A method for identifying urban functional areas based on multi-source traffic travel data and topic model

Abstract

The present invention discloses a method for identifying urban functional areas based on multi-source traffic travel data and topic models, the method comprising: collecting road network data, point of interest data and multi-source traffic travel data; obtaining a number of traffic analysis communities through morphological processing of the road network data; mining the features in the point of interest data and online car-hailing and subway travel data to construct a "vocabulary", using the latent Dirichlet allocation topic model algorithm to obtain the functional probability distribution of each traffic analysis community; using a clustering algorithm to perform functional clustering on all traffic analysis communities, and completing the functional calibration of the clusters through the constructed regional travel-activity analysis indicators. The beneficial effect of the present invention is that it combines the point of interest data representing the static semantic information of the urban functional space and a variety of resident travel data with dynamic travel characteristics, which not only overcomes the influence of a single factor, but also combines the influence of the activity characteristics of different travel modes, and obtains a more extensive and practical division result.

Description

Technical Field

[0001] The invention relates to a method for identifying urban functional areas based on multi-source traffic travel data and a topic model, and belongs to the technical field of traffic big data. Background Art

[0002] The development of cities has promoted the functionalization of different areas in cities to meet certain specific needs of people. The existence of functional areas enables the rational and efficient deployment of social and economic resources and brings a lot of convenience to people's production and life. Urban functional areas refer to areas where specific urban functions can be maximized through the spatial aggregation of natural resources and social services. With the rapid development of urban construction and social economy, the natural landscape characteristics and socio-economic characteristics presented by urban functional areas are becoming increasingly prominent. The accelerated urbanization process has led to the rapid expansion of urban scale. Past urban planning has gradually failed to meet the needs of urbanization expansion, resulting in many problems such as traffic congestion, housing shortages, and environmental deterioration. The classification of urban functional areas is a prerequisite for scientific and reasonable planning and construction. The study of the identification and division of urban functional areas is of great significance for measuring land use, planning various economic and social functions, rationally and efficiently utilizing various resources, and enhancing regional comprehensive competitiveness. The traditional division of urban functional areas is mainly based on land use survey data, and the urban landscape and socio-economic characteristics are quantified by establishing evaluation indicators to achieve the division of urban functions. This method has high data acquisition cost and low efficiency, and the subjective factors in constructing evaluation indicators are strong. It is difficult to accurately describe the natural and social characteristics of different urban functions. There is an urgent need for a new method and approach to divide urban functional areas. With the booming development of big data technology, the urban functional area analysis method based on big data has become one of the hot means of functional area division technology due to its advantages such as wide sources of big data, high spatiotemporal resolution, and rich information. So far, many studies have combined user activity travel data with point of interest data to identify different urban functional areas, such as using point of interest data with bus card swiping data, subway card swiping data, online car-hailing trajectory data, and mobile phone movement data to identify urban functional areas.

[0003] 1) Point of interest data is a kind of data with category attributes, which has the characteristics of large sample size and wide coverage of information. Its category entries have keywords such as accommodation service facilities, shopping services, science and technology culture, scenic spots, and transportation facilities, which can provide semantic analysis support for clustered regional plots.

[0004] 2) Resident travel data includes bus card swiping data, subway card swiping data, online car-hailing order data, and mobile phone data, etc., which contains rich information about residents' travel. A resident travel record generally includes the residents' boarding and alighting time, boarding and alighting location, and travel duration. There is a close relationship between residents' travel and urban spatial structure. Generally speaking, residents usually leave residential areas in the morning on weekdays and return to their original residence after get off work in the evening. The main time to visit entertainment areas is in the evening on weekdays or all day on non-working days; secondly, the different functions of the region are also related to the arrival pattern of the crowd. For example, residents visiting entertainment areas are likely to come from work areas on weekdays and from residential areas on non-working days; in addition, various studies have shown that travel duration and travel mode selection will affect residents' travel activities. Therefore, the travel characteristics of residents contained in resident travel data are closely related to the functions of the region.

[0005] The method of using multi-source travel data combined with point of interest data to identify functional areas can make up for the problem of recognition bias caused by a single data source, and make full use of the more comprehensive and rich resident travel information contained in multiple travel mode data and the spatial semantic information contained in point of interest data, making the identification of functional areas more accurate. Summary of the invention

[0006] Technical problem: The purpose of this application is to provide a method and technology for identifying urban functional areas based on multi-source traffic travel data and topic models, divide traffic analysis areas based on the road network, and combine points of interest and residents' travel characteristics to overcome the influence of single factors, make full use of the functional area characteristics expressed by data from different travel modes, and obtain more practical results.

[0007] Technical solution: To achieve the above purpose, the technical solution adopted by this application is:

[0008] The present application provides a method for identifying urban functional areas based on multi-source traffic travel data and a topic model. The method for identifying urban functional areas based on multi-source traffic travel data and a topic model is used to divide urban space into functional areas, and includes the following steps:

[0009] Step 1: Obtain road network data, point of interest data, and multi-source transportation (online car-hailing, subway) travel data;

[0010] Step 2: Based on the road network data, the urban space is divided into a number of traffic analysis areas using the morphological processing corrosion and expansion method;

[0011] Step 3: Based on the distribution of the POI data and online car-hailing and subway travel data in the traffic analysis community, mining regional travel-activity features to construct vocabulary;

[0012] Step 4: using a latent Dirichlet allocation topic model algorithm to obtain the functional probability distribution of each of the traffic analysis communities;

[0013] Step 5: Based on the functional probability distribution of each traffic analysis cell, the K-means algorithm is used to perform functional clustering on all traffic analysis cells to obtain C clusters for the urban space, and each cluster is functionally labeled according to the regional travel-activity characteristic distribution in each traffic analysis cell.

[0014] Preferably, the POI data is geospatial point data that includes name, address, coordinates and is divided into P categories after data cleaning and reclassification. The multi-source traffic travel data is the travel data of urban online car-hailing and subway obtained after data cleaning and screening. The road network data is screened road network data including highways, urban expressways and urban trunk roads. These roads are morphologically expanded and eroded, and the urban space is naturally divided into different basic units according to the extension of the processed roads to obtain traffic analysis cells.

[0015] Preferably, regional travel-activity characteristics and land use attribute characteristics are constructed based on the POI data and online car-hailing and subway travel data, including:

[0016] Extract geographic information features of points of interest. The frequency of interest categories and the land use mixture weighted by the average area of ​​each category of interest points are used as the vocabulary of the topic model. The category frequency and regional mixture of interest points weighted in all regions are used as the vocabulary library of the topic model to obtain the probability distribution of traffic analysis cells-travel characteristics composed of interest point categories and the probability distribution of interest point themes in each traffic analysis cell. The calculation method of land use mixture adopts the calculation method based on entropy index:

[0017]

[0018] Among them, p i It represents the proportion of the i-th type of interest point to all types of interest points, and P represents the total number of interest point categories.

[0019] Extract travel-activity features. The travel-activity features are the basic flow characteristics of weekdays and weekends. Each travel data contains two travel features: boarding (entering the station) and getting off (exiting the station). The specific form is feature = (direction, T j ), where direction indicates the direction of subway passengers entering or exiting the station or getting on or off the online car-hailing service. For the online car-hailing service boarding point and the subway passenger entry point, direction = S. For the online car-hailing service boarding point and the subway passenger exit point, direction = E. For online car-hailing service orders, T jrepresents the time when passengers get on and off the bus in the online car-hailing order. For subway data, T j Indicates the time when subway passengers enter and exit the station.

[0020] Based on different data sets, the corresponding vocabulary for each functional area is constructed, and the traffic analysis area-functional feature distribution and functional feature-travel feature distribution constructed by different data sets are obtained by solving the topic model respectively.

[0021] Preferably, the latent Dirichlet allocation probability topic model is constructed, and the functional feature vectors of the traffic analysis community are mined using the latent Dirichlet allocation probability topic model. The latent Dirichlet allocation probability topic model allocates the probability of each functional feature vector by extracting the high-dimensional semantic vector contained in each functional feature vector, including:

[0022] The latent Dirichlet allocation probability topic model is used to give the functional distribution of each traffic analysis community in the data set in the form of a probability distribution vector. The latent Dirichlet allocation is an unsupervised Bayesian model that is widely used in topic modeling and document topic classification:

[0023] p(w|d)=p(w|k)×p(k|d)

[0024] in

[0025] d is a document, which is represented as a traffic analysis cell in a specific application scenario;

[0026] w is a vocabulary, which represents the regional travel-activity characteristics of the traffic analysis cell;

[0027] k is the theme, which represents the urban functional characteristics of the traffic analysis area;

[0028] The model uses k as the middle layer, and p(k|d) represents the probability that each traffic analysis cell d in the total set of traffic analysis cells D corresponds to different urban functions k. Its distribution is represented by the vector θ d , p(w|k) represents the probability of each urban function k in the total urban function set generating different regional travel-activity characteristics w, and its distribution is represented by the vector β k ; The topic model algorithm is used to solve the probability distribution of urban function characteristics of each traffic analysis area, as well as the regional travel-activity characteristic distribution in the urban function characteristic distribution; The learning process of the urban function area characteristic recognition model of the latent Dirichlet allocation probability topic model is as follows:

[0029] The regional travel-activity characteristics in the urban traffic analysis area are generated by the following steps:

[0030] 1) The Dirichlet distribution determined by the hyperparameter α generates the probability distribution θ of the multinomial function characteristics of the traffic analysis cell dd .,Right now in, is the hyperparameter of the distribution, which is a K-dimensional vector, where K represents the total number of urban functions in a given traffic analysis area. The probability density function of the Dirichlet distribution is as follows:

[0031]

[0032] where p k represents the probability of occurrence of the kth urban function, α k is the hyperparameter corresponding to the k-th city function.

[0033] 2) The Dirichlet distribution determined by the hyperparameter δ generates the corresponding traffic analysis community-travel characteristic conditional distribution β k ,Right now Among them, η is the hyperparameter of the travel feature distribution, and the dimension is the number of elements V in the travel feature set.

[0034] 3) The functional characteristics of the nth travel characteristics of the traffic analysis cell d are calculated by calculating the functional characteristic distribution θ of the traffic analysis cell d d Multinomial probability distribution z for the nth travel feature d Get, that is, z d =multi(θ d ), the probability density function of the multinomial distribution is:

[0035]

[0036] Among them, p i represents the probability of occurrence of the i-th functional feature, θ d,i is the occurrence frequency of the i-th functional feature of the traffic analysis cell d.

[0037] Determine the number of topics K based on perplexity. When using the latent Dirichlet allocation topic model for modeling, the perplexity of the trained model in the test set is used to determine the optimal number of topics. If the perplexity of the model on the test set is less than the preset value, the current value of the number of functional features K meets the requirements. The perplexity calculation formula is as follows:

[0038]

[0039] Where D represents the total set of all travel characteristics; M is the total number of traffic analysis cells; W d is the travel feature set in the traffic analysis area d; N d is the number of travel characteristics in the traffic analysis area d; P(W d ) is the likelihood value of all travel characteristics in the traffic analysis area d.

[0040] The regional interest point feature data set, the pre-processed online car-hailing, subway weekday and weekend data set are respectively modeled by topic modeling to obtain the corresponding probability distribution of traffic analysis community function features, and form each traffic analysis community function feature vector. The urban regional function feature vector constructed in this way takes into account the traffic analysis community interest point reclassification data, regional mixing degree, online car-hailing weekday data, online car-hailing weekend data, subway weekday data and subway weekend data. It not only takes into account the regional geographic information, but also combines the residents' travel characteristics reflected by different travel modes, enriching the regional residents' travel characteristics representation, thereby obtaining a more extensive and good result.

[0041] Preferably, according to the functional characteristics of each traffic analysis cell, the K-means algorithm is used to perform functional clustering on all traffic analysis cells. C clusters for the urban space are obtained by clustering, and each cluster is functionally labeled according to the functional probability distribution in each cluster. Including:

[0042] Let the functional feature distribution of the traffic analysis area d be the stack of the probability distribution vectors of the traffic analysis area-functional feature obtained by combining the subway, online car-hailing, and points of interest, that is, feature d =(VeC taxiweek，d , Vec taxiweekend，d , Vec metroweek，d , Vec metroweekend，d , Vec poi，d ), where Vec taxiweek,d In the weekday online car-hailing topic model, the probability distribution vector of the traffic analysis community d belonging to each functional feature of the weekday online car-hailing, Vec taxiweekend,d In the weekend ride-hailing topic model, the probability distribution vector of the traffic analysis community d belonging to the various functional features of the weekend ride-hailing service, Vec metroweek,d In the weekday subway model, the probability distribution vector of the traffic analysis area d belonging to each functional feature of the weekday subway, Vec metroweekend,d In the rest day subway model, the probability distribution vector of the traffic analysis area d belonging to each functional feature of the rest day subway, Vec poi,d It represents the probability distribution vector of each functional feature of the point of interest in the point of interest model.

[0043] The number of functional area clusters in the final traffic analysis community is determined by the silhouette coefficient of the cluster evaluation index:

[0044]

[0045] Among them, the vector composed of the functional distribution characteristics of each cluster is taken as a sample, a represents the average distance between the sample and all other elements in the same cluster; traverse all clusters that do not contain the sample point, calculate the average distance from the sample point to all points in these clusters, and the minimum value is b. ;

[0046] Calculate the corresponding silhouette coefficient index under different clustering numbers C, and take the C value that makes the silhouette coefficient index value the largest, which is the C clusters for the urban space traffic analysis area.

[0047] For each cluster obtained by clustering, the specific urban functional land category of the area is determined according to its regional characteristics and travel activity characteristic indicators. Among them, the regional characteristics include the frequency density and category ratio of the points of interest in each traffic analysis area. The frequency density and category ratio in the region are calculated by the following formulas:

[0048]

[0049]

[0050] Among them, i represents the type of interest point, P represents the total number of interest point categories, and n i Represents the number of interest points of the i-th category in a certain area, N i represents the total number of interest points of type i in the study area; FD i represents the frequency density of the i-th type of interest point to the total number of interest points of this type; CR i It represents the ratio of the frequency density of the i-th type of interest point to the frequency density of all types of interest points in a certain area;

[0051] The regional travel activity characteristic index includes the normalized hourly flow of multi-source travel data on weekdays and weekends in the region, which is calculated by the following formula:

[0052]

[0053] Among them, X norm represents the normalized hourly flow rate of the region, X represents the original hourly flow rate of the region, and X max and X min Respectively represent the maximum and minimum hourly flow rates in the area.

[0054] By analyzing the travel characteristics of residents reflected in the normalized hourly flow on weekdays and weekends of regional multi-source travel data and comparing them with the travel characteristics of residents in typical urban functional land, the specific urban function calibration of the region is completed.

[0055] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0056] The urban functional area identification method based on multi-source traffic travel data and topic model of the present invention uses easily accessible and large-scale multi-source traffic travel data as the data source, which makes up for the defects of low sampling rate and high cost of the questionnaire survey method adopted in the traditional urban functional area survey; it expands the semantic information of travel data by combining with point of interest data, so that dynamic travel data can play a greater role in urban functional area identification; the model of the method of the present invention is simple, avoiding a large amount of parameter calibration, and the identification result is accurate and reliable, which is an important application of traffic travel data in urban land use. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a technical flow chart of the urban functional area identification method based on multi-source traffic travel data and topic model of this application;

[0058] Figure 2 A schematic diagram of an embodiment of using road network data to divide traffic analysis into districts;

[0059] Figure 3 Schematic diagram of the latent Dirichlet allocation topic model;

[0060] Figure 4 It is the silhouette coefficient graph corresponding to different numbers of clusters of the K-means method;

[0061] Figure 5 A normalized flow line chart of online ride-hailing services in a functional area on weekdays obtained using the technical method proposed in this patent;

[0062] Figure 6 A schematic diagram of an embodiment of functional area identification obtained using the technical method proposed in this patent. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0065] In one embodiment, a method for identifying urban functional areas based on multi-source traffic travel data and a topic model is provided. The method for identifying urban functional areas based on multi-source traffic travel data and a topic model is used to identify functional areas of urban space.

[0066] like Figure 1 As shown in FIG. 1 , the urban functional area identification method based on multi-source traffic travel data and topic model includes the following steps:

[0067] Step 1: Obtain road network data, point of interest data, and multi-source transportation (online car-hailing, subway) travel data;

[0068] When obtaining road network data, point of interest data, and multi-source transportation (online car-hailing, subway) travel data, you can directly obtain the above data by establishing a data interface with relevant departments or units; you can also collect and integrate the above data through a third-party platform.

[0069] After obtaining the above data, the above raw data can also be selectively subjected to data cleaning processing such as deduplication, screening or error correction to improve the reliability of the basic data. The obtained geospatial point of interest data including name, address and coordinates can be reclassified into P categories for subsequent data processing.

[0070] Step 2: According to the road network data, the urban space is divided into a number of traffic analysis areas through morphological erosion and expansion processing;

[0071] Urban roads are generally divided into expressways, trunk roads, secondary trunk roads, and branch roads. In order to obtain basic units of reasonable size, the road network data used in this embodiment includes: expressways, urban expressways, and urban trunk roads. These roads are subjected to morphological expansion and corrosion processing, and the urban space is naturally divided into different basic units according to the extension of the processed roads to obtain traffic analysis cells.

[0072] like Figure 2 As shown in the figure, the black lines in the area represent the distribution of expressways, urban expressways and urban trunk roads in the urban area. After the roads are morphologically expanded and eroded, the closed areas surrounded by road intersections and extensions are used as different basic units.

[0073] It should be noted that the dilation and erosion operations used in the morphological processing of this embodiment are prior arts, and their principles will not be described in detail.

[0074] Step 3: construct regional travel-activity characteristics and land use attribute characteristics based on the POI data and the resident travel data of online ride-hailing and subway;

[0075] Extract geographic information features of points of interest. The frequency of interest categories and the land use mixture weighted by the average area of ​​each category of interest points are used as the vocabulary of the topic model. The category frequency and regional mixture of interest points weighted in all regions are used as the vocabulary of the topic model to obtain the probability distribution of traffic analysis cells-travel characteristics composed of interest point categories and the probability distribution of interest point themes in each traffic analysis cell. The calculation method of land use mixture adopts the calculation method based on entropy index, as shown below:

[0076]

[0077] Among them, p i It represents the proportion of the i-th type of interest point to all types of interest points, and P represents the number of interest point categories.

[0078] Extract travel-activity features. The travel-activity features are the basic flow characteristics of weekdays and weekends. Each travel data contains two travel features: boarding (entering the station) and getting off (exiting the station). The specific form is: feature = (direction, T j ), where direction indicates the direction of subway passengers entering or exiting the station or getting on and off the train. For the online car-hailing order boarding point and the subway passenger entry station, direction = S is taken. For the online car-hailing order drop-off point and the subway passenger exit station, direction = E is taken. For online car-hailing orders, T j represents the time when passengers get on and off the bus in the online car-hailing order. For subway data, T j Indicates the time when subway passengers enter and exit the station.

[0079] A traffic analysis cell is regarded as a document, and a function in the traffic analysis cell is regarded as a topic. Therefore, an area with multiple functions is similar to a document containing various topics. The corresponding words of each functional area constructed by different data sets are respectively substituted into the topic model, and the document-function probability distribution and traffic analysis cell-travel feature distribution of the topic model constructed by different data sets are respectively solved by the topic model. The mapping relationship from traffic analysis cell function to document topic is shown in Table 1.

[0080] Table 1 Analogy from regional functions to document topics

[0081]

[0082] Step 4: using a latent Dirichlet allocation topic model algorithm to obtain the functional characteristics of each of the traffic analysis communities;

[0083] Based on the corresponding vocabulary of each functional area constructed by different data sets, the traffic analysis community-functional feature distribution and functional feature-travel feature distribution constructed by different data sets are solved respectively, and the latent Dirichlet allocation topic model is used to solve them. The functional feature vector of the traffic analysis community is mined using the latent Dirichlet allocation probability topic model. The latent Dirichlet allocation probability topic model allocates the probability of each functional feature vector by extracting the high-dimensional semantic vector contained in each functional feature vector, including:

[0084] The latent Dirichlet allocation probability topic model is used to give the functional feature distribution of each traffic analysis community in the data set in the form of probability distribution. The latent Dirichlet allocation is an unsupervised Bayesian model that is widely used in topic modeling and document topic classification, where:

[0085] p(w|d)=p(w|k)×p(k|d)

[0086] in

[0087] d is a document, which is represented as a traffic analysis cell in a specific application scenario;

[0088] w is a vocabulary, which represents the regional travel-activity characteristics of the traffic analysis cell;

[0089] k is the theme, which represents the urban functional characteristics of the traffic analysis area;

[0090] The model uses k as the middle layer, and p(k|d) represents the probability that each traffic analysis cell d in the total set of traffic analysis cells D corresponds to different urban functions k. Its distribution is represented by the vector θ d , p(w|k) represents the probability of each urban function k in the total urban function set generating different regional travel-activity characteristics w, and its distribution is represented by the vector β k ; The topic model algorithm is used to solve the probability distribution of urban function characteristics of each traffic analysis area, as well as the regional travel-activity characteristic distribution in the urban function characteristic distribution.

[0091] like Figure 3 As shown in Figure 2, the generation process of identifying functional areas based on the latent Dirichlet allocation probability topic model is as follows:

[0092] 1) The hyperparameter α determines the Dirichlet distribution to generate the functional characteristic probability distribution θ of the traffic analysis cell d d ,Right now in, is the hyperparameter of the distribution, which is a K-dimensional vector. For each functional feature k, the distribution hyperparameter α is taken as α=50 / k based on relevant experience. K represents the total number of urban functions in a given traffic analysis area. The probability density function of the Dirichlet distribution is as follows:

[0093]

[0094] Among them, p k represents the probability of occurrence of the kth urban function, α k is the hyperparameter corresponding to the k-th city function;

[0095] 2) The hyperparameter η determines the conditional distribution β of the travel characteristics generated by the Dirichlet distribution with functional characteristics k k ,Right now Among them, η is the hyperparameter of the travel feature distribution, and the dimension is the number of elements V in the travel feature set;

[0096] 3) The functional characteristics of the nth travel characteristics of the traffic analysis cell d are calculated by calculating the functional characteristic distribution θ of the traffic analysis cell d d Multinomial probability distribution z for the nth travel feature d Get, that is, z d =multi(θ d ), the probability density function of the multinomial distribution is:

[0097]

[0098] Among them, p i represents the probability of occurrence of the i-th functional feature, θ d，i The occurrence frequency of the ith functional feature of the traffic analysis cell d;

[0099] The function feature-travel feature conditional probability distribution corresponding to each function feature k is β k , according to its conjugate polynomial distribution φ k Randomly select the travel characteristics, that is, φ k =multi(β k ), the probability density function of the multinomial distribution is:

[0100]

[0101] Among them, p i represents the probability of occurrence of the i-th travel feature, β k，i is the occurrence frequency of the i-th travel feature;

[0102] When using the latent Dirichlet allocation topic model for modeling, the perplexity of the trained model in the test set is used to determine the optimal number of topics. If the perplexity of the model on the test set is less than the preset value, the current value of the number of topics K meets the requirements. The perplexity calculation formula is as follows:

[0103]

[0104] Where D represents the total set of all travel characteristics; M is the total number of traffic analysis cells; W d is the travel feature set in the traffic analysis area d; N d is the number of travel characteristics in the traffic analysis area d; P(W d ) is the likelihood value of all travel characteristics in the traffic analysis area d.

[0105] The regional interest point feature data set, the pre-processed online car-hailing, subway weekday and weekend data set are respectively used for topic modeling to obtain the corresponding probability distribution of traffic analysis community function features, and form each traffic analysis community function feature vector. The urban area function feature vector constructed in this way takes into account the regional interest point reclassification data, regional mixing degree, online car-hailing, subway weekday and weekend data, and not only takes into account the regional geographic information, but also combines the travel characteristics of residents reflected by different travel modes, enriching the representation of regional residents' travel characteristics, thereby obtaining a more extensive and good result.

[0106] Step 5: Based on the functional probability distribution characteristics of each traffic analysis area, the K-means clustering algorithm is used to perform functional clustering on all traffic analysis areas, and C clusters for the urban traffic analysis areas are obtained by clustering, and each cluster is functionally labeled according to the functional characteristics in each cluster. Including:

[0107] Let the functional characteristics of the traffic analysis community d be the stack of the region-topic probability distribution vectors obtained by the subway, online car-hailing, and points of interest, that is:

[0108] feature d =(Vec taxiweek，d , Vec taxiweekend，d , Vec metroweek，d , Vec metroweekend，d , Vec poi，d ),

[0109] Where Vec taxiweek，d In the weekday online car-hailing topic model, the probability distribution vector of the traffic analysis community d belonging to each functional feature of the weekday online car-hailing, Vec taxiweekend,d In the weekend ride-hailing topic model, the probability distribution vector of the traffic analysis community d belonging to the various functional features of the weekend ride-hailing service, Vec metroweek,d In the weekday subway model, the probability distribution vector of the traffic analysis area d belonging to each functional feature of the weekday subway, Vec metroweekend,d In the rest day subway model, the probability distribution vector of the traffic analysis area d belonging to each functional feature of the rest day subway, Vec poi,dIt indicates the probability distribution vector of the functional characteristics of the traffic analysis cell d belonging to the point of interest in the point of interest model. The number of functional area clusters of the final cell is determined by the clustering evaluation index silhouette coefficient (silhouette coefficient):

[0110]

[0111] The vector composed of the functional distribution characteristics of each cluster is taken as the sample, a represents the average distance between the sample and all other elements in the same cluster, and all clusters that do not contain the sample point are traversed to calculate the average distance from the sample point to all points in these clusters, the minimum of which is b.

[0112] Calculate the corresponding silhouette coefficient index under different clustering numbers C, and take the C value that makes the silhouette coefficient index value the largest, which is the C clusters for the urban space traffic analysis area.

[0113] When calculating the clustering evaluation index, we traverse all possible clustering numbers C values ​​within a certain range, calculate the corresponding silhouette coefficients of different C values, and take the number of clusters that makes the silhouette coefficient value the largest, so as to obtain the C-type functional area for the urban traffic analysis area. The value of the silhouette coefficient is between -1 and 1. The larger the value, the more similar the functional characteristics of the traffic areas divided into the same type of area are, while the functional characteristics of different areas are obviously different, which also means that the corresponding division method is better. Figure 4 As shown, in this embodiment, the number of clusters corresponding to the maximum silhouette coefficient value with a cluster number greater than 2 is finally selected, that is, C=5, and the urban space in the study area is divided into five functional areas.

[0114] For each cluster obtained by clustering, the specific urban functional land category of the region is determined according to its regional characteristics and travel activity characteristic indicators. Among them, the regional characteristics include the frequency density and category ratio of the points of interest in each traffic analysis area. The frequency density and category ratio in the region are calculated by the following formulas:

[0115]

[0116]

[0117] Among them, i represents the type of interest point, P represents the total number of interest point categories, and n i Represents the number of interest points of the i-th category in a certain area, N i represents the total number of interest points of type i in the study area; FD i represents the frequency density of the i-th type of interest point to the total number of interest points of this type; CR i It represents the ratio of the frequency density of the i-th type of interest point to the frequency density of all types of interest points in a certain area;

[0118] The regional travel activity characteristic index includes the normalized hourly flow of multi-source travel data on weekdays and weekends in the region, which is calculated by the following formula:

[0119]

[0120] Where X norm represents the normalized hourly flow rate of the region, X represents the original hourly flow rate of the region, and X max and X min Respectively represent the maximum and minimum values ​​of the regional hourly flow, Figure 5 The normalized flow line graph of each cluster of online ride-hailing vehicles on weekdays in this embodiment is shown.

[0121] By analyzing the travel characteristics of residents reflected in the normalized hourly flow of weekdays and weekends in regional multi-source travel data and comparing them with the travel characteristics of residents in typical urban functional land, regional functions are labeled based on regional characteristics and travel activity characteristics. The final determined regional functional attributes include office areas, residential areas or leisure areas, etc. Figure 6 The figure shows a distribution of functional areas marked in this embodiment.

[0122] It should be noted that the K-means algorithm used in the present embodiment for clustering functional features is a prior art, and the K-means algorithm will not be described in detail.

[0123] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0124] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for identifying urban functional areas based on multi-source traffic travel data and topic models, characterized in that: The method comprises the following steps: Step 1: Obtain road network data, point of interest data, and multi-source traffic travel data; Step 2: Based on the road network data, the urban space is divided into a number of traffic analysis areas using the morphological processing corrosion and expansion method; Step 3: According to the distribution of the POI data and online car-hailing and subway travel data in the traffic analysis community, obtain regional travel-activity features to construct vocabulary; Step 4: using a latent Dirichlet allocation topic model algorithm to obtain the functional probability distribution of each of the traffic analysis communities; Step 5: Based on the functional probability distribution of each traffic analysis cell, the K-means algorithm is used to perform functional clustering on all traffic analysis cells to obtain C clusters for the urban space, and each cluster is functionally labeled according to the regional travel-activity characteristic distribution in each traffic analysis cell; The specific method of step three is: 4.1 Extract the geographic information features of the points of interest, and use the frequency of the categories of points of interest and the land use mixture weighted according to the average area of ​​each category of points of interest as the vocabulary of the topic model. The category frequency and regional mixture of the weighted points of interest in all regions are used as the vocabulary library of the topic model to obtain the probability distribution of the traffic analysis area-travel characteristics composed of the categories of points of interest and the probability distribution of the topics of points of interest in each traffic analysis area; the calculation method of the land use mixture adopts the calculation method based on the entropy index, as shown below: Among them, p i represents the proportion of the i-th type of interest point to all types of interest points, and P represents the total number of interest point categories; 4.2 Extracting travel-activity features, the travel-activity features are the basic flow characteristics of weekdays and weekends; each travel data contains two travel features, namely, boarding and getting off, and the specific form is feature = (direction, T j ), where direction indicates the direction of subway passengers entering or exiting the station or getting on or off the online car-hailing service. For the online car-hailing service boarding point and the subway passenger entry point, direction = S. For the online car-hailing service boarding point and the subway passenger exit point, direction = E. For online car-hailing service orders, T j represents the time when passengers get on and off the bus in the online car-hailing order. For subway data, T j Indicates the time when subway passengers enter and exit the station; The specific method of step 4 is: 5.1 Using the latent Dirichlet allocation probability topic model, the functional distribution of each traffic analysis community in the data set is given in the form of a probability distribution vector, where: p(w|d)=p(w|k)×p(k|d) (1) in d is a document, which is represented as a traffic analysis cell in a specific application scenario; w is a vocabulary, which represents the regional travel-activity characteristics of the traffic analysis cell; k is the theme, which represents the urban functional characteristics of the traffic analysis area; The model uses k as the middle layer, and p(k|d) represents the probability that each traffic analysis cell d in the total set of traffic analysis cells D corresponds to a different urban function k. Its distribution is represented by the vector θ d , p(w|k) represents the probability of each urban function k in the total urban function set generating different regional travel-activity characteristics w, and its distribution is represented by the vector β k ;Use the topic model algorithm to solve the probability distribution of urban function characteristics of each traffic analysis area, as well as the regional travel-activity characteristics distribution in the urban function characteristics distribution; 5.2 The learning process of the urban functional area feature recognition model based on the latent Dirichlet allocation probability topic model is as follows: The regional travel-activity characteristics in the urban traffic analysis area are generated by the following steps: 1) The Dirichlet distribution determined by the hyperparameter α generates the functional characteristic probability distribution θ of the traffic analysis cell d d ,Right now in, is the hyperparameter of the distribution, which is a K-dimensional vector. K represents the total number of urban functions in a given traffic analysis area. The probability density function of the Dirichlet distribution is as follows: Among them, p k represents the probability of occurrence of the kth urban function, α k is the hyperparameter corresponding to the k-th city function; 2) The Dirichlet distribution determined by the hyperparameter η generates the travel characteristic conditional distribution β of the functional characteristic k k ,Right now Among them, η is the hyperparameter of the travel feature distribution, and the dimension is the number of elements V in the travel feature set; 3) The functional characteristics of the nth travel characteristics of the traffic analysis cell d are calculated by calculating the functional characteristic distribution θ of the traffic analysis cell d d The multinomial probability distribution z for the nth travel feature d Get, that is, z d =multi(θ d ), the probability density function of the multinomial distribution is: Among them, p i represents the probability of occurrence of the i-th urban function, θ d，i is the occurrence frequency of the ith functional feature of the traffic analysis cell d; The function feature-travel feature conditional probability distribution corresponding to each function feature k is β k , according to its conjugate polynomial distribution φ k Randomly select the travel characteristics, that is, φ k =multi(β k ), the probability density function of the multinomial distribution is: Among them, p i represents the probability of occurrence of the i-th travel feature, β k，i is the occurrence frequency of the i-th travel feature; 4) Determine the number of urban functions K based on perplexity. When using the latent Dirichlet allocation topic model for modeling, use the perplexity of the trained model in the test set to determine the optimal number of functional features. If the perplexity of the model on the test set is less than the preset value, it means that the current value of the number of functional features K meets the requirements. The perplexity calculation formula is as follows: Where D represents the total set of all traffic analysis cells; M is the total number of traffic analysis cells; W d is the travel feature set in the traffic analysis area d; N d is the number of travel characteristics in the traffic analysis area d; P(W d ) is the likelihood value of all travel characteristics in the traffic analysis area d; 5) The regional interest point feature dataset, the pre-processed online car-hailing, subway weekday and weekend datasets are respectively used for topic modeling to obtain the corresponding probability distribution of traffic analysis community function characteristics and form the function feature vectors of each traffic analysis community.

2. The method for identifying urban functional areas based on multi-source traffic travel data and topic models according to claim 1 is characterized in that: In step one, the point of interest data is geographic spatial point data divided into P categories, including name, address, and coordinates, obtained after data cleaning and reclassification; the multi-source transportation travel data is the travel data of urban online car-hailing and subway obtained after data cleaning and screening; the road network data is screened road network data including highways, urban expressways, and urban trunk roads.

3. The method for identifying urban functional areas based on multi-source traffic travel data and topic models according to claim 1 is characterized in that: In step 2, the road network data is processed by morphological expansion and corrosion to divide the urban space into different basic units.

4. The method for identifying urban functional areas based on multi-source traffic travel data and topic models according to claim 1 is characterized in that: The specific method of step 5 is: 1) Let the functional feature distribution of the traffic analysis area d be the stack of the probability distribution vectors of the traffic analysis area-functional feature obtained by combining the subway, online car-hailing, and points of interest, that is, feature d =(Vec taxiweek，d , Vec taxiweekend，d , Vec metroweek，d , Vec metroweekend，d , Vec poi，d ), where Vec taxiweek，d In the weekday online car-hailing topic model, the probability distribution vector of the traffic analysis community d belonging to each functional feature of the weekday online car-hailing, Vec taxiweekend，d In the weekend ride-hailing topic model, the probability distribution vector of the traffic analysis community d belonging to the various functional features of the weekend ride-hailing service, Vec metroweek，d In the weekday subway model, the probability distribution vector of the traffic analysis area d belonging to each functional feature of the weekday subway, Vec metroweekend，d In the rest day subway model, the probability distribution vector of the traffic analysis area d belonging to each functional feature of the rest day subway, Vec poi，d It represents the probability distribution vector of each functional feature of the point of interest in the point of interest model. 2) The number of functional area clusters in the final traffic analysis area is determined by the silhouette coefficient of the cluster evaluation index: The vector composed of the functional distribution characteristics of each cluster is taken as a sample, a represents the average distance between the sample and all other elements in the same cluster, and all clusters that do not contain the sample point are traversed to calculate the average distance from the sample point to all points in these clusters, where the minimum value is b; Calculate the corresponding silhouette coefficient index under different clustering numbers C, and take the C value that makes the silhouette coefficient index value the largest, which is the C clusters for the urban space traffic analysis area; 3) For each cluster obtained by clustering, the specific urban functional land category of the area is determined according to its regional characteristics and travel activity characteristic indicators; the regional characteristics include the frequency density and category ratio of the points of interest in each traffic analysis area, and the frequency density and category ratio in the region are calculated by the following formulas: Among them, i represents the type of interest point, P represents the total number of interest point categories, and n i Represents the number of interest points of the i-th category in a certain area, N i represents the total number of interest points of type i in the study area; FD i represents the frequency density of the i-th type of interest point to the total number of interest points of this type; CR i It represents the ratio of the frequency density of the ith type of interest point to the frequency density of all types of interest points in a certain area. The regional travel activity characteristic index includes the normalized hourly flow of multi-source travel data on weekdays and weekends in the area, which is calculated by the following formula: Among them, X norm represents the normalized hourly flow rate of the region, X represents the original hourly flow rate of the region, and X max and X min Respectively represent the maximum and minimum hourly flow rates in the area.

Images