second travel matrix to obtain the third OD matrix.

[0114] The feature extraction module is used to extract the features required for modeling the urban residents' travel distribution based on the third OD matrix and the influencing factors of urban residents' travel.

[0115] The model building module is used to build a model based on the characteristics required for modeling the travel distribution of urban residents, combining traditional gravity models and deep learning-based travel distribution models, and considering a comprehensive evaluation index system to achieve the final establishment of the model.

[0116] This embodiment also provides a computing device suitable for deep learning-based urban resident travel distribution modeling methods, including:

[0117] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the deep learning-based urban resident travel distribution modeling method proposed in the above embodiments.

[0118] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the method for modeling the urban resident travel distribution based on deep learning as proposed in the above embodiments.

[0119] The storage medium proposed in this embodiment and the method for modeling urban resident travel distribution based on deep learning proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0120] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0121] Example 2

[0122] Reference Figures 3-6 This is one embodiment of the present invention. Unlike the first embodiment, this embodiment provides a verification test of a deep learning-based method for modeling the travel distribution of urban residents, verifying the beneficial effects of our invention.

[0123] The research data selected for this embodiment of the invention includes resident travel survey data from a certain city in 2019, mobile phone signaling data from 2018, and data from the Seventh National Population Census, along with all characteristic variable data for the corresponding years. It should be noted that, for ease of modeling and accuracy, the study area was divided into 1000m x 1000m grids (7200 grids in total), and the dependent and independent variable characteristics were aggregated based on this grid. Furthermore, in the resident travel OD matrices obtained based on the resident travel survey data and the mobile phone signaling data, verification was performed at the street scale (226 grids in total) for ease of comparison.

[0124] Based on resident travel survey data, a travel OD matrix is ​​obtained after data preprocessing, origin-destination mapping, and travel expansion.

[0125] Based on mobile signaling data, a travel origin-destination (OD) matrix is ​​obtained after data preprocessing, stop point identification, travel chain extraction, resident user identification, and travel expansion.

[0126] The trip origin-destination (OD) matrices were obtained from two independent data sources, and the results were validated based on spatial distribution similarity and multiple trip features. The results show that the Pearson correlation coefficient between the two trip OD matrices at the street scale reaches 0.93. For details on the spatial distribution, please refer to [link to relevant documentation]. Figure 3 .

[0127] Verification has shown that urban resident travel distribution obtained based on mobile signaling data has advantages such as a large sample size, high coverage, and the ability to capture more short-distance trips. Therefore, the travel distribution OD matrix obtained based on mobile signaling data was used for subsequent deep learning-based travel distribution modeling.

[0128] Based on previous research on the impact of urban residents' travel, five categories of feature variables were selected for modeling, as shown in Table 1:

[0129] Table 1. Description of Independent Variables

[0130]

[0131]

[0132] The travel distribution model is based on deep learning. For details on the deep neural network architecture design, please refer to [link / reference needed]. Figure 2 The different network layers mainly utilize the Embedding layer to model the spatial relationship between origin and destination traffic zones, the fully connected layer to mine the complex relationship between massive OD traffic flow and various feature variables, the activation function to introduce nonlinear relationships between feature variables, and the Dropout layer to prevent overfitting. Specific hyperparameter settings are shown in Table 2.

[0133] Table 2 Hyperparameter settings for deep learning models

[0134] hyperparameters Detailed settings Learning rate The initial setting is 5e-05, and it decreases to 30% of its original value every 5 epochs. Batch size 128 Number of iterations 50 times Optimizer Adam

[0135] The traditional gravity model used in this embodiment of the invention is specifically a single-constraint gravity model based on the total traffic generation. A comparison of the model outputs of the traditional gravity model and the deep learning-based travel distribution model is shown in the figure below. Figures 4-6 As shown in Table 3, the comparison of CPC, RMSE, NRMSE, and JSD results for the comprehensive evaluation index system is as follows:

[0136] Table 3 Comparison of metrics between traditional gravity models and deep learning-based travel distribution models

[0137]

[0138]

[0139] By combining Table 3 Figure 4 , Figure 5 and Figure 6 It can be seen that the deep learning-based travel distribution model not only better matches the original travel origin-destination (OD) matrix in terms of spatial distribution, but also comprehensively outperforms the traditional gravity model in all aspects. Specifically, CPC accuracy is improved by 77.6%, RMSE error is reduced by 46.6%, and JSD dissimilarity is reduced by 42.4%.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A deep learning-based urban resident trip distribution modeling method, characterized in that, The application relates to a method for obtaining a third OD matrix of urban residents. The method comprises the following steps: obtaining a first OD matrix of urban residents according to resident travel survey data and a second OD matrix of urban residents according to mobile phone signaling data, expanding the matrices according to population census data to obtain a first travel matrix expansion and a second travel matrix expansion; obtaining the first OD matrix according to the resident travel survey data comprises the following steps: resident travel survey data preprocessing, origin-destination mapping and first travel matrix expansion based on population census data expansion; obtaining the second OD matrix according to the mobile phone signaling data comprises the following steps: mobile phone signaling data preprocessing, stop point identification, travel chain extraction, permanent user identification and second travel matrix expansion based on population census data expansion; performing multi-dimensional verification on the first travel matrix expansion and the second travel matrix expansion to obtain a third OD matrix, wherein the multi-dimensional verification comprises spatial distribution similarity verification and per capita daily travel frequency verification, calculating per capita daily travel frequency by using the first OD matrix expansion and the second OD matrix expansion, and comparing the per capita daily travel frequency with per capita daily travel frequency of existing travel survey data; if the difference frequency is less than a first threshold value, the verification is successful; if the difference frequency is greater than the first threshold value, the verification fails; calculating travel volume spatial distribution by using the first OD matrix expansion and the second OD matrix expansion; if the spatial distribution similarity calculated by using the first OD matrix expansion and the second OD matrix expansion is greater than a second threshold value, the verification is successful; if the spatial distribution similarity calculated by using the first OD matrix expansion and the second OD matrix expansion is less than the second threshold value, the verification fails; obtaining the third OD matrix based on the spatial distribution similarity and the per capita daily travel frequency of the successful verification; combining the third OD matrix with influencing factors of urban resident travel to extract required features of urban resident travel distribution modeling; modeling based on a deep learning-based travel distribution model according to the required features of the urban resident travel distribution modeling, and finally establishing the model by considering a comprehensive evaluation index system; the modeling based on the deep learning-based travel distribution model comprises network design of the travel distribution model, super parameter setting and model optimization; the model network design comprises an input layer, a middle hidden layer and an output layer; the input layer uses an Embedding layer to convert spatial numbers of origins and destinations into low-dimensional vector spaces and to represent spatial relationships of origin and destination regions, and inputs the travel distribution model with the required feature variables of the urban resident travel distribution modeling; the middle hidden layer mainly comprises a full connection layer, introduces a feature variable nonlinearity through an activation function and controls overfitting through a Dropout layer; the output layer adopts a ReLU activation function to control the output result to be a non-negative number; the super parameter setting at least comprises a learning rate, a batch size, an iteration number and an optimizer. The learning rate setting adopts an exponential decay method, and decreases to 30% of the original learning rate after every 5 iteration rounds; the batch size is set according to the performance of the computing device and the prediction accuracy of the model, and the batch size is set not to exceed the device GPU memory and the loss function converges; the number of iterations is determined according to the data volume of the model, and the number of iterations is the same as the number of loss function convergences, and the optimizer adopts the Adam optimizer. 2.The deep learning based urban resident trip distribution modeling method of claim 1, wherein, The required features for modeling the trip distribution of urban residents include: trip generation / attraction volume, urban land use properties, traffic accessibility, social and economic attributes, and trip impedance. 3.The deep learning based urban resident trip distribution modeling method of claim 2, wherein, The comprehensive evaluation index system includes: commuter index CPC, root mean square error RMSE, normalized root mean square error NRMSE, and Jensen-Shannon divergence JSD. 4.The deep learning based urban resident trip distribution modeling method of claim 3, wherein, Further comprising: The commuter index CPC is expressed as: where y g (l i ,l j ) is the model predicted flow for the O i D j pair, y r (l i ,l j ) is the true observed flow for the O i D j pair, l i denotes region i, and l j denotes region j; The root mean square error RMSE is expressed as: wherein N is O i Number of all OD pairs originated The normalized root mean square error NRMSE is expressed as: where y r (l i ,l j ) max is the maximum true observed flow for the pair O i D j y r (l i ,l j ) min is the minimum large true observed flow for the pair O i D j ; The Jensen-Shannon divergence JSD is expressed as: Where P is the probability distribution of the first sample data, Q is the probability distribution of the second sample data, M represents the conditional distribution of P under distribution Q, and KL(P|M) is the relative entropy between P and Q.

5. A deep learning based urban resident trip distribution modeling system, applied to the method of any one of claims 1-4, characterized in that, Comprising: A matrix acquisition expansion module is configured to acquire a first OD matrix of urban resident trips according to resident trip survey data, and acquire a second OD matrix of urban resident trips according to mobile phone signaling data, and expand the matrices in combination with population census data to obtain first trip matrix expansion and second trip matrix expansion; A dimension verification module is configured to perform multi-dimensional verification on the first trip matrix expansion and the second trip matrix expansion to obtain a third OD matrix; A feature extraction module is configured to extract required features for modeling the trip distribution of urban residents according to the third OD matrix in combination with influencing factors of the trip distribution of urban residents; A model building module is configured to model in combination with a traditional gravity model and a trip distribution model based on deep learning according to the required features for modeling the trip distribution of urban residents, and to finally build the model by considering a comprehensive evaluation index system. 6.An electronic device comprising: a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, which, when executed by the processor, implement the steps of the method for modeling the trip distribution of urban residents based on deep learning according to any one of claims 1 to 4. 7.A computer readable storage medium storing computer executable instructions, which, when executed by a processor, implement the steps of the method for modeling the trip distribution of urban residents based on deep learning according to any one of claims 1 to 4.

Images