Resident mileage influence factor analysis method based on dual machine learning model

Abstract

The invention belongs to the technical field of data analysis, and particularly relates to a resident mileage influence factor analysis method based on a dual machine learning model, which comprises the following steps: extracting built environment data including population density, distance to downtown center, intersection density, bus station number and land utilization entropy based on TAZ analysis; obtaining travel data of residents, and performing preprocessing; constructing a model for evaluating the vehicle ownership quantity and the influence of the vehicle ownership quantity on the daily mileage of the resident based on the built-up environment characteristics; a five-fold cross validation method is adopted to verify an evaluation model, model parameters are adjusted, the defects in the prior art are overcome, an innovative double-machine learning method is adopted, the influence of the regularization deviation problem is remarkably avoided, and the risk of model mismatching is greatly reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular relates to a method for analyzing factors affecting residents' mileage based on a dual machine learning model. Background Art

[0002] The increasing popularity of motor vehicle use has significantly exacerbated traffic congestion, air pollution and other problems in urban areas around the world, which has attracted widespread attention from relevant scholars. As part of solving the problem of the number of motor vehicles, the relationship between the built environment and motor vehicle ownership has always been a hot area of ​​research. In addition, understanding the factors that affect the daily mileage of vehicles (Vehicle Kilometers Traveled, VKT) is crucial for formulating effective transportation and urban planning policies.

[0003] At present, existing studies have confirmed that built environment factors affect motor vehicle ownership and frequency of use. The built environment of the place of residence usually involves residential density, accessibility of public facilities, diversity of land use, etc. These factors not only affect residents' daily travel choices, but may also have an important impact on motor vehicle ownership by changing residents' travel needs and patterns. Similarly, the built environment of the workplace, such as public transportation accessibility and work density, also has a significant impact on commuting behavior. By considering the built environment of the place of residence and the place of work at the same time, we can fully understand the compound impact of the built environment on daily mileage. In addition, as a mediating variable, motor vehicle ownership plays a key role in the impact of the built environment of the place of residence and the place of work on daily mileage.

[0004] In order to explore this mediating effect more deeply and reveal a more detailed impact mechanism, this paper adopts a double machine learning model (DML). This method can separate the direct impact of the built environment from the mediating role of motor vehicle ownership, and more accurately estimate the causal relationship by considering the nonlinear associations and interactions between variables. Summary of the invention

[0005] The purpose of the present invention is to provide a method for analyzing factors affecting residents' mileage based on a dual machine learning model, which overcomes the shortcomings of the prior art and significantly avoids the influence of the regularization bias problem by adopting an innovative dual machine learning method, and greatly reduces the risk of model mismatch.

[0006] To solve the above problems, the technical solutions adopted by the present invention are as follows:

[0007] The method for analyzing factors affecting residents' mileage based on the dual machine learning model includes the following steps:

[0008] S1. Extract built environment data including population density, distance to the city center, intersection density, number of bus stops, and land use entropy based on TAZ analysis;

[0009] S2, obtain residents’ travel data and perform pre-processing;

[0010] S3. Construct an assessment model for motor vehicle ownership and its impact on residents’ daily mileage based on built environment characteristics;

[0011] S4. Use the five-fold cross-validation method to verify the evaluation model and adjust the model parameters.

[0012] Furthermore, the land use entropy in S1 adopts the POI data provided by Amap, and the POI data is composed of data on transportation facilities, life services, shopping consumption, companies, science and education culture, medical care, hotel accommodation, leisure and entertainment, financial institutions, tourist attractions, sports and fitness, catering, motor vehicle related, and commercial residences. The land use entropy is the mixed entropy of the POI data, and its expression is:

[0013]

[0014] Where E is the mixed entropy, ranging from [0,1]. The higher the land use mixed degree, the greater the entropy value; n is the number of POI types in the grid; k is the i is the proportion of POI type i in the grid.

[0015] Furthermore, the travel data of the residents in S2 include age, income, population density, land use entropy, number of bus stops, population density, distance from residence to CBD, population density at work, land use entropy at work, number of bus stops at work, intersection density at work, distance from work to CBD, motor vehicle ownership, and driving distance.

[0016] Furthermore, the construction of the evaluation model of motor vehicle ownership and its impact on residents' daily driving mileage based on built environment characteristics specifically includes the following steps:

[0017] Let Z be the dependent variable, Q be the binary treatment variable, W = (W1,…,W p ) is a vector of independent variables, the model can be expressed as:

[0018] Z=Qα0+h1(W)+∈, E[∈|W, Q]=0

[0019] Q=n1(W)+ν,E[ν|W]=0

[0020] Among them, h1(W) and n1(W) are unknown functions, ò and ν are random errors, and α0 is the parameter of the variable of interest;

[0021] By using machine learning methods to predict h1(W), we get:

[0022]

[0023] The estimated value of α0 can be calculated by the following least squares estimation, which is expressed as:

[0024]

[0025] Estimated value The convergence speed is usually and The deviation converges faster than The convergence rate of right The influence of , the expression of estimation error is:

[0026]

[0027] In order to effectively deal with regularization errors and biases, the sample size N needs to be large enough to ensure the accuracy of the model; the optimal estimate of the model will be achieved based on the following formula:

[0028]

[0029] Combining the above models and derivation results, a comprehensive summary formula is obtained to describe the overall impact of mileage, which is expressed as follows:

[0030] Z=α1·Q+β1·h1(W)+γ2·n1(W)+δ+ò

[0031] Where: α1, β1 and γ2 are parameters, reflecting the influence of different factors on mileage; h1(W) and n1(W) are nonlinear functions, representing the relationship between the independent variable W and the motor vehicle ownership situation Q; δ is a constant term, representing the fixed influencing factors in the model; ò is an error term, including random factors not explained in the model.

[0032] Furthermore, the evaluation model is validated by using the five-fold cross validation method described in S4, and the model parameters are adjusted. The specific steps are as follows:

[0033] (1) Dataset partitioning: The original dataset is evenly divided into five subsets, usually randomly;

[0034] (2) Model training and testing: Select one of the subsets as the test set and the other four subsets as the training set, train the model and evaluate it on the test set, repeat five times, and ensure that each subset is used as the test set once;

[0035] (3) Performance measurement: For each test, various performance metrics, including accuracy, precision, recall, and F1 score, are used to evaluate the performance of the model on the test set;

[0036] (4) Summarize the results: Take the average of the performance measurement results of the five tests to obtain the final performance evaluation of the model.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention significantly avoids the impact of the regularization bias problem and greatly reduces the risk of model mismatch by adopting an innovative dual machine learning method.

[0039] When analyzing the impact of built environment factors on motor vehicle mileage, the present invention particularly considers the important mediating effect of motor vehicle ownership, thereby significantly improving the accuracy of the analysis.

[0040] The present invention provides strong support for managers to formulate effective measures to control and reduce the number of motor vehicles and mileage, helps to ease urban traffic pressure, and further promotes the sustainable development of cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The figure is a flowchart of the method for analyzing factors affecting residents’ mileage based on a dual machine learning model.

[0042] Figure 2 It is a division diagram of the study area in the embodiment. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] like Figure 1 As shown, the method for analyzing factors affecting residents' mileage based on the dual machine learning model of the present invention comprises the following steps:

[0045] Step 1: Select the research area

[0046] The research area selected by the present invention is the main urban area of ​​Changchun. The Changchun map crawled from Open Street Map is converted into an SHP file and then imported into ArcGIS. The main urban area (Luyuan District, Kuancheng District, Erdao District, etc.) is cut out and divided into traffic areas. Figure 2 shown.

[0047] Step 2: Extraction of built environment elements

[0048] Built environment data are systematically collected based on TAZ analysis, using a method to establish a 1-km radius buffer zone with the centroid of the transportation community as the core.

[0049] This paper uses five built environment attributes: population density, distance to the city center, intersection density, number of bus stops, and land use entropy, based on World Pop population data, Open Street Map road network, and Amap application programming interface (API), and calibrates them in residential areas and work areas respectively. The specific calculation method is as follows:

[0050] (1) Population density:

[0051] Density can be characterized by population density. The data comes from the World Pop project, which records the spatial distribution of Changchun residents in a grid with a resolution of 100 meters. The extracted population raster information is imported into ArcGIS, and the spatial distribution of Changchun residents is recorded in a grid with a resolution of 100 meters. The "Summary Statistics" tool is used to calculate the total population in each research traffic area, and then the population density is calculated.

[0052] (2) Distance to city center:

[0053] The latitude and longitude of the residents' starting point and the city center are obtained through Amap, and the path distance between the two places is obtained through the obtained latitude and longitude to analyze the accessibility between different areas

[0054] (3) Intersection density:

[0055] Intersection density is measured by the number of four-way intersections within the buffer zone, which represents the street network design and is obtained using road network data provided by Open Street Map.

[0056] (4) Number of bus stops:

[0057] The number of bus stops is characterized by the bus stop data provided by Amap to evaluate the convenience of the public transportation system in the region. Area of ​​Interest (AOI) data, bus routes and station data are obtained through the API provided by Amap. The obtained AOI data is filtered and reclassified through the ArcGIS platform, and then the number of bus stops in each research grid is counted using the “Summary Statistics Tool” in ArcGIS.

[0058] (5) Land use entropy:

[0059] Land use entropy is calculated from POI data provided by Amap. POI data is divided into 14 main categories (transportation facilities, life services, shopping and consumption, companies and enterprises, science and education and culture, medical care, hotel accommodation, leisure and entertainment, financial institutions, tourist attractions, sports and fitness, catering and food, motor vehicle related, commercial and residential), with about 132,273 records. Land use entropy is calculated by mixed entropy of 14 categories of POI. The entropy calculation formula is as follows:

[0060]

[0061] Where E is the mixed entropy, ranging from [0,1]. The higher the land use mixed degree, the greater the entropy value; n is the number of POI types in the grid; k is the i is the proportion of POI type i in the grid.

[0062] Step 3: Preprocessing of individual residents and travel data

[0063] The individual residents and travel data in this step come from the 2012 Changchun Family Travel Survey, which uses face-to-face interviews to record 24-hour travel information of 51,909 members of 20,000 households in five districts of Changchun, with a sampling rate of 0.68%. The obtained data is processed using Python software.

[0064] The survey data is divided into three archive data sets: family archive, personal archive and travel archive, covering the travel and socioeconomic information of all members of the surveyed family. The 14 key fields required by the present invention are extracted from the original data, namely: age, income, population density, land use entropy, number of bus stops, population density, distance from residence to CBD, population density at work, land use entropy at work, number of bus stops at work, intersection density at work, distance from work to CBD, motor vehicle ownership, and driving distance. After error checking, outlier and missing value cleaning of the survey data, 6487 individual motor vehicle travel records were finally used for analysis.

[0065] Step 4: Construct an assessment model for motor vehicle ownership and its impact on residents’ daily mileage based on built environment characteristics.

[0066] The present invention adopts a new method, namely dual machine learning, which combines statistical inference in a data-driven way to solve the problem of affecting vehicle mileage. Since the machine learning method does not require pre-defined relationships, the method can capture the nonlinear relationship between independent variables, vehicle ownership, and mileage.

[0067] Assume that Z is the dependent variable (i.e., mileage), Q is the binary treatment variable (i.e., whether or not a person owns a motor vehicle), and W = (W1, ..., W p ) is a vector of independent variables (such as income, gender, distance to the city center, etc.). The model can be expressed as:

[0068] Z=Qα0+h1(W)+∈, E[∈|W, Q]=0

[0069] Q=n1(W)+ν,E[ν|W]=0

[0070] Where h1(W) and n1(W) are unknown functions, ò and ν are random errors, and α0 is the parameter of the variable of interest. Formula (1) describes the mediating effect (i.e., the impact of motor vehicle ownership on individual mileage), while formula (2) defines the dependency of the intermediate variable (motor vehicle ownership) on the independent variable.

[0071] By using machine learning methods to predict h1(W), we can get:

[0072] Then, the estimated value of α0 can be calculated by the following least squares estimation as shown below:

[0073]

[0074] Estimated value The convergence speed is usually and The deviation converges faster than In order to understand the convergence rate of right The estimated error can be expressed as:

[0075]

[0076] In order to effectively deal with regularization errors and biases, the sample size N needs to be large enough to ensure the accuracy of the model. Ultimately, the optimal estimate of the model will be achieved based on the following formula:

[0077]

[0078] Combining the above models and derived results, we can derive a comprehensive summary formula to describe the overall impact of mileage:

[0079] Z=α1·Q+β1·h1(W)+γ2·n1(W)+δ+ò

[0080] Where: α1, β1 and γ2 are parameters, reflecting the influence of different factors on mileage; h1(W) and n1(W) are nonlinear functions, representing the relationship between the independent variable W and the motor vehicle ownership situation Q; δ is a constant term, representing the fixed influencing factors in the model; ò is an error term, including random factors not explained in the model.

[0081] Step 5: Optimize model performance

[0082] In order to further optimize the prediction performance of the model and effectively avoid overfitting, the present invention adopts a five-fold cross-validation method to adjust the model parameters. The specific steps are as follows:

[0083] (1) Dataset partitioning: The original dataset is evenly divided into five subsets, usually randomly.

[0084] (2) Model training and testing: Select one of the subsets as the test set and the other four subsets as the training set, train the model and evaluate it on the test set. This process is repeated five times to ensure that each subset is used as a test set once.

[0085] (3) Performance measurement: For each test, various performance metrics such as accuracy, precision, recall, F1 score, etc. can be used to evaluate the performance of the model on the test set.

[0086] (4) Summarize the results: Take the average of the performance measurement results of the five tests to obtain the final performance evaluation of the model.

[0087] In the process of parameter optimization, key parameters such as learning rate, maximum depth of tree and minimum number of samples required for node partitioning were mainly adjusted. Through random search technology, the performance of the model under different parameter configurations was comprehensively compared, and the optimal parameter combination was used to improve the generalization ability and prediction accuracy of the model on the independent test set.

[0088] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for analyzing factors affecting residents’ mileage based on a dual machine learning model, characterized by: The following steps are involved: S1. Extract built environment data including population density, distance to the city center, intersection density, number of bus stops, and land use entropy based on TAZ analysis; S2, obtain residents’ travel data and perform pre-processing; S3. Construct an assessment model for motor vehicle ownership and its impact on residents’ daily mileage based on built environment characteristics; S4. Use the five-fold cross-validation method to verify the evaluation model and adjust the model parameters.

2. The method for analyzing factors affecting mileage of residents based on a dual machine learning model according to claim 1, characterized in that: The land use entropy in S1 adopts the POI data provided by Amap. The POI data consists of data on transportation facilities, life services, shopping, companies, science and education, health care, hotel accommodation, leisure and entertainment, financial institutions, tourist attractions, sports and fitness, catering, motor vehicle-related, and commercial residences. The land use entropy is the mixed entropy of the POI data, and its expression is: Where E is the mixed entropy, ranging from [0,1]. The higher the land use mixed degree, the greater the entropy value; n is the number of POI types in the grid; k is the i is the proportion of POI type i in the grid.

3. The method for analyzing factors affecting mileage of residents based on a dual machine learning model according to claim 1, characterized in that: The travel data of residents described in S2 include age, income, population density, land use entropy, number of bus stops, population density, distance from residence to CBD, population density at work, land use entropy at work, number of bus stops at work, intersection density at work, distance from work to CBD, motor vehicle ownership, and driving distance.

4. The method for analyzing factors affecting mileage of residents based on a dual machine learning model according to claim 1, characterized in that: The specific steps of constructing the evaluation model of motor vehicle ownership and its impact on residents' daily driving mileage based on built environment characteristics include: Let Z be the dependent variable, Q be the binary treatment variable, W = (W1,...,W p ) is a vector of independent variables, the model can be expressed as: Z=Qα0+h1(W)+∈,E[∈|W,Q]=0. Q=n1(W)+ν,E[ν|W]=0 Among them, h1(W) and n1(W) are unknown functions, ò and ν are random errors, and α0 is the parameter of the variable of interest; By using machine learning methods to predict h1(W), we get: The estimated value of α0 can be calculated by the following least squares estimation, which is expressed as: Estimated value The convergence speed is usually and The deviation converges faster than The convergence rate of right The influence of , the expression of estimation error is: In order to effectively deal with regularization errors and biases, the sample size N needs to be large enough to ensure the accuracy of the model; the optimal estimate of the model will be achieved based on the following formula: Combining the above models and derivation results, a comprehensive summary formula is obtained to describe the overall impact of mileage, which is expressed as follows: Z=α1·Q+β1·h1(W)+γ2·n1(W)+δ+ò Where: α1, β1 and γ2 are parameters, reflecting the influence of different factors on mileage; h1(W) and n1(W) are nonlinear functions, representing the relationship between the independent variable W and the motor vehicle ownership situation Q; δ is a constant term, representing the fixed influencing factors in the model; ò is an error term, including random factors not explained in the model.

5. The method for analyzing factors affecting mileage of residents based on a dual machine learning model according to claim 1, characterized in that: As described in S4, the evaluation model was validated using the five-fold cross validation method and the model parameters were adjusted. The specific steps were as follows: (1) Dataset partitioning: The original dataset is evenly divided into five subsets, usually randomly; (2) Model training and testing: Select one of the subsets as the test set and the other four subsets as the training set, train the model and evaluate it on the test set, repeat five times, and ensure that each subset is used as the test set once; (3) Performance measurement: For each test, various performance metrics, including accuracy, precision, recall, and F1 score, are used to evaluate the performance of the model on the test set; (4) Summarize the results: Take the average of the performance measurement results of the five tests to obtain the final performance evaluation of the model.

Images