CN105206048B - A system and method for discovering transfer modes of urban resident groups based on traffic OD data - Google Patents

Abstract

The invention discloses a system and method for discovering transfer modes of urban residents based on traffic OD data, including a data preprocessing module, a transfer trajectory identification module, a transfer hotspot area discovery module and a group transfer mode mining module; the invention establishes In a large number of real residents' bus travel card swiping data, it is robust and universally applicable. The discovery of transfer modes within the city provides data support and reasonable suggestions for the optimization of public transportation routes and dynamic vehicle scheduling.

Description

A system for discovering transfer modes of urban residents based on traffic OD data and its method

technical field

The invention belongs to the technical field of intelligent transportation, and proposes a system and method for discovering transfer modes of urban resident groups based on traffic OD data.

Background technique

Public transportation is not only an economical and green way of travel for residents, but also an important means of alleviating urban congestion. According to statistics, 60% of the urban population spends at least 1 hour a day on public transportation (how many people take the bus, how many people take the subway). However, the continuous development of urban construction has led to dynamic changes in the distribution of urban functional areas and points of interest, which has changed the distribution of urban passenger flow. For example, newly built shopping malls and commercial districts will stimulate new travel needs. Therefore, how to improve residents' public travel efficiency and dynamically discover new public travel needs is very important for transportation service providers and urban planning.

At present, public transport options for citizens to travel mainly include buses, subways, and public rental bicycles. Buses and subways are the economical and convenient modes of travel preferred by users. The advantages of this type of transportation are fast, wide coverage, and time-saving and labor-saving; another feature is that it is often necessary to transfer from the departure point to a destination. The transfer with the bus or the transfer between the bus and the subway has also become a difficult point in public transportation planning.

Transfer is an important part of multi-modal transportation. Taking the morning rush hour in Beijing as an example, one fifth of people cannot reach their destination directly by a single journey and need to transfer in the middle. Transfers increase travel costs, increase walking distance and travel time, and significantly reduce travel efficiency. According to research, 80% of the residents choose the direct plan when traveling and avoid transfers, so it is reasonable to use transfers as a key factor to measure travel efficiency. In addition, the deficiencies of the existing transportation system can be found through transfers. For example, when the transfer passenger flow exceeds a certain threshold and the frequency of this situation is relatively high, it can be found that the bus planning is unreasonable, and further through the travel route The research can provide the demand basis for setting up direct bus lines; when special events such as holidays or concerts occur, there will be irregular transfer needs, which can provide data support for dynamic vehicle scheduling and traffic congestion relief. From the perspective of the transfer area, it is also possible to understand and grasp the passenger's travel changes, and the real destination of the traveler can be calculated through the records of the transfer points.

However, it is difficult to accurately collect the transfer behavior of passengers. Traditional methods such as questionnaires have regional limitations and are expensive. One-cards are used for urban residents to take the main means of transportation, buses and subways, which imply important information about bus and subway transfers. Compared with cellphones and mobile phones, it describes residents' public travel characteristics in an approximate full-sample and full-time manner to meet the needs. The data used in the present invention is based on the bus and subway data of AFCS, covering 95% of the bus transfer trips of urban residents.

In addition, in urban public transportation, the research on individual travel behavior has great reference value for the formulation and management of urban transportation policies, such as commonly used travel modes and travel distances. Since individual travel behavior is personalized and random to a certain extent, the analysis of individual travel behavior cannot reflect the overall travel characteristics of the city. Due to the planning of public transport stations, residents who use public transport to travel often form passenger flow "groups". This "group" represents a group of people with similar travel purposes in a certain period of time and space. Due to the commonality of groups, larger travel groups can reflect characteristics that cannot be reflected in individual trips in travel groups, such as tidal changes in platform congestion and changes in the spatial-temporal density of urban passenger flows. Therefore, when analyzing the urban public transport transfer mode, it is necessary to analyze the group transfer mode.

Bus-to-bus exchange passenger flow, subway-to-bus exchange passenger flow has always been a key point in the study of travel characteristics, and the study of transfer stations is of great significance in the field of urban planning.

Since it is difficult to accurately collect passengers' transfer behaviors, and the actual number of transfers in the public transportation network is difficult to obtain, the evaluation of traditional transfer times indicators is often based on network analysis of the network topology, such as adjacency matrix analysis, Floyd algorithm, etc. , has become one of the main obstacles restricting the evaluation of the service level of the bus network. Effective identification of transfers is the key to reasonable planning of public transport.

There are some foreign studies on the transfer based on the one-card data. By using the passenger boarding data to obtain the transfer information of public transportation and describe an iterative classification algorithm, this algorithm divides the passenger boarding data into two categories: transfer travel and single trip. Analysis of transfer node identification matrix, waiting time distribution map and spatial first and second station distribution map. This algorithm can provide traffic dispatchers with important information: passenger boarding, transfer and waiting time and other relevant information. This information can provide powerful data support for their transportation planning and policy formulation.

Due to the limitation of basic data, there are few researches on the transfer mode of public transportation at home and abroad. Modeling and analysis of residents' transfer behavior between integrated public transportation.

Contents of the invention

The technical problem of the present invention is to overcome the deficiencies of the prior art, and provide a system and method for discovering transfer modes of urban residents based on traffic OD data, which is based on a large number of real residents' bus travel card swiping data, which is robust and reliable. Universal applicability, solves the transfer demand problem of existing urban residents, and provides data support for public transportation route optimization and dynamic vehicle scheduling.

Technical solution of the present invention: a system for discovering transfer modes of urban residents based on traffic OD data, including the following modules: data preprocessing module, transfer track identification module, transfer hotspot area discovery module, group transfer mode Mining modules.

It belongs to the field of intelligent transportation technology. The transfer mode represents the spatio-temporal characteristics of transfer trajectories with frequent large passenger flows, and reflects the transfer needs of urban residents. The model includes a data preprocessing module, a transfer trajectory identification module, a transfer hotspot area discovery module and a group transfer pattern mining module. The data preprocessing module is used to process the card swiping data of buses and subways to obtain OD data; the transfer trajectory identification module judges whether the OD data of continuous travel of individuals is transfer from the input data OD data of this model through the time threshold and space threshold, and obtains personal Transfer trajectory data; the hotspot area discovery module judges the degree of passenger flow aggregation of the station and its adjacent stations through a clustering algorithm to obtain the hotspot area. On this basis, the hotspot area is merged and divided twice through the traffic area data to obtain the hot traffic area. That is, transfer hotspot areas; the group transfer mode mining module obtains group transfer trajectory data by aggregating individual transfer trajectories, and defines group transfer modes. The traffic data itself has high-dimensional characteristics. By establishing a high-dimensional data description model and using the reduced Dimensional method to mine out the group transfer mode of large passenger flow between hotspot areas. This model is based on a large number of real residents' bus card swiping data. The model is robust and universally applicable. The discovery of transfer modes within the city provides data support and reasonable suggestions for the optimization of public transportation lines and dynamic vehicle scheduling.

Data preprocessing module: used to process bus and subway card swiping data to obtain personal complete bus travel OD data as the basic input data of the subsequent transfer identification module; the specific implementation is as follows:

The input data is bus card swiping data, subway card swiping data and other data sources. Firstly, valid fields are extracted, abnormal data are eliminated, and then the time of getting on and off the bus, location and related information are filled, and the travel mode is marked; finally, by card number Steps such as merging and time serialization to obtain bus OD data;

The OD data from different data sources are merged by card number, and sorted by boarding time to obtain personal complete bus travel spatio-temporal OD data, which mainly includes boarding and boarding time information, spatial location information of boarding and boarding stations, etc.

Transfer track recognition module: through the time-space OD data of urban residents’ bus travel, extract according to the same card number, judge two consecutive OD data, set the transfer recognition time threshold Th and space threshold Td, and time threshold judge two consecutive OD data Whether the time interval between the previous time of getting off the bus and the next time of getting on the bus is less than Th, the spatial threshold determines whether the distance between the previous time of getting off the bus and the next time of getting on the bus is less than Td, and if the above two conditions are met, it is judged as one time The transfer trajectories are deduced sequentially, and multiple transfer trajectories in a row can be judged.

Transfer hotspot area discovery module: define the passenger flow density of the station, use the passenger flow at the station as the density and the average distance between stations as the radius through density clustering, take the DBSACN method as an example to obtain the hotspot area, and conduct secondary analysis on the hotspot area through the traffic area data Secondary merging and division to obtain hotspot traffic areas, division method: a. If the distance (surface distance) between two stations in the clustered hotspot area is greater than the threshold dis, segmentation is required. b. If more than half of the sites in the hotspot area are distributed in a certain traffic area, it will be included in this traffic area. Repeat the above steps until all the hotspot areas are processed, and the hotspot traffic area is obtained as the transfer hotspot area.

Group transfer mode mining module: defines the group transfer mode and uses a high-dimensional tensor model to represent the multi-dimensional spatio-temporal characteristics of the group transfer mode. The group transfer mode is defined as follows: for the transfer trajectory R _O → R _T → R _D the departure hotspot area is R _O , after passing through the hot transfer area R _T , the arrival hotspot area is R _D , the group transfer mode is defined as an n-dimensional vector:

each element Represents the passenger flow whose departure time is in the i-th time slice and the transfer trajectory is R _O → R _T → R _D.

Build a 4D tensor model Describe the spatio-temporal characteristics of the group transfer mode, including one-dimensional time information and three-dimensional space information, that is, departure, transfer, and arrival areas. tensor The value of the (i,j,k,l)th element of is equal to In view of the unbalanced spatio-temporal distribution of traffic data, the data is log-smoothed, that is,

Using non-negative tensor factorization (implemented by CP decomposition) to decompose the fourth-order tensor X into a finite sum of R rank tensors X ^(r) , the decomposition is as follows:

The optimal projection matrix is obtained by ALS algorithm. Among them, a two-stage method is used to extract the significant transfer mode for the decomposed single rank tensor: candidate mode selection and candidate mode ranking, considering the importance of the area and λ _r in each factor vector, respectively. Candidate mode selection is to sort the spatial factor vectors of the rank tensor by size, introduce a threshold ω to indicate the magnitude of the value drop and limit the number of modes, using the formula Select the salient candidate region R _i to represent the O region, and obtain the salient T and D regions in the same way. Candidate mode sorting means that each rank tensor selects significant regions corresponding to O, T, and D through candidate mode selection. Suppose O selects m ₁ regions, T selects m ₂ regions, and D selects m ₃ regions, the total number of significant modes obtained is m ₁ *m ₂ *m ₃ . According to the selected group transfer mode, the number of people in the transfer mode group is restored, and further screening is performed to ensure the large passenger flow characteristics of the significant transfer mode.

Compared with the prior art, the present invention has the advantages of obtaining the complete personal travel trajectory through the one-card data processing, and the transfer is often ignored in previous studies and only analyzed for OD; in addition, a clustering method based on passenger flow density is proposed instead of Only based on geographical location clustering, the effectiveness of hotspot area discovery is guaranteed; high-dimensional spatio-temporal modeling and tensor decomposition methods are proposed to mine significant transfer modes, which represent groups with great transfer needs and their transfer spatio-temporal characteristics. Traffic route optimization and dynamic vehicle scheduling provide data support and reasonable suggestions.

Description of drawings

Fig. 1 is the overall flowchart of the present invention;

Figure 2 is the distribution map of hot spots in Beijing;

Figure 3 is a schematic diagram of non-negative tensor decomposition;

Figure 4 is the spatio-temporal distribution diagram of significant transfer modes.

detailed description

The present invention proposes an urban residents group transfer mode discovery system based on traffic OD data ("O" comes from English ORIGIN, pointing out the starting point of the trip, and "D" comes from English DESTINATION, pointing out the destination of the trip) including data Preprocessing module, transfer trajectory identification module, transfer hotspot area discovery module and group transfer mode mining module. In order to make the object, technical solution and advantages of the present invention clearer, each module of the present invention will be further described in detail below in conjunction with the accompanying drawings.

As shown in Figure 1, the specific implementation of the present invention is as follows:

1. Data preprocessing module

The input data is bus card swiping data, subway card swiping data and other data sources. Firstly, valid fields are extracted, abnormal data are eliminated, and then the time of getting on and off the bus, location and related information are filled, and the travel mode is marked; finally, by card number Steps such as merging and time serialization to obtain bus OD data;

The OD data from different data sources are merged by card number, and sorted by boarding time to obtain personal complete bus travel spatio-temporal OD data, which mainly includes boarding and boarding time information, spatial location information of boarding and boarding stations, etc.

2. Transfer trajectory recognition module

The definitions involved:

(1) Personal Transfer

Personal transfer T _r means that a passenger departs from station S _O at time h _O and arrives at station S _D at time h _D , forming an OD record: T _r = {S _O , h _O , S _D , h _D }(h _O < _hD ).

(2) Personal trajectory

An individual trajectory TR is a collection of individual transfers over a day: Each T _r has timing.

(3) Transfer trajectory

If a certain trajectory of a person contains two adjacent transitions,

with

That is, if the time interval between the previous alighting time and the next boarding time is less than Th, Th=30min, whether the distance between the previous alighting point and the next boarding point is less than Td, Td=1km, if the above two conditions are met, it is judged as one trip Change track.

3. Discovery module of transfer hotspot area

The definitions involved:

(1) Passenger flow density

Define the site S _x , take it as the center, and a unit density clustering area with a radius of r K refers to the total number of stations falling in this area, if there is only one station S _x in this area, then A _x =S ^x , K=1. S _x site traffic vector where t is the number of time slices, is the site passenger flow intensity of the site in the i-th time slice. To scan all stations within the area A _x , define is the passenger flow density in the area A _x in the i-th time slice, and the passenger flow density in the area A _x at this time

if Satisfy That is, the regional aggregation intensity threshold per unit time slice In this dimension, the area A _{x is} taken as a candidate hotspot area H _x , otherwise, the site S _x is marked as an outlier. Take the new area A _y generated by any other station S _y in the area A _x and calculate the area A _y in the same way as above If in any dimension there is Then add the sites contained in A _y to H _x , H _x ＝H _y ＝H _x ∪H _y , otherwise mark S _y as an outlier site. Iterate in this way until all stations are processed, and finally obtain the set H _x ={S _k } constituting the hotspot area.

Carry out secondary merging and division of the hotspot area through the traffic area data to obtain the hotspot traffic area. The division method: a. If the distance (surface distance) between two stations in the clustered hotspot area is greater than the threshold 1km, then need to split. b. If more than half of the sites in the hotspot area are distributed in a certain traffic area, it will be included in this traffic area. Repeat the above steps until all the hotspot areas are processed, and the hot traffic area is obtained as the transfer hotspot area, as shown in Figure 2.

4. Group transfer mode mining module

The definitions involved:

(1) Group transfer mode

For the transfer trajectory R _O → R _T → R _D , the departure hotspot area is R _O , passes through the hotspot transfer area R _T , and arrives at the hotspot area R _D , the group transfer mode is defined as an n-dimensional vector each element Represents the passenger flow whose departure time is in the i-th time slice and the transfer trajectory is R _O → R _T → R _D.

(2) Significant group transfer mode

If the group transfer mode is significant in a certain time slice, its passenger flow value is dominant among all modes.

Build a 4D tensor model Describe the spatio-temporal characteristics of the group transfer mode, including one-dimensional time information and three-dimensional space information, that is, departure, transfer, and arrival areas. tensor The value of the (i,j,k,l)th element of is equal to In view of the unbalanced spatio-temporal distribution of traffic data, log smoothing is performed on the data, that is,

Using non-negative tensor factorization (CP decomposition) to decompose the fourth-order tensor X into a finite sum of R rank tensors X ^(r) , the decomposition is as follows: As shown in Figure 3, the projection matrices O, T, D, and H of each dimension are composed of vectors of rank tensors, O=[o ₁ o ₂ ...o _R ], and T, D are obtained by the same method , H.

Each X ^(r) has a weight, and the vector λ(λ ₁ ,λ ₂ ,λ ₃ ,...,λ _R ) represents the weight vector of each decomposed X ^(r) . The decomposed elementalization is expressed as

The rank tensor is expressed as X ^(r) = λ _r o _r οt _r οd _r οh _r . Solve the formula by ALS algorithm:

Add non-negative constraints O≥0, T≥0, D≥0, H≥0, for tensors where the Frobenius norm is Get a rank tensor consisting of the optimal projection matrix.

Among them, a two-stage method is used to extract the significant transfer pattern P' for each rank tensor after decomposition: candidate pattern selection and candidate pattern sorting, respectively considering the importance of the area and λ _r in each factor vector.

Sort the spatial factor vectors of the rank tensor by size, introduce the threshold ω to indicate the magnitude of the value drop and limit the number of modes, using the formula:

Select the salient candidate region R _i to represent the O region, and obtain the salient T and D regions in the same way.

After the above steps, the significant regions corresponding to O, T, and D are selected, assuming that O selects m1 regions, T selects m2 regions, and D selects m3 regions, then the total number of significant patterns obtained is m1 *m2*m3 pieces. Restore the number of group transfers according to the selected group transfer mode,

The set of significant transfer modes with large passenger flow is filtered out as follows:

P′=P′ ₁ ∪P′ ₂ ∪...P′ _r

An example of its spatio-temporal characteristics is shown in Figure 4. The time distribution can be seen to be the characteristics of the morning peak, and the spatial distribution shows that this group pattern starts from the areas numbered R ₅₀ and R ₂₀ , transfers to the R ₂ area, and finally arrives at the R ₄ area. The groups corresponding to each transfer route are 300 and 240 people respectively.

So far, the construction and implementation of the method for discovering the transfer mode of urban residents based on traffic OD data has been completed.

Claims (10)

1. A system for discovering transfer patterns of urban residents based on traffic OD data, characterized in that: it includes a data preprocessing module, a transfer trajectory identification module, a transfer hotspot area discovery module, and a group transfer pattern mining module; Data preprocessing module: used to process bus and subway card swiping data to obtain OD data of urban residents' bus travel, which is used as the basic input data of the transfer track identification module; Transfer trajectory identification module: Based on the OD data of urban residents' bus travel, the transfer trajectory data of urban residents' travel is obtained according to the time threshold parameters and space threshold parameters; Transfer hotspot area discovery module: define the station passenger flow density, take the station passenger flow as the density, take the average distance between stations as the radius, obtain the hotspot area through density clustering, and use the traffic area data to merge and divide the hotspot area twice to obtain Hot traffic area; Group transfer mode mining module: According to the transfer trajectory identification module to obtain the transfer trajectory data of urban residents and the hot traffic districts obtained by the transfer hotspot area discovery module, obtain group transfer trajectory data, define group transfer mode, and traffic data It has high-dimensional characteristics, establishes a high-dimensional tensor model, and uses the dimensionality reduction decomposition method to dig out the group transfer mode of large passenger flow between hotspot areas. 2. the urban resident group transfer pattern discovery system based on traffic OD data according to claim 1, is characterized in that: described data preprocessing module is concretely realized as follows: The input data is bus card swiping data and subway card swiping data from multiple data sources. Firstly, valid fields are extracted, abnormal data are eliminated, and then the time and location of getting on and off the bus are filled, and the travel mode is marked; finally, the card number is merged and time series The personal OD data is obtained through the steps of simplification; the OD data from different data sources are combined by card number, and sorted by the boarding time to obtain the complete personal travel time and space OD data, which mainly includes the time information of boarding and boarding, and the spatial location information of boarding and boarding stations. 3. The urban resident group transfer mode discovery system based on traffic OD data according to claim 1, characterized in that: the transfer trajectory identification module is implemented as: based on the OD data of urban resident bus travel, extracting according to the same card number , to determine two consecutive OD data, set the transfer recognition time threshold Th and space threshold Td, the time threshold judges whether the time interval between the previous alighting time and the next boarding time of two consecutive OD data is less than Th, and the spatial threshold judges whether two consecutive OD data Whether the distance between the previous alighting point and the next boarding point of the OD data is less than Td, if the above two conditions are met, it is judged as a transfer trajectory, and it is recursively deduced in turn to determine the continuous multiple transfer trajectories, that is, the travel of urban residents transfer trajectory data. 4. the urban resident group transfer mode discovery system based on traffic OD data according to claim 1, is characterized in that: the method for dividing hot spot traffic sub-district in the described transfer hotspot area discovery module is: a. if clustering is complete If the distance between two stations in the hotspot area is greater than the threshold dis, it needs to be divided; b. If more than half of the stations in the hotspot area are distributed in a certain traffic area, it will be classified into this traffic area; repeat The above steps are performed until all the hotspot areas are processed, and the hot traffic area is obtained as the transfer hotspot area. 5. the urban resident group transfer pattern discovery system based on traffic OD data according to claim 1, is characterized in that: the defined group transfer pattern in the described group transfer pattern mining module is as follows: for transfer track R _O → R _T → R _D The departure hotspot area is R _O , after passing through the transfer hotspot area R _T , the arrival hotspot area is R _D , and the group transfer mode is defined as an n-dimensional vector: <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>R</mi> <mi>O</mi> </msub> <mo>&amp;RightArrow;</mo> <msub> <mi>R</mi> <mi>T</mi> </msub> <mo>&amp;RightArrow;</mo> <msub> <mi>R</mi> <mi>D</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mo>(</mo> <mrow> <msubsup> <mi>v</mi> <mn>1</mn> <mrow> <msub> <mi>R</mi> <mi>O</mi> </msub> <mo>&amp;RightArrow;</mo> <msub> <mi>R</mi> <mi>T</mi> </msub> <mo>&amp;RightArrow;</mo> <msub> <mi>R</mi> <mi>D</mi> </msub> </mrow> </msubsup> <mo>,</mo> <mn>...</mn> <mo>,</mo> <msubsup> <mi>v</mi> <mi>n</mi> <mrow> <msub> <mi>R</mi> <mi>O</mi> </msub> <mo>&amp;RightArrow;</mo> <msub> <mi>R</mi> <mi>T</mi> </msub> <mo>&amp;RightArrow;</mo> <msub> <mi>R</mi> <mi>D</mi> </msub> </mrow> </msubsup> </mrow> <mo>)</mo> <mo>,</mo> </mrow> each element Represents the passenger flow whose departure time is in the i-th time slice and the transfer trajectory is R _O → R _T → R _D. 6. The urban resident group transfer pattern discovery system based on traffic OD data according to claim 1, characterized in that: the high-dimensional tensor model set up in the group transfer pattern mining module is: Among them, I ⁽ⁱ⁾ represents the number of elements in the i-th dimension, and N represents the largest dimension. The high-dimensional tensor model X contains spatio-temporal information. The unit, the spatial dimension includes the place of departure and the place of arrival, and the three information of the place of transfer is included when transfer is considered. 7. The system for discovering transfer modes of urban residents based on traffic OD data according to claim 6, characterized in that: said model is less than or equal to four dimensions. 8. The system for discovering transfer modes of urban residents based on traffic OD data according to claim 1 or 6, characterized in that: aiming at the unbalanced spatio-temporal distribution of traffic data, the high-dimensional tensor model data is smoothed. 9. The urban resident group transfer mode discovery system based on traffic OD data according to claim 1, characterized in that: the process of obtaining the significant group transfer mode of large passenger flow by the method of dimension reduction and decomposition is: (1) Use non-negative tensor factorization to decompose the high-order tensor X into a finite sum of R rank tensors X ^(r) with weight λ _r , R is a positive integer, and each rank tensor implicitly Contains significant transfer modes independent of other modes, broken down as follows: <mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <mi>X</mi> <mo>&amp;ap;</mo> <mrow> <mo>&amp;lsqb;</mo> <mrow> <mi>&amp;lambda;</mi> <mo>;</mo> <msup> <mi>A</mi> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </msup> <mo>,</mo> <msup> <mi>A</mi> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </msup> <mo>,</mo> <mn>...</mn> <mo>,</mo> <msup> <mi>A</mi> <mrow> <mo>(</mo> <mi>N</mi> <mo>)</mo> </mrow> </msup> </mrow> <mo>&amp;rsqb;</mo> </mrow> <mo>&amp;equiv;</mo> <msup> <mi>X</mi> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </msup> <mo>+</mo> <msup> <mi>X</mi> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </msup> <mo>+</mo> <mn>...</mn> <mo>+</mo> <msup> <mi>X</mi> <mrow> <mo>(</mo> <mi>r</mi> <mo>)</mo> </mrow> </msup> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>&amp;equiv;</mo> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>r</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>R</mi> </munderover> <msub> <mi>&amp;lambda;</mi> <mi>r</mi> </msub> <mrow> <mo>(</mo> <mrow> <msubsup> <mi>a</mi> <mi>r</mi> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </msubsup> <mo>&amp;times;</mo> <msubsup> <mi>a</mi> <mi>r</mi> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </msubsup> <mo>&amp;times;</mo> <mn>...</mn> <mo>&amp;times;</mo> <msubsup> <mi>a</mi> <mi>r</mi> <mrow> <mo>(</mo> <mi>N</mi> <mo>)</mo> </mrow> </msubsup> </mrow> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced> Among them, the vector Find optimal projection matrix A ^(r) by ALS algorithm; (2) For the decomposed single rank tensor X ^(r), a two-stage method is used to extract significant transfer modes: candidate mode selection and candidate mode ranking, respectively considering the region R _i and weight in each factor vector The importance of λ _r ; candidate mode selection is about the spatial factor vector of the rank tensor Sorting by size, introducing the threshold ω to indicate the magnitude of the value drop and limiting the number of modes, using the formula Select the salient candidate region R _i to represent the O region, and obtain the salient T and D regions in the same way, and T represents the transfer; the candidate mode sorting means that each rank tensor selects the salient points corresponding to O, T and D through candidate mode selection. region, assuming that O selects m ₁ regions, T selects m ₂ regions, and D selects m ₃ regions, then the total number of significant patterns obtained is m ₁ *m ₂ *m ₃ , according to The selected group transfer mode restores the number of groups in the transfer mode for further screening to ensure the large passenger flow characteristics of the significant transfer mode. 10. A method for discovering transfer patterns of urban residents based on traffic OD data, characterized in that: the implementation steps are as follows: (1) For the one-card swiping data, the OD data of urban residents' bus travel are obtained through the steps of effective field extraction, abnormal data cleaning, boarding time filling, station basic information filling, and sorting by card number and boarding time; (2) Carry out transfer trajectory identification based on urban resident travel OD data, and obtain transfer trajectory data according to time threshold parameters and space threshold parameters; (3) Define the passenger flow density of the station, take the passenger flow at the station as the density, take the average distance between the stations as the radius, obtain the hotspot area through density clustering, and use the traffic area data to merge and divide the hotspot area twice to obtain the hot traffic area; (4) According to the transfer trajectory data of urban residents travel obtained in step (1) and the hot traffic area obtained in step (2), the group with the same transfer trajectory based on the region is obtained, and the group transfer mode is given Define and establish a high-dimensional tensor model to describe the spatio-temporal characteristics of the transfer mode; (5) Based on the high-dimensional tensor model of step (4), and using the dimensionality reduction decomposition method to obtain the significant group transfer mode of large passenger flow.