A method for predicting online car-hailing travel demand based on land use mix

Abstract

The present invention discloses a method for predicting online car-hailing travel demand based on land use mixture, which belongs to the technical field of traffic information processing. The method comprises the following steps: obtaining multi-source geographic spatial data and online car-hailing order data, and pre-processing the online car-hailing order data; establishing grid units for a study area, calculating grid areas, and counting the number of online car-hailing orders in different time periods in each grid, and obtaining built environment indicators based on multi-source geographic spatial data, wherein the built environment indicators include density indicators and land use mixture; establishing a multi-scale spatiotemporal geographic weighted regression model by taking the built environment indicators as independent variables and the online car-hailing order data as dependent variables; calibrating the model by using a reverse fitting algorithm, and predicting the online car-hailing travel demand by using the calibrated model.

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic information processing, and specifically relates to a method for predicting online car-hailing travel demand based on land use mixture. Background Art

[0002] As an important part of shared transportation, understanding the factors affecting online ride-hailing travel can clarify the heterogeneity of the built environment on the spatial distribution of online ride-hailing travel in cities, rationally plan the spatial layout of online ride-hailing travel, shorten the waiting time for passengers, and alleviate urban traffic congestion. However, the rapid growth of the online ride-hailing market has also caused some negative effects, such as high operating costs, low efficiency, and long idle time. Therefore, understanding the influencing mechanism of online ride-hailing travel can better analyze travel patterns and provide support for optimizing layout.

[0003] Built environment refers to all man-made environments that can be intervened, built and transformed by humans. It is also a three-dimensional spatial combination that affects online ride-hailing travel. The relationship between built environment and online ride-hailing travel behavior is mainly reflected in density (building density, subway station density, etc.), public transportation proximity, land use mix, and social demographic characteristics. Related studies all use the type of POI inside the grid as an indicator for calculating land use mix. However, the perception of the surrounding three-dimensional space by passengers in the analysis unit is continuous rather than discrete. The diversity obtained by the previous method cannot well reflect the continuity of passengers' actual spatial perception. Secondly, in the existing research on the method of constructing analysis models, the impact of the built environment is mainly analyzed by regression coefficients. Among them, the geographically weighted regression model is mainly used because it considers the local regression effect and can solve the problem of spatial non-stationarity and is widely used. However, the geographically weighted regression and spatiotemporal geographically weighted regression models ignore the scale effect of bandwidth, which may lead to biased analysis results. Therefore, the subsequent research on the built environment and online ride-hailing travel distribution cannot accurately reflect the impact of the built environment on the distribution of online ride-hailing travel, and thus cannot make reasonable optimization of the spatial layout of online ride-hailing. Summary of the invention

[0004] In view of the above-mentioned problems existing in the prior art, the present invention proposes a method for predicting online car-hailing travel demand based on land use mixture, which is reasonably designed, solves the shortcomings of the prior art and has good results.

[0005] A method for predicting online car-hailing travel demand based on land use mixture degree includes the following steps:

[0006] S1. Obtain multi-source geospatial data and online ride-hailing order data;

[0007] S2. Preprocessing the online car-hailing order data;

[0008] S3. Establish grid cells for the study area, calculate the grid area, count the number of online ride-hailing orders in each grid at different time periods, and obtain built environment indicators based on multi-source geospatial data. Built environment indicators include density indicators and land use mix;

[0009] S4. Using built environment indicators as independent variables and online ride-hailing order data as dependent variables, a multi-scale spatiotemporal geographic weighted regression model was established;

[0010] S5. Use the reverse fitting algorithm to calibrate the model, and use the calibrated model to predict the travel demand for online ride-hailing vehicles.

[0011] Furthermore, the multi-source geospatial data includes city POI data, road network data, and Worldpop population data, and the online car-hailing order data includes order starting point time and location information.

[0012] Furthermore, the S2 comprises the following sub-steps:

[0013] S201: Eliminate order data with a blank start time;

[0014] S202: Eliminate data with a running time of less than 1 minute;

[0015] S203: Eliminate order data that is not in the research area.

[0016] Furthermore, in S3, the density indicators include population density, road network density, transportation facility density, residential area density, public management and service density, commercial service density, green space and open space density and company enterprise density.

[0017] Furthermore, in S3, the urban POI data is reclassified into residential areas, government agencies, scientific research, culture, education, sports and leisure, medical care, shopping, catering, business and finance, companies, bus stops, transportation hubs, parks, and squares. A buffer zone is created with the center of each grid as the origin. The radius of the buffer zone is not less than the length of the grid side. The number of POI types falling in the buffer zone is counted and brought into the information entropy to calculate the land use mixture degree. The expression is:

[0018] (1);

[0019] in, For the The proportion of POI types in the grid, is the total number of POI types in the grid.

[0020] Furthermore, the built environment index is used as the independent variable and the online car-hailing order data is used as the dependent variable to establish a multi-scale spatiotemporal geographic weighted regression model, which is expressed as follows:

[0021] (2);

[0022] in, is the dependent variable, i.e., the online car-hailing travel demand data. Indicates the first The estimated coefficients of the independent variables; is the intercept, For the The coordinates of the sampling points, For the The spatial position coordinates of the sampling points, For sampling point The time coordinates, For the independent variables, namely the built environment indicators, , is the random error term;

[0023] Estimation in the model The form is as follows:

[0024] (3);

[0025] in, is the weight matrix, is the explanatory variable;

[0026] Find the weight matrix When , the kernel function used is the Gaussian function, and the expression is:

[0027] (4);

[0028] In the formula, for The weight of For the Sampling point to The space-time distance of the regression points, is the space-time bandwidth;

[0029] The optimal bandwidth is selected using the AICc criterion and the Akaike information criterion correction value is a measure of the goodness of model fit. The lower the value, the better the model fit, expressed as:

[0030] (5);

[0031] In the formula, is the estimated standard deviation of the error term, is the hat matrix traces, is the sample size; when When it is minimum, it is the optimal bandwidth.

[0032] Furthermore, the reverse fitting algorithm is used for model calibration. The multi-scale spatiotemporal geographically weighted regression model is essentially a generalized additive model, so the reverse fitting algorithm that maximizes the log-likelihood is used for calibration. The GAM form of the multi-scale spatiotemporal geographically weighted regression model is;

[0033] (6);

[0034] In the formula, is the response variable, It is The smooth function of the explanatory variables, is the error term, Total number of explanatory variables;

[0035] The model calibration process consists of the following sub-steps:

[0036] S51. For all addition terms Set the initial value to , then the corresponding error term is ;

[0037] S52, for initialization and , and Combination Perform MGTWR model regression analysis and calculate the corresponding space-time bandwidth , thus Update, recorded as ;

[0038] S53, according to Calculate the corresponding , repeat step S52 to obtain the corresponding space-time bandwidth , thus Update, recorded as ;

[0039] S53, repeat step S53, update all variables corresponding to and , thus completing the first round of iteration;

[0040] S55, repeating the iterative process until a convergence criterion is reached;

[0041] The convergence criterion of MGTWR is: the iteration ends when the difference between consecutive iterations is small enough, and the change value is adopted To reflect the difference between the previous and next iterations, when The iteration ends when The expression is:

[0042] (7);

[0043] In the formula, For the +1 addition term in the iteration, For the The addition term in the round iteration, is the number of iterations.

[0044] Beneficial technical effects brought by the present invention:

[0045] The hybrid calculation idea of ​​land use proposed in this invention can well reflect the continuity of passengers' perception of surrounding objects within the analysis unit, and can provide data support for analyzing the relationship between the built environment and the distribution of online car-hailing travel. Secondly, the multi-scale spatiotemporal geographic weighted regression model can be used to explore the spatiotemporal heterogeneity of different variables on online car-hailing travel, which can further provide a scientific basis for analyzing the potential impact of the built environment, provide suggestions for optimizing the layout of online car-hailing, and promote the development of green cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a method for predicting online car-hailing travel demand based on land use mixture in the present invention.

[0047] Figure 2 This is a flow chart for calculating the land use mixture degree in the present invention.

[0048] Figure 3 Schematic diagram of POI distribution in calculating land use mixture.

[0049] Among them, (a) is a traditional POI distribution map; (b) is a POI distribution map in the present invention. DETAILED DESCRIPTION

[0050] The specific implementation of the present invention is further described below in conjunction with specific embodiments:

[0051] A method for predicting online car-hailing travel demand based on land use mix, such as Figure 1 As shown, the following steps are included:

[0052] S1: Obtain multi-source geospatial data and online ride-hailing order data;

[0053] S2. Preprocessing the online car-hailing order data;

[0054] S3. Establish grid cells for the study area, calculate the grid area and the number of online ride-hailing orders in each grid at different time periods, and obtain built environment indicators based on multi-source geospatial data. Built environment indicators include density indicators and land use mix;

[0055] S4. Using built environment indicators as independent variables and online ride-hailing order data as dependent variables, a multi-scale spatiotemporal geographic weighted regression model was established;

[0056] S5. The model is calibrated using a reverse fitting algorithm, and the calibrated model is used to predict the travel demand of online ride-hailing vehicles.

[0057] Specifically, in step S1, the multi-source geographic spatial data includes city POI data, road network data, and Worldpop population data, and the online car-hailing order data includes the order starting point time and location information.

[0058] The online car-hailing data in this embodiment comes from the Didi Chuxing data in Haikou in October 2017, as shown in Table 1; the city POI data comes from the Amap API (Application Program Interface), the city road network data comes from the open source map Open Street Map, and the population data comes from the open source website Worldpop;

[0059] Table 1 Original order data fields

[0060] .

[0061] Specifically, in step S2, in order to ensure the accuracy and completeness of the data, data cleaning methods such as missing value filtering and outlier removal are adopted. The online car-hailing order data preprocessing steps are:

[0062] S201: Eliminate order data with a blank start time;

[0063] S202: Eliminate data with a running time of less than 1 minute;

[0064] S203: Eliminate order data that is not in the research area.

[0065] After preprocessing, the amount of online ride-hailing order data is 1,710,637.

[0066] Specifically, in step S3, grid units are established for the study area, the grid area is calculated, and the number of online car-hailing orders in different time periods within each grid is calculated; density indicators include population density, road network density, transportation facility density, residential area density, public management and service density, commercial service density, green space and open space density, and company enterprise density.

[0067] In this embodiment, the main urban area of ​​Haikou City is selected as the research area, and grid units with a side length of 500m are created for each grid, and the area of ​​each grid is calculated. The data of online car-hailing orders in each grid are counted, and the average hourly travel volume distribution is obtained by aggregation, and then the research period is divided into 24 time periods. The density of built environment data in each grid is counted, including: transportation facility density, residential area density, public management and service density, commercial service density, green space and open space density and company enterprise density. The population density is obtained by aggregating the population data selected from Worldpop data to obtain the population density corresponding to the grid. In order to reduce the data error, the statistical yearbook data is used for correction. The road network density is obtained by taking the ratio of the road length in the grid to the grid area.

[0068] like Figure 2 As shown in the figure, the built environment indicators also include land use mixture. The urban POI data is reclassified into residential areas, government agencies, scientific research, culture, education, sports and leisure, medical care, shopping, catering, business and finance, companies, bus stops, transportation hubs, parks, and squares. A buffer zone is created with the center of the grid as the origin. The buffer zone radius is not less than the grid radius. The number of POI types falling in the buffer zone is counted and brought into the information entropy to calculate the land use mixture. The expression is:

[0069] (1);

[0070] in, For the The proportion of POI types in the grid, is the total number of POI types in the grid.

[0071] Previous studies have used the types of POIs within the grid as an indicator for calculating information entropy. However, the perception of the surrounding three-dimensional space by passengers within the analysis unit is continuous rather than discrete. The diversity obtained by previous methods cannot well reflect the continuity of passengers' perception of the actual space, such as Figure 3 As shown in (a), this embodiment adopts the method of setting the search distance or buffer zone (the search radius is set to 1000m when generating the neighbor table in ArcGIS 10.4 (the living circle radius is mostly 1000m)), and calculates the information entropy index of the POI in the buffer zone as the information entropy index of the central grid, as shown in Figure 3 as shown in (b).

[0072] The definition of community life circle is: a basic unit that meets the various needs of urban and rural residents throughout their life cycle, including work and life, within a suitable daily walking range, integrating multiple functions of "suitable for work, suitable for living, suitable for tourism, suitable for nursing, and suitable for learning", and leading a healthy and low-carbon lifestyle for the future. It is divided into two levels of community life circles: 15 minutes and 5-10 minutes. The 15-minute life circle is determined based on the characteristics of residents' life and travel and actual needs, and its radius is 1000m. Therefore, the search radius of this embodiment is set to 1000m.

[0073] Specifically, in step S4, the built environment indicators are used as independent variables and the online car-hailing order data are used as dependent variables to establish a multi-scale spatiotemporal geographic weighted regression model.

[0074] Specifically, in the spatiotemporal geographically weighted regression model (GTWR), the spatiotemporal bandwidth of each variable is the same, and the "average bandwidth" is used for all variables in the detection process of the local regression coefficient, which means that the impact of the explanatory variable on the dependent variable defaults to the same scale (bandwidth), while the actual explanatory variable has different impact scales on the response variable. The GTWR model may have limitations in dealing with spatiotemporal stationarity issues, causing regression results to be biased. Therefore, GTWR is extended to the multiscale spatiotemporal geographically weighted regression model (Multiscale GTWR, MGTWR), allowing each variable to have its own spatiotemporal bandwidth, which can reflect the actual impact scale of the explanatory variable. In MGTWR, bandwidth is the key to controlling the scale of operation. Small spatiotemporal bandwidths often reveal short-term patterns at the micro scale, while large spatiotemporal bandwidths often reveal long-term patterns at the macro scale. The form of MGTWR is:

[0075] (2);

[0076] in, is the dependent variable, i.e., the online car-hailing travel demand data. Indicates the first The estimated coefficients of the independent variables; is the intercept, For the The coordinates of the sampling points, For the The spatial position coordinates of the sampling points, For sampling point The time coordinates, For the independent variables, namely the built environment indicators, , is the random error term;

[0077] Estimation in the model The form is as follows:

[0078] (3);

[0079] in, is the weight matrix, is the explanatory variable;

[0080] When calculating the weight matrix W in the MGTWR model, the kernel function used is the Gaussian function, and the expression is:

[0081] (4);

[0082] In the formula, For the Sampling point to The space-time distance of the regression points, is the space-time bandwidth;

[0083] The optimal bandwidth is selected using the AICc criterion and the Akaike information criterion correction value is a measure of the goodness of model fit. The lower the value, the better the model fit, expressed as:

[0084] (5);

[0085] In the formula, is the estimated standard deviation of the error term, is the hat matrix traces, is the sample size; when When it is minimum, it is the optimal bandwidth.

[0086] For MGTWR, the value of the bandwidth varies with the number of independent variables. Therefore, the weighted least squares method is not suitable for the calibration of MGTWR. In essence, MGTWR is a generalized additive model (GAM), which operates under the same technical framework as the multiscale geographically weighted regression model. Therefore, the backfitting algorithm is the best choice for model calibration because it assumes that all other terms are known and calibrates the unknown terms separately.

[0087] The reverse fitting algorithm is used for model calibration. The MGTWR model is essentially a generalized additive model, so the reverse fitting algorithm that maximizes the log-likelihood is used for calibration. The GAM form of the MGTWR model is;

[0088] (6);

[0089] In the formula, is the response variable, It is The smooth function of the explanatory variables, is the error term, Total number of explanatory variables;

[0090] The model calibration process consists of the following sub-steps:

[0091] S51. For all addition terms Set the initial value to , then the corresponding error term is ;

[0092] S52, for initialization and , and Combination Perform MGTWR model regression analysis and calculate the corresponding space-time bandwidth , thus Update, recorded as ;

[0093] S53, according to Calculate the corresponding , repeat step S52 to obtain the corresponding space-time bandwidth , thus Update, recorded as ;

[0094] S53, repeat step S53, update all variables corresponding to and , thus completing the first round of iteration;

[0095] S55, repeating the iterative process until a convergence criterion is reached;

[0096] The convergence criterion of MGTWR is: the iteration ends when the difference between consecutive iterations is small enough, and the change value is adopted To reflect the difference between the previous and next iterations, when The iteration ends when The expression is:

[0097] (7);

[0098] In the formula, For the +1 addition term in the iteration, For the The addition term in the round iteration, is the number of iterations.

[0099] In order to evaluate the model performance gap between the online car-hailing travel demand forecasting method proposed in this paper and the traditional online car-hailing travel demand forecasting method, the , , AICc, and residual sum of squares (RSS) for comparison. and The higher the AICc and RSS, the better the model fit. On this basis, the mean absolute error (MAE) and root mean square error (RMSE) are calculated for further comparison. and , lower AICc and RSS, lower MAE and RMSE indicate better performance of the model.

[0100] (8);

[0101] (9);

[0102] In the formula, is the actual value of the impact of the built environment on online ride-hailing travel, is the predicted value of the model, is the number of built-in environment variables.

[0103] As shown in Table 2, according to different evaluation indicators, it can be seen that the calculation method in this example has higher model interpretability and better goodness of fit, and the MAE and RMSE are also lower.

[0104] Table 2 is a comparison between the method of the present invention and the traditional prediction method

[0105] .

[0106] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for predicting online car-hailing travel demand based on land use mix, characterized in that: The following steps are involved: S1. Obtain multi-source geospatial data and online ride-hailing order data; S2. Preprocessing the online car-hailing order data; S3. Establish grid cells for the study area, calculate the grid area, count the number of online ride-hailing orders in each grid at different time periods, and obtain built environment indicators based on multi-source geospatial data. Built environment indicators include density indicators and land use mix; S4. Using built environment indicators as independent variables and online ride-hailing order data as dependent variables, a multi-scale spatiotemporal geographic weighted regression model was established; S5. Use the reverse fitting algorithm to calibrate the model, and use the calibrated model to predict the travel demand for online ride-hailing vehicles; In S3, the urban POI data is reclassified into residential areas, government agencies, scientific research, culture, education, sports and leisure, medical care, shopping, catering, business and finance, companies, bus stops, transportation hubs, parks, and squares. A buffer zone is created with the center of each grid as the origin. The radius of the buffer zone is not less than the length of the grid side. The number of POI types falling in the buffer zone is counted and brought into the information entropy to calculate the land use mixture degree. The expression is: (1)； in, For the The proportion of POI types in the grid, is the total number of POI types in the grid; Taking the built environment index as the independent variable and the online car-hailing order data as the dependent variable, a multi-scale spatiotemporal geographic weighted regression model is established, and the expression is: (2)； in, is the dependent variable, i.e., the online car-hailing travel demand data. Indicates the first The estimated coefficients of the independent variables; is the intercept, For the The coordinates of the sampling points, For the The spatial position coordinates of the sampling points, For sampling point The time coordinates, For the independent variables, namely the built environment indicators, , is the random error term; Estimation in the model The form is as follows: (3)； in, is the weight matrix, is the explanatory variable; Find the weight matrix When , the kernel function used is the Gaussian function, and the expression is: (4)； In the formula, for The weight of For the Sampling point to The space-time distance of the regression points, is the space-time bandwidth; The optimal bandwidth is selected using the AICc criterion and the Akaike information criterion correction value is a measure of the goodness of model fit. The lower the value, the better the model fit, expressed as: (5)； In the formula, is the estimated standard deviation of the error term, is the hat matrix traces, is the sample size; when When it is minimum, it is the optimal bandwidth; The convergence criterion of MGTWR is: the iteration ends when the difference between consecutive iterations is small enough, and the change value is adopted To reflect the difference between the previous and next iterations, when The iteration ends when The expression is: (7)； In the formula, For the +1 addition term in the iteration, For the The addition term in the round iteration, is the number of iterations.

2. The method for predicting online car-hailing travel demand based on land use mix according to claim 1, characterized in that: The multi-source geographic spatial data includes city POI data, road network data, and Worldpop population data, and the online car-hailing order data includes order starting point time and location information.

3. The method for predicting online car-hailing travel demand based on land use mix according to claim 2, characterized in that: The S2 comprises the following sub-steps: S201: Eliminate order data with a blank start time; S202: Eliminate data with a running time of less than 1 minute; S203: Eliminate order data that is not in the research area.

4. The method for predicting online car-hailing travel demand based on land use mix according to claim 3, characterized in that: In S3, the density indicators include population density, road network density, transportation facility density, residential area density, public management and service density, commercial service density, green space and open space density and company enterprise density.

5. The method for predicting online car-hailing travel demand based on land use mix according to claim 4, characterized in that: The reverse fitting algorithm is used for model calibration. The multi-scale spatiotemporal geographically weighted regression model is essentially a generalized additive model. Therefore, the reverse fitting algorithm that maximizes the log-likelihood is used for calibration. The GAM form of the multi-scale spatiotemporal geographically weighted regression model is; (6)； In the formula, is the response variable, It is The smooth function of the explanatory variables, is the error term, Total number of explanatory variables; The model calibration process consists of the following sub-steps: S51. For all addition terms Set the initial value to , then the corresponding error term is ; S52, for initialization and , and Combination Perform MGTWR model regression analysis and calculate the corresponding space-time bandwidth , thus Update, recorded as ; S53, according to Calculate the corresponding , repeat step S52 to obtain the corresponding space-time bandwidth , thus Update, recorded as ; S53, repeat step S53, update all variables corresponding to and , thus completing the first round of iteration; S55, repeating the iterative process until a convergence criterion is reached; The convergence criterion of MGTWR is: the iteration ends when the difference between consecutive iterations is small enough, and the change value is adopted To reflect the difference between the previous and next iterations, when The iteration ends when The expression is: (7)； In the formula, For the +1 addition term in the iteration, For the The addition term in the round iteration, is the number of iterations.

Images