Passenger public transport dependence level identification and influence factor sensitivity analysis method

Abstract

The invention discloses a passenger public transport dependence level identification and influence factor sensitivity analysis method, and relates to the technical field of public transport, and the method comprises the steps: obtaining the public transport travel behavior data and individual attribute data of a passenger; based on the data, constructing a public transport dependency level measurement index; a Twood-step clustering algorithm is adopted, category data and continuous data are processed, the optimal cluster number is automatically determined, and clustering recognition is conducted on the public transport dependence level of passengers; on the basis of sensitivity analysis of an Ordered Logit regression model, evaluating and identifying key factors influencing the public transport dependency level of passengers; and quantitatively analyzing the influence of the ratio to the independent variable, and determining the sensitivity degree of each key factor on the dependence level of passenger public transport. The invention provides an efficient, comprehensive and accurate public traffic behavior and influence analysis framework, which not only is helpful for improving the operation efficiency of a public traffic system, but also can provide a scientific basis for optimization of traffic policies and design of personalized services.

Description

Technical Field

[0001] The present invention relates to the technical field of public transportation, and in particular to a method for identifying a passenger's public transportation dependence level and analyzing the sensitivity of influencing factors. Background Art

[0002] With the continuous advancement of urbanization, the public transportation system, as an important part of urban transportation, bears an increasingly heavy travel demand. In this context, understanding and evaluating the level of public transportation dependence of passengers is of great significance for optimizing transportation policies, improving the quality of public transportation services and formulating precise transportation management measures. The level of public transportation dependence refers to the degree of dependence of passengers on public transportation in daily travel, which is usually affected by many factors, including individual attributes of passengers, travel behavior, transportation policies and accessibility of public transportation systems. The study of the level of public transportation dependence can help transportation management departments better understand the travel needs of different passenger groups and provide data support for urban transportation planning and public transportation facility layout. For example, groups that are highly dependent on public transportation may be concentrated in areas with convenient transportation, while groups with low dependence may be more inclined to use private cars or other travel modes. Therefore, by accurately identifying and dividing the level of public transportation dependence of passengers, it can provide an important basis for the rational allocation of public transportation resources, the prediction of travel demand and the formulation of policies.

[0003] At present, the research on public transportation dependence level mainly relies on methods such as questionnaire survey and traffic big data analysis. Questionnaire survey is to collect data such as passengers' personal attributes and travel behavior by designing targeted questionnaires, and use statistical analysis methods to evaluate passengers' dependence level. However, traditional questionnaire surveys have problems such as insufficient sample representativeness and complex data processing, and the results are greatly affected by survey design and sampling methods. On the other hand, traffic big data analysis methods usually rely on smart card swiping records, GPS trajectory data, etc., and identify passengers' travel patterns and traffic dependence through data mining technology. It has high accuracy, but it is insufficient in individual attribute analysis. Summary of the invention

[0004] In order to solve the above technical problems, a method for identifying the level of passenger dependence on public transportation and sensitivity analysis of influencing factors is provided. This technical solution solves the above problems.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A method for identifying the level of passenger public transportation dependence and sensitivity analysis of influencing factors, comprising:

[0007] Obtain passengers' public transportation travel behavior data and individual attribute data;

[0008] Based on the data, construct public transportation dependency level measurement indicators, the measurement indicators include travel behavior indicators and individual attribute indicators, forming a multi-dimensional dependency level measurement system;

[0009] A two-step clustering algorithm is used to process categorical and continuous data, automatically determine the optimal number of clusters, and perform cluster identification on passengers’ public transportation dependence levels;

[0010] Based on the sensitivity analysis of the Ordered Logit regression model, the key factors affecting the level of passengers' public transportation dependence are evaluated and identified;

[0011] The influence of the odds ratio independent variables was quantitatively analyzed to determine the sensitivity of each key factor to the level of passengers' public transportation dependence.

[0012] Preferably, the acquisition of the public transportation travel behavior data and individual attribute data of the passenger specifically includes:

[0013] Design a questionnaire covering public transportation travel behavior and individual attributes, including travel behavior data, individual attribute data, and transportation policy data;

[0014] Travel behavior data include the proportion of travel days, the average number of daily trips, and the round-trip nature of travel modes;

[0015] Individual attribute data include age, occupation, education level, income, and car availability;

[0016] Through stratified sampling, sample groups were selected according to research objectives and data requirements, and questionnaire surveys were conducted; the collected questionnaire data were preliminarily screened to eliminate invalid data;

[0017] Clean the collected data and delete duplicate, erroneous and invalid records;

[0018] In the questionnaire, the passengers' smart card numbers are collected, and the travel behavior data is associated with the card swiping transaction records. By matching the data with the card swiping records in the public transportation system background, the detailed travel data of each passenger is obtained;

[0019] Combine all questionnaires and smart card data to build a complete travel dataset;

[0020] Standardize individual attribute and travel behavior data, including normalizing continuous variables such as income, travel days ratio Y, and average daily travel times, and convert categorical data into numerical form of categorical variables;

[0021] The collected valid data are organized and stored in the database.

[0022] Preferably, based on the data, a public transportation dependency level measurement index is constructed, and the measurement index includes a travel behavior index and an individual attribute index, and the formation of a multi-dimensional dependency level measurement system specifically includes:

[0023] Define the measurement objectives of public transport dependence level, including travel behavior indicators and individual attribute indicators;

[0024] The travel behavior indicators include: the proportion of travel days, data sources, measurement methods, average daily trip times, data sources, measurement methods and travel mode round trip;

[0025] The individual attribute indicators include: age, data source, measurement method, data source, measurement method, education level, income and car availability;

[0026] Obtain the required travel behavior data and individual attribute data based on the questionnaire and traffic data, and build a public transportation dependence level measurement model;

[0027] By using the weighted summation method, the travel behavior and individual attribute indicators are combined into a public transportation dependence level value, where the weighted summation formula is:

[0028]

[0029] Where E is the dependency level, w i is the weight of the ith indicator, x i is the value of the ith indicator.

[0030] Preferably, the two-step clustering algorithm is used to process the categorical data and the continuous data, automatically determine the optimal number of clusters, and cluster the public transportation dependence level of passengers, specifically including:

[0031] Based on the Two-step algorithm, a recognition model for public transportation dependence level is established to automatically determine the optimal number of clusters;

[0032] The Two-step algorithm clustering process includes:

[0033] Read the data points in the data set one by one, generate a CF tree, remove outliers on the CF tree, and pre-cluster the data points in the dense area to form subclusters. When pre-clustering, process mixed data and perform hierarchical clustering based on the log-likelihood distance. The log-likelihood distance between two clusters i and j is:

[0034] d(i,j)=ζ i +ζ j -ζ ＜i,j＞

[0035]

[0036] In the formula, ζ i and j are the likelihood function values ​​of clusters i and j, respectively. The index 〈i,j〉 represents the new cluster generated by combining clusters i and j. K1 is the total number of continuous variables, K2 is the total number of categorical variables, and L k is the number of k-th categorical variables, N S is the total amount of data in cluster S, N skl The categorical variable k in the S cluster is divided into l groups, is an estimate of the variance of the continuous variable k in the entire data set, and is the variance estimate of the continuous variable k in cluster S, is the estimated mean of the kth continuous variable in cluster s.

[0037] By inputting the sub-clusters of the final CF leaf elements in the pre-clustering stage, the clusters with the closest distance are recursively merged using the idea of ​​agglomeration until the optimal number of clusters is obtained. The optimal number of clusters J * The Bayesian information criterion is combined with the ratio of the nearest cluster distance r(J) to determine the calculation formula as follows:

[0038]

[0039] Among them, J1 and J2 are the number of clusters corresponding to the two largest subsets in the r(J) set.

[0040] The BIC criterion is selected accurately, and the goodness of fit is compared for a series of clustering results with different numbers of clusters generated by combining leaf nodes based on the agglomerative hierarchical clustering method, so as to automatically determine the best clustering scheme.

[0041] Preferably, the BIC criteria specifically include:

[0042] Among them, for cluster J, the calculation formula of the BIC criterion is as follows:

[0043]

[0044] Where N is the number of public transportation travel dependence level measurement indicator data sets, is the maximum value of the likelihood function, m J is the number of clustering model parameters, L k is the number of k-th categorical variables.

[0045] Preferably, the categorical variables and continuous variable data are input into the identification model in sequence. Based on the existing literature results and the iterative operation results after multiple model parameter adjustments, the maximum number of branches of the model CF tree is set to 8, the maximum tree depth is set to 3, and the optimal number of classifications of public transportation dependence level is determined according to the BIC information criterion. Finally, the group is divided into 4 categories according to the public transportation travel dependence level indicator. The clustering silhouette coefficient value is 0.8, and the proportions of the groups with high, relatively high, relatively low and low public transportation travel dependence levels are 50%, 17%, 15% and 18% respectively. Among them, more than two-thirds of the passengers in the high and relatively high dependence level groups mainly commuter travel, which proves the effectiveness of the public transportation dependence level identification model and questionnaire survey data.

[0046] Preferably, the sensitivity analysis based on the Ordered Logit regression model to evaluate and identify key factors affecting the level of passenger public transportation dependence specifically includes:

[0047] Based on the multivariate ordered logit model, a sensitivity analysis of the factors affecting the level of public transportation dependence is conducted. The multiple categories of the dependent variable are divided into multiple binary logistic regressions in turn, and the assumption is that the independent variable coefficients of the multiple binary logistic regression functions after the split are equal but the constant terms are unequal. Among them, the utility function U of the public transportation traveler's dependence level k on the traveler i under the transportation mode choice t is ikt The expression is as follows:

[0048]

[0049] Where, X ikt x is the factor affecting the level of public transportation dependence i , i=1,2,...,N,k=1,2,3,4,t is a 0-1 binary variable, i.e., whether or not to choose public transportation; is the regression parameter vector corresponding to the factors affecting the level of public transportation dependence; ε represents the random error terms that are independent and identically distributed, ε ikt It is the sum of the effects of other factors that are difficult to observe but have an impact on the level of public transportation travelers’ reliance on public transportation. ikt is the utility function of public transport traveler dependence level k for travel individual i under transportation mode choice t;

[0050] When ε follows a Gumbel distribution, the Logit model is transformed into an OrderedLogit regression model.

[0051] Preferably, the public transportation dependence level of individual i is expressed as probability L ij The calculation formula is:

[0052]

[0053] Where, L ij is the probability of the public transportation dependence level of travel individual i, β is the regression parameter vector corresponding to the factors affecting the public transportation dependence level, and X is the independent variable that affects the traveler's dependence on public transportation.

[0054] Preferably, the complementary double logarithmic function is used as the link function of the multivariate OrderedLogit regression model;

[0055] The sensitivity of the changes in the variables of the influencing factors related to the level of public transportation travelers’ dependence on their dependence level was analyzed, and the odds ratio indicator was used for comparative analysis, which means the probability of the event occurring in the experimental group p i The probability of the event occurring in the control group p j The ratio is calculated as:

[0056]

[0057] In the formula, OR i is the odds ratio, P i is the probability of the event occurring in the experimental group, P j The probability of an event in the control group, β i are regression parameter variables corresponding to the influencing factors of public transportation dependence level.

[0058] Preferably, the key influencing factors of public transportation dependence level are screened based on the multivariate correlation analysis method, that is, the total time from home and destination to the site, income, car availability and public transportation accessibility are used as the independent variables of the multivariate Ordered Logit regression model, and the public transportation dependence level with 4 levels is the dependent variable of the model. The initial confidence level of the model is taken as 90%, that is, when the significance parameter is less than 0.100, the public transportation dependence level group has a significant sensitivity effect on the independent variable, and the model data set is input into the multivariate Ordered Logit regression model based on complementary double logarithm, and the final result is obtained after multiple iterative calculations.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The present invention proposes an efficient, comprehensive and accurate analysis framework, which can not only improve the operational efficiency of the public transportation system, but also provide a scientific basis for the optimization of transportation policies and the design of personalized services. It has important practical significance and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a step flow framework diagram of the present invention. DETAILED DESCRIPTION

[0062] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0063] Reference Figure 1 As shown, a method for identifying the level of passenger public transportation dependence and sensitivity analysis of influencing factors includes:

[0064] Step 1:

[0065] Design a questionnaire covering public transportation travel behavior and individual attributes, including travel behavior data, individual attribute data, and transportation policy data;

[0066] Travel behavior data include the proportion of travel days, the average number of daily trips, and the round-trip nature of travel modes;

[0067] Individual attribute data include age, occupation, education level, income, and car availability;

[0068] Through stratified sampling, sample groups were selected according to research objectives and data requirements, and questionnaire surveys were conducted; the collected questionnaire data were preliminarily screened to eliminate invalid data;

[0069] Clean the collected data and delete duplicate, erroneous and invalid records;

[0070] In the questionnaire, the passengers' smart card numbers are collected, and the travel behavior data is associated with the card swiping transaction records. By matching the data with the card swiping records in the public transportation system background, the detailed travel data of each passenger is obtained;

[0071] Combine all questionnaires and smart card data to build a complete travel dataset;

[0072] Standardize individual attribute and travel behavior data, including normalizing continuous variables such as income, travel days ratio Y, and average daily travel times, and convert categorical data into numerical form of categorical variables;

[0073] The collected valid data are organized and stored in the database.

[0074] Step 2:

[0075] Define the measurement objectives of public transport dependence level, including travel behavior indicators and individual attribute indicators;

[0076] The travel behavior indicators include: the proportion of travel days, the average number of daily trips and the round trip nature of travel modes;

[0077] The individual attribute indicators include: age, education level, income, and car availability;

[0078] Obtain the required travel behavior data and individual attribute data based on the questionnaire and traffic data, and build a public transportation dependence level measurement model;

[0079] By using the weighted summation method, the travel behavior and individual attribute indicators are combined into a public transportation dependence level value, where the weighted summation formula is:

[0080]

[0081] Where E is the dependency level, w i is the weight of the ith indicator, x i is the value of the ith indicator.

[0082] Step 3:

[0083] Based on the Two-step algorithm, a recognition model for public transportation dependence level is established to automatically determine the optimal number of clusters;

[0084] The Two-step algorithm clustering process includes:

[0085] Read the data points in the data set one by one, generate a CF tree, remove outliers on the CF tree, and pre-cluster the data points in the dense area to form subclusters. When pre-clustering, process mixed data and perform hierarchical clustering based on the log-likelihood distance. The log-likelihood distance between two clusters i and j is:

[0086] d(i,j)=ζ i +ζ j -ζ ＜i,j＞

[0087]

[0088] In the formula, ζ i and j are the likelihood function values ​​of clusters i and j, respectively. The index 〈i,j〉 represents the new cluster generated by combining clusters i and j. K1 is the total number of continuous variables, K2 is the total number of categorical variables, and L k is the number of k-th categorical variables, N S is the total amount of data in cluster S, N skl The categorical variable k in the S cluster is divided into l groups, is an estimate of the variance of the continuous variable k in the entire data set, and is the variance estimate of the continuous variable k in cluster S, is the estimated mean of the kth continuous variable in cluster s;

[0089] By inputting the sub-clusters of the final CF leaf elements in the pre-clustering stage, the clusters with the closest distance are recursively merged using the idea of ​​agglomeration until the optimal number of clusters is obtained. The optimal number of clusters J * The Bayesian information criterion is combined with the ratio of the nearest cluster distance r(J) to determine the calculation formula as follows:

[0090]

[0091] Among them, J1 and J2 are the number of clusters corresponding to the two largest subsets in the r(J) set.

[0092] The BIC criterion is selected accurately, and the goodness of fit is compared for a series of clustering results with different numbers of clusters generated by combining leaf nodes based on the agglomerative hierarchical clustering method, so as to automatically determine the best clustering scheme.

[0093] Preferably, the BIC criteria specifically include:

[0094] Among them, for cluster J, the calculation formula of the BIC criterion is as follows:

[0095]

[0096] Where N is the number of public transportation travel dependence level measurement indicator data sets, is the maximum value of the likelihood function, m J is the number of clustering model parameters, L k is the number of k-th categorical variables.

[0097] The categorical variables and continuous variable data were input into the identification model in sequence. Based on the existing literature results and the iterative operation results after multiple model parameter adjustments, the maximum number of branches of the model CF tree was set to 8, and the maximum tree depth was set to 3. The optimal number of classifications for public transportation dependence levels was determined according to the BIC information criterion. Finally, the groups were divided into 4 categories according to the public transportation travel dependence level indicators. The clustering silhouette coefficient value was 0.8, and the proportions of the groups with high, relatively high, relatively low and low public transportation travel dependence levels were 50%, 17%, 15% and 18% respectively. Among them, more than two-thirds of the passengers in the high and relatively high dependence level groups mainly commuted, which proved the effectiveness of the public transportation dependence level identification model and questionnaire survey data.

[0098] Step 4:

[0099] Based on the multivariate ordered logit model, a sensitivity analysis of the factors affecting the level of public transportation dependence is conducted. The multiple categories of the dependent variable are divided into multiple binary logistic regressions in turn, and the assumption is that the independent variable coefficients of the multiple binary logistic regression functions after the split are equal but the constant terms are unequal. Among them, the utility function U of the public transportation traveler's dependence level k on the traveler i under the transportation mode choice t is ikt The expression is as follows:

[0100]

[0101] Where, X ikt x is the factor affecting the level of public transportation dependence i , i=1,2,...,N,k=1,2,3,4,t is a 0-1 binary variable, i.e., whether or not to choose public transportation; is the regression parameter vector corresponding to the factors affecting the level of public transportation dependence; ε represents the random error terms that are independent and identically distributed, ε ikt It is the sum of the effects of other factors that are difficult to observe but have an impact on the level of public transportation travelers’ reliance on public transportation. ikt is the utility function of public transport traveler dependence level k for travel individual i under transportation mode choice t.

[0102] When ε follows a Gumbel distribution, the Logit model is transformed into an OrderedLogit regression model.

[0103] The probability L of the public transportation dependence level of travel individual i ij The calculation formula is:

[0104]

[0105] Where, L ij is the probability of the public transportation dependence level of travel individual i, β is the regression parameter vector corresponding to the factors affecting the public transportation dependence level, and X is the independent variable that affects the traveler's dependence on public transportation.

[0106] Based on the complementary double logarithmic function as the link function of the multivariate OrderedLogit regression model;

[0107] The sensitivity of the changes in the variables of the influencing factors related to the level of public transportation travelers' dependence on their dependence level was analyzed, and the odds ratio index was used for comparative analysis, which means that the probability of the event occurring in the experimental group p i The probability of the event occurring in the control group p j The ratio is calculated as:

[0108]

[0109] In the formula, OR i is the odds ratio, P i is the probability of the event occurring in the experimental group, P j The probability of an event in the control group, β i are regression parameter variables corresponding to the influencing factors of public transportation dependence level.

[0110] Based on the multivariate correlation analysis method, the key influencing factors of public transportation dependence level were screened, namely the total time from home and destination to the station, income, car availability and public transportation accessibility were used as the independent variables of the multivariate ordered logit regression model, and the public transportation dependence level with 4 levels was the dependent variable of the model. The initial confidence level of the model was taken as 90%, that is, when the significance parameter was less than 0.100, the public transportation dependence level group had a significant sensitivity effect on the independent variable. The model data set was input into the multivariate ordered logit regression model based on complementary double logarithm, and the final result was obtained after multiple iterative calculations.

[0111] Step 5:

[0112] After multiple iterations, the final results are shown in Table 1:

[0113] The study used the parallel line test method to verify the model assumptions, that is, the coefficients of the three binary logistic regression independent variables were equal. The results showed that the model test significance sig. = 0.047 was close to 0.05, and the assumption was established, indicating that the model construction was effective. In addition, the model fitting results showed that χ 2 =72.091, df=15, sig.=0.000, and the goodness of fit is 0.39, which is greater than the model accuracy acceptability of 0.20, indicating that the final model has a good fitting effect;

[0114] Table 1 - Multivariate OrderedLogit Model Results

[0115]

[0116] From the results in Table 1, we can see that the variables related to the four key influencing factors of public transportation dependence all show significant sensitivity characteristics. Among them, the public transportation dependence behavior of travelers is significantly sensitive to the total time variable from home and destination to transportation stations. When other policy variables are kept fixed, for every 1 minute reduction in the total time from home and destination to transportation stations, the possibility of the public transportation traveler's dependence level increasing by one level is 1.031 times. This also reflects that the time spent on the "last mile" needs to be significantly reduced through methods such as line station optimization in order to further improve the level of public transportation traveler dependence.

[0117] In addition, car availability, income, and public transportation accessibility are multi-classification variables and are split into multiple dummy variables in the model calculation, among which:

[0118] (1) The car availability variable was split into four dummy variables, and travelers' public transportation dependence behavior was significantly sensitive to the two dummy variables of low and high car availability. When other policy variables were kept fixed, the probability of increasing public transportation dependence in the low and high car availability groups was 2.199 times and 1.547 times that of the high car availability group, respectively. The results showed that the lower the car availability in Beijing, the higher the probability and sensitivity of increasing public transportation dependence.

[0119] (2) The income variable was split into six dummy variables. The first two categories of low-income people were mainly undergraduate and graduate students, and did not show a sensitive effect on their public transportation dependence. The probability of an increase in public transportation dependence among the third category of low-income workers was 2.071 times that of the high-income reference group, which means that the probability and sensitivity of an increase in public transportation dependence among low-income groups were higher than those among high-income groups.

[0120] (3) Public transportation accessibility is split into four dummy variables, all of which have significant impact sensitivity.

[0121] The probability of increasing the level of public transportation dependence of the four groups with low to high public transportation accessibility is 0.274, 0.322, 0.462 and 0.592 times that of the reference group, respectively. The results show that public transportation accessibility and public transportation dependence are positively correlated, which is consistent with the results of Section 4.2.3. However, the lower the public transportation accessibility, the more difficult it is for the group to improve its public transportation dependence level. The reason may be that the relevant travel groups mostly belong to the level of low public transportation dependence, which is significantly affected by their own travel psychological inertia and reduces their sensitivity to this variable. That is, if travelers are in a state of low public transportation accessibility for a long time, they will form a fixed negative perception psychology.

[0122] In summary, the advantages of the present invention are:

[0123] By combining travel behavior indicators with individual attribute indicators, a multi-dimensional dependency level measurement system is constructed. This measurement system can more comprehensively reflect the level of passengers' public transportation dependence and avoid the problem that a single dimension cannot fully reflect the characteristics of passengers' dependence.

[0124] The two-step clustering algorithm is used to cluster and identify passengers' public transportation dependence levels. It can process both categorical data and continuous data and automatically determine the optimal number of clusters. This method can efficiently identify different dependence levels of passenger groups and avoid the complexity and inaccuracy of traditional clustering methods in data preprocessing and cluster number selection.

[0125] Through the ordered logit regression model, the sensitivity analysis of the factors affecting the public transportation dependence level can be carried out, which can accurately identify and quantify the impact of different factors on the dependence level. This model takes into account the multi-level structure of the dependent variable, can effectively handle the dependence level data with multi-classification characteristics, and avoids the simplified assumptions of the binary classification model.

[0126] Quantitative analysis using odds ratios can clearly show the impact of each key factor on the level of passengers' public transportation dependence. Especially when analyzing the relationship between independent variables and dependent variables, odds ratios provide intuitive explanations, which facilitates policymakers and transportation planners to make decisions.

[0127] Through accurate identification and sensitivity analysis of passenger dependence levels, valuable insights can be provided to managers of public transportation systems to help them identify key factors affecting public transportation dependence, thereby making more precise adjustments and optimizations in policy design, transportation resource allocation, and personalized services.

[0128] This method has good scalability and can adapt to different cities and different types of public transportation systems. By collecting more data and further adjusting model parameters, it can achieve accurate analysis of different groups and different transportation needs, and help local transportation management departments formulate targeted policies.

[0129] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for identifying the level of passenger public transportation dependence and sensitivity analysis of influencing factors, characterized in that: include: Obtain passengers' public transportation travel behavior data and individual attribute data; Based on the data, construct public transportation dependency level measurement indicators, the measurement indicators include travel behavior indicators and individual attribute indicators, forming a multi-dimensional dependency level measurement system; A two-step clustering algorithm is used to process categorical and continuous data, automatically determine the optimal number of clusters, and perform cluster identification on passengers’ public transportation dependence levels; Based on the sensitivity analysis of the Ordered Logit regression model, the key factors affecting the level of passengers' public transportation dependence are evaluated and identified; The influence of the odds ratio independent variables was quantitatively analyzed to determine the sensitivity of each key factor to the level of passengers' public transportation dependence.

2. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 1 is characterized in that: The acquisition of the public transportation travel behavior data and individual attribute data of the passengers specifically includes: Design a questionnaire covering public transportation travel behavior and individual attributes, including travel behavior data, individual attribute data, and transportation policy data; Through stratified sampling, sample groups were selected according to research objectives and data requirements, and questionnaire surveys were conducted; the collected questionnaire data were preliminarily screened to eliminate invalid data; Clean the collected data and delete duplicate, erroneous and invalid records; In the questionnaire, the passengers' smart card numbers are collected, and the travel behavior data is associated with the card swiping transaction records. By matching the data with the card swiping records in the public transportation system background, the detailed travel data of each passenger is obtained; Combine all questionnaires and smart card data to build a complete travel dataset; Standardize individual attributes and travel behavior data; The collected valid data are organized and stored in the database.

3. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 2 is characterized in that: Based on the data, a public transportation dependency level measurement index is constructed, and the measurement index includes a travel behavior index and an individual attribute index, and a multi-dimensional dependency level measurement system is formed, which specifically includes: Define the measurement objectives of public transport dependence level, including travel behavior indicators and individual attribute indicators; The travel behavior indicators include: proportion of travel days, average number of daily trips, and round trip nature of travel modes; The individual attribute indicators include: age, education level, income, and car availability; Obtain the travel behavior data and individual attribute data required by the questionnaire and traffic data; By using the weighted summation method, the travel behavior and individual attribute indicators are combined into a public transportation dependence level value, where the weighted summation formula is: Where E is the dependency level, w i is the weight of the ith indicator, x i is the value of the ith indicator.

4. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 3 is characterized in that: The two-step clustering algorithm is used to process categorical data and continuous data, automatically determine the optimal number of clusters, and perform clustering identification on the public transportation dependence level of passengers, specifically including: Based on the Two-step algorithm, a recognition model for public transportation dependence level is established to automatically determine the optimal number of clusters; The Two-step algorithm clustering process includes: Read the data points in the data set one by one, generate a CF tree, remove outliers on the CF tree, and pre-cluster the data points in the dense area to form subclusters. When pre-clustering, process mixed data and perform hierarchical clustering based on the log-likelihood distance. The log-likelihood distance between two clusters i and j is: d(i,j)=ζ i +g j -g ＜i,j＞ In the formula, ζ i and j are the likelihood function values ​​of clusters i and j, respectively. The index 〈i,j〉 represents the new cluster generated by combining clusters i and j. K1 is the total number of continuous variables, K2 is the total number of categorical variables, and L k is the number of k-th categorical variables, N S is the total amount of data in cluster S, N skl The categorical variable k in the S cluster is divided into l groups, is an estimate of the variance of the continuous variable k in the entire data set, and is the variance estimate of the continuous variable k in cluster S, is the estimated mean of the kth continuous variable in cluster s. By inputting the sub-clusters of the final CF leaf elements in the pre-clustering stage, the clusters with the closest distance are recursively merged using the idea of ​​agglomeration until the optimal number of clusters is obtained. The optimal number of clusters J * The Bayesian information criterion is combined with the ratio of the nearest cluster distance r(J) to determine the calculation formula as follows: Among them, J1 and J2 are the number of clusters corresponding to the two largest subsets in the r(J) set, J * is the optimal number of clusters; The BIC criterion is selected accurately, and the goodness of fit is compared for a series of clustering results with different numbers of clusters generated by combining leaf nodes based on the agglomerative hierarchical clustering method, so as to automatically determine the best clustering scheme.

5. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 4 is characterized in that: The BIC guidelines specifically include: Among them, for cluster J, the calculation formula of the BIC criterion is as follows: Where N is the number of public transportation travel dependence level measurement indicator data sets, is the maximum value of the likelihood function, m J is the number of clustering model parameters, L k is the number of k-th categorical variables.

6. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 5 is characterized by: The categorical variables and continuous variable data were input into the identification model in sequence. Based on the existing literature results and the iterative operation results after multiple model parameter adjustments, the maximum number of branches of the model CF tree was set to 8, and the maximum tree depth was set to 3. The optimal number of classifications for public transportation dependence levels was determined according to the BIC information criterion. Finally, the groups were divided into 4 categories according to the public transportation travel dependence level indicators. The clustering silhouette coefficient value was 0.8, and the proportions of the groups with high, relatively high, relatively low and low public transportation travel dependence levels were 50%, 17%, 15% and 18% respectively. Among them, more than two-thirds of the passengers in the high and relatively high dependence level groups mainly commuted, which proved the effectiveness of the public transportation dependence level identification model and questionnaire survey data.

7. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 6 is characterized in that: The sensitivity analysis based on the OrderedLogit regression model evaluates and identifies the key factors that affect the level of passengers' public transportation dependence, including: Based on the multivariate OrderedLogit model, a sensitivity analysis of the factors affecting the level of public transportation dependence is conducted. The multiple categories of the dependent variable are divided into multiple binary logistic regressions in turn, and the assumption is that the independent variable coefficients of the multiple binary logistic regression functions after the split are equal but the constant terms are unequal. Among them, the utility function U of the public transportation traveler's dependence level k on the traveler i under the transportation mode choice t is ikt The expression is as follows: Where, X ikt x is the factor affecting the level of public transportation dependence i , i=1,2,...,N,k=1,2,3,4,t is a 0-1 binary variable, i.e., whether or not to choose public transportation; is the regression parameter vector corresponding to the factors affecting the level of public transportation dependence; ε represents the random error terms that are independent and identically distributed, ε ikt It is the sum of the effects of other factors that are difficult to observe but have an impact on the level of public transportation travelers’ reliance on public transportation. ikt is the utility function of public transport traveler dependence level k for travel individual i under transportation mode choice t; When ε follows a Gumbel distribution, the Logit model is transformed into an OrderedLogit regression model.

8. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 7 is characterized by: The probability L of the public transportation dependence level of travel individual i ij The calculation formula is: Where, L ij is the probability of the public transportation dependence level of travel individual i, β is the regression parameter vector corresponding to the factors affecting the public transportation dependence level, and X is the independent variable that affects the traveler's dependence on public transportation.

9. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 8 is characterized by: The complementary double logarithmic function is used as the link function of the multivariate ordered logit regression model; The sensitivity of the changes in the variables of the influencing factors related to the level of public transportation travelers’ dependence on their dependence level was analyzed, and the odds ratio indicator was used for comparative analysis, which means the probability of the event occurring in the experimental group p i The probability of the event occurring in the control group p j The ratio is calculated as: In the formula, OR i is the odds ratio, P i is the probability of the event occurring in the experimental group, P j The probability of an event in the control group, β i are regression parameter variables corresponding to the influencing factors of public transportation dependence level.

10. The method for identifying the level of passenger public transportation dependence and analyzing the sensitivity of influencing factors according to claim 9, characterized in that: Based on the multivariate correlation analysis method, the key influencing factors of public transportation dependence level were screened, namely the total time from home and destination to the station, income, car availability and public transportation accessibility were used as the independent variables of the multivariate ordered logit regression model, and the public transportation dependence level with 4 levels was the dependent variable of the model. The initial confidence level of the model was taken as 90%, that is, when the significance parameter was less than 0.100, the public transportation dependence level group had a significant sensitivity effect on the independent variable. The model data set was input into the multivariate ordered logit regression model based on complementary double logarithm, and the final result was obtained after multiple iterative calculations.

Images