When conducting urban-level passenger flow demand forecasts for city transportation projects, the four-stage model can capture the change of passenger demand. Based on the four-stage method of changing passenger demands, this paper obtained the total amount generated by the travel demands of urban residents, the distribution of travel demands in each traffic zone, the probability of residents choosing different modes of transportation, and the way it distributes on the traffic network to achieve the propose of travel.
According to the existing data and the calculation range of Baoji, a carbon emission model based on passenger demanding forecast can be established for cities prior to opening rail transit. This differs from a traditional assessment by collecting transportation capacity, distance, and energy consumption. Based on the travel demand of urban passengers, this paper predicts the change of carbon emissions, which will be brought about by the adjustment of the proportion of use of each travel mode after the opening of rail transit in 2023. Here, it is predicted that the choice of transportation mode within traffic zones in 2023 has become the basis of carbon emission reduction calculation [
29]. According to the traditional four-stage method, it is not difficult to get the annual passenger flow demand for Baoji, for the planning year. However, according to the characteristics of individual travelers and their preference characteristics, accounting for the travel demands attracted by different modes of transportation will be an essential part of this paper.
3.1. The Basic Principles of the Mixed Logit Model based on the Choice of Residents’ Travel Mode
The Mixed Logit model has great physical significance, and with the improvement of computer speed and the appearance of the simulation algorithm, it has been widely used. The Mixed Logit model overcomes two critical flaws in the traditional Logit model [
30], the LRTV (Limitation of Random Taste Variation) and the IIA(Independent from Irrelevant Alternative)requirement can express personal preferences. It is a highly adaptable model. McFadden [
31] demonstrates that the Mixed Logit model can simulate any random utility model (including Logit, Probit Nested Logit, and so on), that is, it can be infinitely close to any random utility model when the mixed distribution function is chosen appropriately.
This paper designed and conducted the behavioral survey (Revealed Preference) and the intention survey (Stated Preference) of urban rail transportation choice from the perspective of the traveler. Based on the Mixed Logit, the key factors of transportation mode selection and their changes on the travel structure were studied. Data on residents and travel behavior were collected, and the travel time and cost factors were comprehensively considered and analyzed. The paper selected the travel-related attributes such as urban rail transit dependence, travel purpose, the age and other personal socio-economic characteristics as utility function feature variables, assuming that the arrival time, time spent on the vehicle and the cost as random variables and obey the log-normal distribution. Using questionnaires generated by D-efficient design method, an empirical survey of the preference of urban rail transit mode was carried out in Baoji.
As for the traveler, the mode choice model was established based on the Mixed Logit that reflects the preferences to individual heterogeneity, including setting parameters and selecting available data. When describing individual behavior selection, the utility function consists of two addable parts, one observable and the other random [
32,
33,
34,
35], expressed as by Equation (1).
where
q represents the individual who travels;
j is an item or branch in the selection set;
qj is the random item of the utility function;
qj is an observable utility, usually expressed as
qj=
Tqj; here,
qj is the observed parameter containing the characteristics of the subject
q and object
j. The vector β is the coefficient to be estimated. The research and development of discrete choice models was based on the assumptions and processing of random variables. The selection set of individuals in this paper are public bus, car (private owned), taxi (including online car-hailing), AND urban rail transit (metro), expressed by Equation (2).
The properties that affect the utility of the selection branch were determined. Six travel characteristic attributes were set, including travel purpose, waiting time, travel distance, arrival time (defined as the time spent to reach the main transportation from travel origins), the time spent on board, and fare. As the time onboard is highly correlated with the trip distance, the change in the average speed reflects the service level of the time spent onboard. It is also assumed that the traveler is also affected by socioeconomic attributes like gender, age, occupation, income, and car access, as well as the expected dependence on the rail transit. Therefore, the utility function of individual
q was determined to the selection branch
j, which is represented by Equation (3).
where
indicates the inherent constant of each item in the selection set;
represents the usage frequency of rail transit for individual
q, indicating expectations for rail transit;
represents the characteristic attribute of travel;
shows the socioeconomic attributes of the individual;
,
,
are the corresponding parameters for related attributes; and
is a utility random item.
In the classic Multinomial Logit model, is a fixed parameter, and is an independent variable and obeys the Gumbel distribution. The Mixed Logit model assumes that variable parameters and utility random items change between individuals and selection branches. Random factors were added to to introduce individual heterogeneity and to accommodate the correlation between the selection branches. This paper assumes that two kinds of attribute parameters, namely, travel consumption time and fare, are random variables subject to the lognormal distribution. To reflect the heterogeneity of individual in time value, the parameters of the time attribute were divided into four specific attribute parameters including bus, car, taxi, and rail transit—,,,. Similarly, the trip fare was also divided into those four categories of attribute parameters, namely, ,,,. The remaining attributes were fixed parameters.
Normally, the expected value of consumption time and expense parameters should be negative, that is, the increase in the above two properties’ values reduces the corresponding utility. The values of the lognormal distribution are in a non-negative interval, so the above attributes are applied to the symbolic change. The car is set to the comparison item; therefore, the determined item of each mode utility function can be expressed as Equations (4)–(7):
When a fixed parameter is replaced by a random number that obeys a distribution, the probability of
j being chosen without conditions should be the probability of
traversing all possible values, that is, the probability function of the Mixed Logit model can be regarded as the integration of Multinomial Logit probability functions on the
density function,
, as shown in Equation (8). For a more reliable estimate of the parameters, the survey data are encoded in
Table 1 as a Mixed Logit model.
3.2. Individual Trips Distance Calculation
Trips between different traffic zones are through the city road network, including bus routes, rail transit lines (planned for 2023), bus stops along the way, subway stations, taxi stops, and car parking stations. The travel path and distance depend on transportation mode. Through the network analysis module in GIS, this study marked the OD points of individual trips in different means and obtained the travel paths of varied transportation modes. The passengers’ travel path on the road network refers to individual travel OD path, including travel within one and two different zones. The travel distance used in the GIS network measurement tool refers to the length from the origin to the destination on one trip line. The origin does not apply to the beginning of a trip, it is set as the position where the walking or cycling to reach the primary transportation mode (including, buses, cars, taxis, and rail transits) ends. The destination refers to the position where the traveler is separated from the primary mode. The negative emission by passengers using bicycles and walking is negligible when calculating individual trip distance.
The traditional aggregate model was used in this study to predict passenger demand for the planning year. The travel path was set according to divided traffic zones. The travel distance within one zone and between two zones was calculated. The travel demand for each transportation mode was obtained based on the proportion through Mixed Logit model training. It was assumed that the structure and scale of the city’s road network will not change except for the planned rail transit. Traffic zones 9 and 14 were used as examples to illustrate the extraction method of travel distance. The passenger flow was generated between traffic zones 9 and 14 for the year 2023, while the proportion of rail transit, bus, car, and taxi was predicted. Thus, the travel demands for these four transportations were obtained. This paper provides two reasonable hypotheses to calculate the individual travel distance between two zones. First, the origin and destination for bus travel are set within 500 m of the bus stop; second, the starting point for cars and taxis was established as the centroid position of the traffic zone due to their unfixed stops. The travel OD routes generated on the bus lines, city road network, and rail transit lines are shown in
Figure 2 and
Figure 3. Through the travel distance combined with the emission factors of buses, cars, taxis, and trail transit, the carbon emissions between traffic zones 9 and 14 were calculated. Similarly, the carbon emissions generated by the demand of passenger trip in other traffic zones could be calculated. The combined calculation of its values shows the carbon emissions of different modes of transportation in Baoji. This provides the necessary parameters for the calculation model of rail transit carbon emission reductions.