Big data analysis method for macroscopic travel mode

Abstract

The invention relates to a multi-day continuous big data analysis method for macroscopic trip mode recognition, and the method comprises the steps: obtaining trip data which comprises a trip track chain, trip environment data and holiday and festival data of passengers taking public transport within a certain time period; constructing the travel trajectory chains in the same time period into travel OD matrixes, and superposing the travel OD matrixes in a plurality of different time periods to obtain travel space-time tensors of all public transport in a certain time period; expanding the travel space-time tensor along a time dimension to obtain a continuous space-time travel matrix, and decomposing the continuous space-time travel matrix through GoDec decomposition to obtain a low-rank matrix L, a sparse matrix S and a noise matrix N; analyzing the low-rank matrix L to obtain travel space distribution characteristics with certain regularity; and in combination with the travel environment data and the travel time, analyzing the sparse matrix S to obtain the influence of the travel environment data and the travel time on the travel demand.

Description

Technical Field

[0001] The present invention relates to the technical field of transportation travel demand analysis, and in particular to a big data analysis method for macro travel patterns. Background Art

[0002] Travel pattern analysis is of great significance in modern traffic management, urban planning and sustainable development. It can provide data support for urban planning, help decision makers allocate resources rationally, and provide support for formulating differentiated traffic policies and refined traffic planning and management.

[0003] With the popularization and development of intelligent public transportation systems, massive amounts of data have been accumulated in the charging, dispatching and other subsystems of intelligent public transportation systems. How to use advanced data mining technology to analyze this massive amount of information, discover the implicit travel modes and rules, and obtain the travel patterns in space that are affected by time.

[0004] For example, the Chinese patent with publication number CN114936785A discloses "a travel characteristics analysis system, including a travel distribution characteristics module, a travel statistics characteristics module, a job-residence characteristics statistics module, a job-residence balance analysis module and a data statistics analysis module, which analyzes the characteristics of various travel modes to clarify the current status of multi-mode transportation." However, this analysis system only focuses on the mining and analysis of travel patterns in a single dimension of space, and does not break through the two-dimensionality of space to capture macro travel characteristics from the dual dimensions of time and space.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] In view of the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a big data analysis method for macro travel patterns, which captures macro travel characteristics from the dual dimensions of time and space.

[0007] The technical solution of the present invention is as follows:

[0008] A multi-day continuous big data analysis method for macro travel pattern identification comprises: obtaining travel data, the travel data comprising a travel trajectory chain of passengers taking public transportation within a certain time period, travel environment data and holiday data; constructing the travel trajectory chain within the same time period into a travel OD matrix, superimposing the travel OD matrices within multiple different time periods, and obtaining a travel space-time tensor of all public transportation within a certain time period; expanding the travel space-time tensor along the time dimension to obtain a continuous space-time travel matrix, and decomposing the continuous space-time travel matrix through GoDec decomposition to obtain a low-rank matrix L, a sparse matrix S and a noise matrix N; analyzing the low-rank matrix L to obtain travel spatial distribution characteristics with certain regularity; combining the travel environment data and travel time, analyzing the sparse matrix S to obtain the influence of the travel environment data and travel time on travel demand.

[0009] A further technical solution is to uniformly number all the stations that the public transportation passes through, and the row and column indexes of the travel OD matrix correspond to the starting point and the end point of the travel trajectory chain respectively; correspondingly, the travel time-space tensor of all public transportation within a certain period of time is:

[0010] D∈R t×m×n

[0011] Wherein t represents the date of travel; m represents the starting point of the travel trajectory chain, that is, the passenger's boarding station; n represents the end point of the travel trajectory chain, that is, the passenger's alighting station.

[0012] 11. The multi-day continuous big data analysis method for macro travel mode identification according to claim 1 is characterized in that: the GoDec decomposition takes minimizing the matrix decomposition error as the objective function, and its mathematical model is as follows:

[0013]

[0014] Among them, ||·|| F represents the F norm; r and c represent the preset thresholds; rank(L) represents the rank of the low-rank matrix L; card(S) represents the potential of the sparse matrix s, that is, the number of non-zero elements in the matrix;

[0015] Further decomposing the above model until the convergence condition is met, we get:

[0016]

[0017] Then iterate each other L t and S t , and set the hard threshold to solve the global optimal solution, we have:

[0018]

[0019] Among them, L t By DS t-1 Update the first r singular vectors of S t By DL t The largest non-zero subset of c elements is updated.

[0020] A further technical solution is that the continuous space-time travel matrix is ​​decomposed by GoDec decomposition to obtain a low-rank matrix L and a sparse matrix S, and the specific steps include:

[0021] (1) Set the decomposition error ε, set the thresholds r and c, and input the space-time travel matrix And initialize the iteration coefficient t=0;

[0022] (2) Initialize the low-rank matrix L0 and the sparse matrix S0 by the following formula;

[0023]

[0024] Among them, med(D i )∈R t×1 is the median of the i-th row of matrix D; 1 1×n2 For 1 row n 2 The unit vector of the column;

[0025] (3) Perform convergence calculation. If the convergence condition is met, stop the subsequent iteration and output the low-rank matrix L t and the sparse matrix S t , otherwise continue with the subsequent iterations:

[0026] like Then stop the iteration, otherwise execute the next step;

[0027] (4) Let t = t + 1, and use the matrix power strategy. Alternative DS t-1 , improve the degree of singular value attenuation, thereby significantly speeding up the decomposition while ensuring accuracy, where q>0 is the power index and T is the matrix transpose;

[0028] (5) Combine bilateral random projection to replace SVD singular value decomposition to reduce the computational time complexity. Generate two random matrices through standard Gaussian distribution Each element in the matrix is ​​generated by an independent and identically distributed Gaussian random variable with mean 0 and variance 1. Based on A1 and A2, the following left and right random projections are constructed:

[0029]

[0030] In order to further reduce the low-rank approximation error, the following improved method is used to construct bilateral random projections Y1 and Y2 to obtain higher approximation accuracy:

[0031] Construct right random projection through random matrix A1 Pick Construct left random projection renew The right random projection is updated to

[0032] (6) To ensure that the matrix rank is ≤ r and improve the computational efficiency, QR matrix decomposition is applied to the bilateral random projection:

[0033]

[0034] (7) Combining (4) to (6), the low-rank matrix is ​​iteratively updated according to the following formula:

[0035]

[0036] (8) Iteratively update the sparse matrix according to the following formula

[0037] S t =P Ω (DL t )

[0038] Among them, P Ω (·) represents the projection of the matrix onto the element set Ω, which is the matrix DL t The non-zero subset consisting of the first c largest elements in;

[0039] (9) Convergence judgment is performed. If Then stop the iteration and output the low-rank matrix L t and the sparse matrix S t , otherwise return to (4) for the next round of update iteration;

[0040] After multiple iterations, the low-rank matrix L = L t , sparse matrix S = S t , and the noise matrix N = DL t -S t A further technical solution is that the analysis method of the low-rank matrix L is to screen the main column vectors in the matrix and convert them into the numbers of the starting and ending sites through the column index; the analysis method of the sparse matrix S is to screen the non-zero components with larger absolute values ​​in the matrix and convert them into the numbers of the starting and ending sites through the column index.

[0041] A further technical solution is that the travel data is obtained by analyzing raw data; the raw data includes GPS data of public transportation, public transportation stops and passengers' ride records within the same time period, and the ride records include the train number and boarding time.

[0042] A further technical solution is that, corresponding to the GPS data and the ride record, the starting point of the ride record, that is, the starting point of the travel trajectory chain, can be obtained.

[0043] A further technical solution is that the original data also includes POI points of interest and land use properties near the public transportation station; the attraction coefficient of the station is determined by the POI points of interest and land use properties, and the probability of getting off at the station is calculated; the getting off probabilities of all possible getting off stations are combined, and the passenger's getting off station, that is, the end point of the travel trajectory chain is determined by a roulette algorithm.

[0044] A further technical solution is that the probability of getting off the bus at the station is determined by the following formula:

[0045]

[0046] Where i is the boarding station; j is the subsequent station on the same line (i<j); D ij is the attraction strength of the site with respect to the travel distance, which is calculated as follows:

[0047]

[0048] F j is the site attraction strength regarding the distribution of POI points of interest, which is calculated as follows:

[0049] F j =-∑f jk ·logf jk +∑N jk

[0050] L j is the site attraction intensity with respect to land use mix, which is calculated as follows:

[0051] L j =-∑α k h jk logh jk

[0052] Among them, N jk 、f jk The number and proportion of the kth type of interest points within 500 meters of the corresponding station j; α k 、h jkis the attraction coefficient and area ratio of the kth land type within 1 km of the site j. A further technical solution is to set a time threshold and the spatial threshold γ s For the same passenger's ride record, if the boarding time difference between two consecutive ride records is less than the time threshold And the train route of the previous ride record and the ride station of the subsequent ride record are less than the spatial threshold γ s , it can be determined that the two ride records are the same trip, that is, the terminal station of the previous ride record is the starting station of the later ride record.

[0053] The beneficial technical effects of the present invention are as follows

[0054] (1) The big data analysis method of the macro travel mode in the present invention performs big data analysis based on travel data. The travel data includes the travel trajectory chain and travel environment data of passengers taking public transportation within a certain period of time. After data processing, GoDec decomposition can be used to obtain a low-rank matrix L, a sparse matrix S and a noise matrix N. The low-rank matrix L can reflect the spatial distribution characteristics of passenger flow with certain time regularity. Differentiated bus vehicle operation and scheduling plans can be formulated according to the component sizes corresponding to these travel trajectory chains to improve the quality and efficiency of urban bus services. In addition, the noise matrix N separates and suppresses the strong noise components in the sparse matrix S, and cooperates with the travel environment data and time that change with time to obtain the changes in travel demand caused by time and travel environment data. Combining the spatial distribution characteristics of travel with certain regularity and the impact of time and travel environment on travel, the macro travel characteristics captured from the dual dimensions of time and space are obtained, which is convenient for transportation planners to accurately predict future travel trends and formulate dynamic optimization and adjustment plans for bus lines, thereby optimizing public transportation services.

[0055] (2) Furthermore, the travel data of passengers does not need to be obtained directly. It can be obtained by analyzing the GPS data of public transportation vehicles, including the ride records of the train number and boarding time, which simplifies the difficulty of collecting travel data.

[0056] (3) Furthermore, a time threshold is set and the spatial threshold γ s , to determine whether two consecutive ride records of the same passenger are the same trip. If they are the same trip, the terminal station of the first ride record can be directly obtained, and the terminal station is the starting station of the later ride record, which improves the accuracy of travel data and further improves the accuracy of travel pattern analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1A flow chart of a big data analysis method for macro travel patterns in the first embodiment of the present invention is shown.

[0058] Figure 2 A flowchart is shown in which a continuous space-time travel matrix is ​​decomposed to obtain a low-rank matrix L and a sparse matrix S in Embodiment 1 of the present invention.

[0059] Figure 3 The flowchart of calculating travel data from original data in the second embodiment of the present invention is shown. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical scheme and advantages of the present invention clearer, the device proposed by the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. According to the following description, the advantages and features of the present invention will be clearer. It should be noted that the accompanying drawings adopt a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, please refer to the accompanying drawings. It should be noted that the structure, proportion, size, etc. illustrated by the accompanying drawings of this specification are only used to match the content disclosed in the specification, so that people familiar with this technology can understand and read, and are not used to limit the limiting conditions for the implementation of the present invention, so they do not have technical substantive significance. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effect that the present invention can produce and the purpose that can be achieved, should still fall within the scope of the technical content disclosed by the present invention.

[0061] In the description of the present invention, the orientations or positional relationships indicated by terms such as “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside”, “outside”, “axial”, “radial”, and “circumferential” are defined as orientations or positional relationships based on those shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be understood as limiting the present invention.

[0062] Embodiment 1

[0063] A multi-day continuous big data analysis method for macro travel mode identification, the specific implementation steps include:

[0064] S110, obtaining travel data, the travel data includes the travel trajectory chain of passengers taking public transportation within a certain period of time, travel environment data and holiday data. The travel trajectory chain includes the boarding and alighting stations when passengers take public transportation. The travel environment data includes external related data such as weather conditions and road conditions within the travel period.

[0065] S120: constructing the travel trajectory chain in the same time period into a travel OD matrix, and superimposing the travel OD matrices in multiple different time periods to obtain the travel space-time tensor of all public transportation within a certain period of time.

[0066] This includes uniformly numbering all the stops along the public transportation route as No. = 1, 2, ... n for easy recording. The row and column indexes of the travel OD matrix correspond to the starting point and end point of the travel trajectory chain, that is, the numbers of different stops.

[0067] Correspondingly, the travel time and space tensor of all public transportation within a certain period of time is:

[0068] D∈R t×m×n

[0069] Where t represents the date of travel. m represents the starting point of the travel trajectory chain, i.e. the passenger's boarding station. n represents the end point of the travel trajectory chain, i.e. the passenger's alighting station.

[0070] In addition, the obtained travel trajectory chains can be classified, for example, they can be classified according to the age of passengers, and travel OD matrices can be constructed separately. The travel OD matrices can be superimposed to obtain the travel time-space tensors of public transportation for different groups of people after classification, which is convenient for analyzing travel patterns for different types of people.

[0071] S130, expand the travel space-time tensor D∈R along the time dimension t×m×n , and obtain the continuous space-time travel matrix D, and decompose the continuous space-time travel matrix D through GoDec decomposition to obtain the low-rank matrix L, sparse matrix S and noise matrix N.

[0072] Among them, GoDec decomposition takes minimizing the matrix decomposition error as the objective function, and its mathematical model is as follows:

[0073]

[0074] Among them, ||·|| F represents the F norm of the matrix. r and c represent the preset thresholds. rank(L) represents the rank of the low-rank matrix L. card(S) represents the cardinality of the sparse matrix s, that is, the number of non-zero elements in the matrix.

[0075] Further decomposing the above model until the convergence condition is met, we get:

[0076]

[0077] Then iterate each other L t and S t , and set the hard threshold to solve the global optimal solution, we have:

[0078]

[0079] Among them, L t By DS t-1 Update the first r singular vectors of S t By DL t The largest non-zero subset of c elements is updated.

[0080] Specifically, the specific steps of decomposing the continuous space-time travel matrix by GoDec decomposition to obtain the low-rank matrix L and the sparse matrix S include:

[0081] (1) Set the decomposition error ε, set the thresholds r and c, and input the space-time travel matrix And initialize the iteration coefficient t=0;

[0082] (2) Initialize the low-rank matrix L0 and sparse matrix S0 using the following formula.

[0083]

[0084] Among them, med(D i )∈R t×1 is the median of the i-th row of matrix D. 1×n2 For 1 row n 2 The unit vector of the column.

[0085] (3) Perform convergence calculation. If the convergence condition is met, stop the subsequent iteration and output the low-rank matrix L t and the sparse matrix S t , otherwise continue with the subsequent iterations:

[0086] like Then stop the iteration, otherwise execute the next step.

[0087] (4) Let t = t + 1, and use the matrix power strategy. Alternative DS t-1 , improve the degree of singular value attenuation, thereby significantly accelerating the decomposition while ensuring accuracy, where q>0 is the power exponent and T is the matrix transpose.

[0088] (5) Combine bilateral random projection to replace SVD singular value decomposition to reduce the computational time complexity. Generate two random matrices through standard Gaussian distribution Each element in the matrix is ​​generated by an independent and identically distributed Gaussian random variable with mean 0 and variance 1. Based on A1 and A2, the following left and right random projections are constructed: Y1∈R t×t :

[0089]

[0090] In order to further reduce the low-rank approximation error, the following improved method is used to construct bilateral random projections Y1 and Y2 to obtain higher approximation accuracy:

[0091] Construct right random projection through random matrix A1 Pick

[0092] Construct left random projection renew The right random projection is updated to

[0093] (6) To ensure that the matrix rank is ≤ r and improve the computational efficiency, QR matrix decomposition is applied to the bilateral random projection:

[0094]

[0095] (7) Combining (4) to (6), the low-rank matrix is ​​iteratively updated according to the following formula

[0096]

[0097] (8) Iteratively update the sparse matrix according to the following formula

[0098] S t =P Ω (DL t )

[0099] Among them, P Ω (·) represents the projection of the matrix onto the element set Ω, which is the matrix DL t The non-zero subset consisting of the first c largest elements in .

[0100] (9) Convergence judgment is performed. If Then stop the iteration and output the low-rank matrix L t and the sparse matrix S t , otherwise return to (4) for the next update iteration.

[0101] Specifically in this embodiment, q is taken as 2, and the decomposition error ε is taken as 10 -7 .

[0102] After multiple iterations, the low-rank matrix L = L t , sparse matrix S = S t , and the noise matrix N = DL t -S t .

[0103] S140, Analyzing Low-Rank Matrices Obtain travel spatial distribution characteristics with certain regularity. Combine travel environment data and travel time to analyze the sparse matrix Obtain travel environment data and the impact of travel time on travel demand.

[0104] Specifically, the low-rank matrix The analysis method is to select the main column vectors in the matrix and convert them into the numbers of the starting and ending stations through the column index. In this way, a travel trajectory chain with continuous and stable bus passenger flow demand can be obtained, which can reflect the spatial distribution characteristics of passenger flow with certain time regularity. Differentiated bus vehicle operation and scheduling plans can be formulated according to the size of the components corresponding to these travel trajectory chains, thereby improving the quality and efficiency of urban bus services.

[0105] Specifically, the sparse matrix S∈R t×n2 The analysis method is to combine external related data such as meteorological conditions and road conditions in the same time period and holiday data, select non-zero components with larger absolute values ​​in the matrix and convert them into the numbers of the starting and ending stations through column indexes. The changes in passenger travel demand caused by meteorological conditions such as cooling, rainy and foggy weather, and slippery roads and holiday data in the time period can be analyzed. According to the spatial distribution characteristics of these elastic demands, a dynamic optimization and adjustment plan for the bus network can be formulated. Through flexible supply, high-standard all-weather travel services can be achieved, and the overall traffic operation efficiency can be improved.

[0106] Furthermore, if in step S120, the acquired travel trajectory chain is classified in advance, for example, by age, then the low-rank matrix and the sparse matrix reflect the travel patterns of people of different ages, as well as the changes in travel demand of people of different ages caused by meteorological conditions and holidays.

[0107] It can be seen that in the above-mentioned embodiment 1, a continuous spatiotemporal travel matrix D is constructed through travel data, and GoDec is used for decomposition processing and noise reduction to obtain macro travel characteristics captured from the dual dimensions of time and space, which is convenient for traffic planners to accurately predict future travel trends and formulate dynamic optimization and adjustment plans for bus lines, thereby optimizing public transportation services.

[0108] Embodiment 2

[0109] Based on the first embodiment, the second embodiment discloses a method for obtaining travel data, where the travel data is obtained by analyzing the original data. The specific analysis steps include:

[0110] S210. The original data includes GPS data of public transportation, public transportation stops and passengers' riding records within the same time period. The riding records include the bus number and boarding time.

[0111] Taking the bus system as an example, the original data includes bus CPS data, bus IC card data and bus station information. When collecting bus IC card data, it is necessary to clean the data and delete duplicate and missing data.

[0112] S220 corresponds to the GPS data and the ride record, and the starting point of the ride record, that is, the starting point of the travel trajectory chain, can be obtained. Specifically, the various data are first classified according to the vehicle operation route and the vehicle number and sorted by time to facilitate the execution of subsequent steps. Then the boarding time in the ride record is compared with the entry and exit time of the vehicle with the corresponding vehicle number at different stations. If the boarding time is within the entry and exit time period of a certain stop station, the starting point of the ride record is at the stop station, that is, the starting point of the travel trajectory chain formed by the ride record is the stop station.

[0113] Furthermore, due to recording errors, the boarding time of some ride records cannot find the corresponding entry and exit time period. At this time, a time error value can be set according to the error size. Even if a boarding time is not within a certain entry and exit time period, as long as the difference between the boarding time and the entry and exit time period is less than the time error value, the boarding station corresponding to the boarding time is determined to be the station where the entry and exit time period is located.

[0114] S230. Setting a time threshold and the spatial threshold γ s For the same passenger's ride record, if the boarding time difference between two consecutive ride records is less than the time threshold And the train route in the previous ride record and the boarding station in the subsequent ride record are less than the spatial threshold γ s , it can be determined that the two ride records are the same trip, that is, the terminal station of the earlier ride record is the starting station of the later ride record.

[0115] Furthermore, different time thresholds can be set for different groups of passengers. and the spatial threshold γ s For example, for older passengers, who have longer transfer times, a larger time threshold can be set. For young users, they are more sensitive to the value of time and spend relatively less time on transfers, so a smaller time threshold is set.

[0116] S240. The original data also includes POI points of interest and land use properties near the stations of public transportation, which can be obtained through various geographic map websites. Among them, POI points of interest can count the information entropy of points of interest within 500 meters of the station, and land use properties can count the land use types and areas within 1 kilometer around the station. Combined with the established attraction coefficients of different land use types, the attraction degree and mixed degree of land use around the station are calculated.

[0117] Then, the attraction coefficient of the station is determined by the POI and land use characteristics, and the probability of getting off the station is calculated. For passengers who are not traveling on the same trip, the probability of getting off at all possible getting off stations is combined, and the passenger's getting off station, that is, the end point of the travel trajectory chain, is determined by the roulette algorithm.

[0118] Specifically, the probability of getting off at a station is determined by the following formula:

[0119]

[0120] Where i is the boarding station and j is the subsequent station on the same line (i<j); D ij is the site attraction intensity with respect to travel distance, which is calculated as follows:.

[0121]

[0122] F j is the site attraction strength regarding the distribution of POI points of interest, which is calculated as follows:

[0123] F j =-∑f jk ·logf jk +∑N jk

[0124] L j is the site attraction intensity with respect to land use mix, which is calculated as follows:

[0125] L j =-∑α k h jk logh jk

[0126] Among them, N jk 、f jk The number and proportion of the kth type of interest points within 500 meters of the corresponding station j. k 、h jk is the attraction coefficient and area ratio of the kth land type within 1 km of site j.

[0127] This embodiment does not need to directly obtain the travel trajectory chain of passengers taking public transportation. It can be deduced only through the GPS data of public transportation tools, public transportation stop data, and riding records including train numbers and boarding times, which greatly reduces the difficulty of obtaining data for analysis.

[0128] 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.

[0129] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A multi-day continuous big data analysis method for macro travel mode identification, characterized in that: include: Acquire travel data, wherein the travel data includes a travel trajectory chain of passengers taking public transportation within a certain period of time, travel environment data, and holiday data; The travel trajectory chain in the same time period is constructed into a travel OD matrix, and the travel OD matrices in multiple different time periods are superimposed to obtain the travel space-time tensor of all public transportation within a certain period of time; Expand the travel space-time tensor along the time dimension to obtain a continuous space-time travel matrix, and decompose the continuous space-time travel matrix through GoDec decomposition to obtain a low-rank matrix L, a sparse matrix S and a noise matrix N; analyze the low-rank matrix L to obtain travel space distribution characteristics with certain regularity; combine the travel environment data and travel time to analyze the sparse matrix S to obtain the impact of the travel environment data and travel time on travel demand.

2. The multi-day continuous big data analysis method for macro travel mode identification according to claim 1 is characterized by: All the stations that the public transportation passes through are uniformly numbered, and the row and column indexes of the travel OD matrix correspond to the starting point and the end point of the travel trajectory chain respectively; correspondingly, the travel time-space tensor of all public transportation within a certain period of time is: D∈R t×m×n Wherein t represents the date of travel; m represents the starting point of the travel trajectory chain, that is, the passenger's boarding station; n represents the end point of the travel trajectory chain, that is, the passenger's alighting station.

3. The multi-day continuous big data analysis method for macro travel mode identification according to claim 1 is characterized by: The GoDec decomposition takes minimizing the matrix decomposition error as the objective function, and its mathematical model is as follows: Among them, ||·|| F represents the F norm; r and c represent the preset thresholds; rank(L) represents the rank of the low-rank matrix L; card(S) represents the potential of the sparse matrix s, that is, the number of non-zero elements in the matrix; Further decomposing the above model until the convergence condition is met, we get: Then iterate each other L t and S t , and set the hard threshold to solve the global optimal solution, we have: Among them, L t By DS t-1 Update the first r singular vectors of S t By DL t The largest non-zero subset of c elements is updated.

4. The multi-day continuous big data analysis method for macro travel mode identification according to claim 3 is characterized by: The specific steps of decomposing the continuous space-time travel matrix by GoDec decomposition to obtain the low-rank matrix L and the sparse matrix S include: (1) Set the decomposition error ε, set the thresholds r and c, and input the space-time travel matrix And initialize the iteration coefficient t=0; (2) Initialize the low-rank matrix L0 and the sparse matrix S0 by the following formula; Among them, med(D i )∈R t×1 is the median of the i-th row of matrix D; For 1 row n 2 The unit vector of the column; (3) Perform convergence calculation. If the convergence condition is met, stop the subsequent iteration and output the low-rank matrix L t and the sparse matrix S t , otherwise continue with the subsequent iterations: like Then stop the iteration, otherwise execute the next step; (4) Let t = t + 1, and use the matrix power strategy. Alternative DS t-1 , improve the degree of singular value attenuation, thereby significantly speeding up the decomposition while ensuring accuracy, where q>0 is the power index and T is the matrix transpose; (5) Combine bilateral random projection to replace SVD singular value decomposition to reduce the computational time complexity; generate two random matrices through standard Gaussian distribution Each element in the matrix is ​​generated by an independent and identically distributed Gaussian random variable with mean 0 and variance 1; based on A1 and A2, the following left and right random projections are constructed: Y1∈R t×t : In order to further reduce the low-rank approximation error, the following improved method is used to construct bilateral random projections Y1 and Y2 to obtain higher approximation accuracy: Construct right random projection through random matrix A1 Pick Construct left random projection renew The right random projection is updated to (6) To ensure that the matrix rank is ≤ r and improve the computational efficiency, QR matrix decomposition is applied to the bilateral random projection: (7) Combining (4) to (6), the low-rank matrix is ​​iteratively updated according to the following formula (8) Iteratively update the sparse matrix according to the following formula S t =P Ω (D-L t ) Among them, P Ω (·) represents the projection of the matrix onto the element set Ω, which is the matrix DL t The non-zero subset consisting of the first c largest elements in; (9) Convergence judgment is performed. If Then stop the iteration and output the low-rank matrix L t and the sparse matrix S t , otherwise return to (4) for the next round of update iteration; After multiple iterations, the low-rank matrix L = L t , sparse matrix S = S t , and the noise matrix N = DL t -S t .

5. The multi-day continuous big data analysis method for macro travel mode identification according to claim 2, characterized in that: The analysis method of the low-rank matrix L is to screen the main column vectors in the matrix and convert them into the numbers of the starting and ending sites through column indices; the analysis method of the sparse matrix S is to screen the non-zero components with larger absolute values ​​in the matrix and convert them into the numbers of the starting and ending sites through column indices.

6. The multi-day continuous big data analysis method for macro travel mode identification according to claim 1 is characterized by: The travel data is obtained by analyzing the original data; the original data includes GPS data of public transportation, public transportation stops and passengers' riding records within the same time period, and the riding records include the train number and boarding time.

7. The multi-day continuous big data analysis method for macro travel mode identification according to claim 6 is characterized by: Corresponding to the GPS data and the ride record, the starting point of the ride record, that is, the starting point of the travel trajectory chain, can be obtained.

8. The multi-day continuous big data analysis method for macro travel mode identification according to claim 6 is characterized by: The original data also includes POI points of interest and land use properties near the station of the public transportation vehicle; the attraction coefficient of the station is determined by the POI points of interest and the land use properties, and the probability of getting off at the station is calculated; The alighting probabilities of all possible alighting locations are comprehensively considered, and the alighting location of the passenger, i.e., the end point of the travel trajectory chain, is determined through a roulette algorithm.

9. The multi-day continuous big data analysis method for macro travel mode identification according to claim 8, characterized in that: The determination of the probability of getting off the bus at the station is derived from the following formula: Where i is the boarding station; j is the subsequent station on the same line (i<j); D ij is the attraction strength of the site with respect to the travel distance, which is calculated as follows: F j is the site attraction strength regarding the distribution of POI points of interest, which is calculated as follows: F j =-∑f jk ·logf jk +∑N jk L j is the site attraction intensity with respect to land use mix, which is calculated as follows: L j =-∑α k h jk forgive jk Among them, N jk 、f jk The number and proportion of the kth type of interest points within 500 meters of the corresponding station j; α k 、h jk is the attraction coefficient and area ratio of the kth land type within 1 km of site j.

10. The multi-day continuous big data analysis method for macro travel mode identification according to claim 7, characterized in that: Set a time threshold and the spatial threshold γ s For the same passenger's ride record, if the boarding time difference between two consecutive ride records is less than the time threshold And the train route of the previous ride record and the ride station of the subsequent ride record are less than the spatial threshold γ s , it can be determined that the two ride records are the same trip, that is, the terminal station of the previous ride record is the starting station of the later ride record.

Images