A method for generating travel activity chains based on nested dynamic discrete choice

Abstract

The present invention discloses a travel activity chain generation method based on nested dynamic discrete choice, comprising the following steps: S1, defining the parameters of a dynamic discrete choice model, and constructing a traveler's full-day travel activity chain model; S2, defining a state transfer matrix, and establishing a connection between the state-choice pairs of consecutive stages; S3, deriving the expected utility function of the traveler in each stage based on the rational traveler and the utility maximization decision criterion; S4, defining the distribution of disturbance terms, and obtaining a nested dynamic discrete choice model with a closed form; S5, defining an immediate utility function according to the selection characteristics, and calculating the option-specific utility function; (6) estimating the parameters of various immediate utility functions in step (5), and generating a travel activity chain. The present invention introduces the distribution of disturbance terms based on the nested choice model to characterize the choice correlation of travel modes, and can better predict the complete travel activity chain of travelers.

Description

Technical Field

[0001] The invention relates to a travel activity chain generation method, in particular to a travel activity chain generation method based on nested dynamic discrete selection. Background Art

[0002] At present, my country's social economy and urbanization are developing rapidly. The scale of cities is expanding, the population is gathering rapidly, and the number of cars is growing rapidly. The tertiary industry, which is mainly based on entertainment and services, accounts for nearly 77%, indicating that people's social and economic activities have become more diverse and rich compared to traditional commuting activities. Travel is a derivative demand generated by activities. In the limited urban space and main activity time, it is necessary to reasonably and accurately predict the activity needs of residents to solve the problem of urban traffic congestion, adopt precise traffic management methods, effectively induce residents to travel, and optimize the supply and demand structure of urban traffic. All of the above have brought severe challenges to the prediction of urban traffic demand.

[0003] The traditional traffic demand model is mainly based on the four-stage method. Its OD thinking mode is not responsive enough to the built environment, spatial relations, spatial activities, spatial networks and spatial policies of urban space. It can only consider the aggregated utility generated by travelers. Therefore, it has the problems of single analysis angle and poor accuracy. The activity-based traffic demand model takes the travel activity chain as the analysis unit and each traveler as the analysis object. It is a non-aggregate travel demand prediction model. The model believes that travelers arrange activities according to their own wishes and connect activities with travel to form a chain, which makes up for the defects of the traditional model from the modeling principle.

[0004] The existing activity-based traffic demand forecasting models are mainly divided into constraint-based models, discrete choice-based models, computational process-based models and agent-based models according to the modeling methods. The common problem of current research is that the time, location, purpose and travel mode of a single trip are taken as the research objects, and the constraints and influence relationships between different trips are ignored. For example, the long duration of the previous activity will affect the location, travel mode and even the existence of the next activity; the possible combination of future activity trips will affect the activity decisions of travelers at the current stage (activity duration, etc.). In addition, decisions related to travel activities belong to multi-dimensional selection problems, which often have certain correlations. Ignoring this time and space correlation of activity trips will incorrectly reflect the choice behavior of travelers, resulting in lower demand forecasting accuracy. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a travel activity chain generation method based on nested dynamic discrete choice that can predict the complete travel activity chain of travelers.

[0006] Technical solution: The travel activity chain generation method of the present invention comprises the following steps:

[0007] S1, according to the properties of Markov chain, define the parameters of dynamic discrete choice model and construct the traveler's full-day travel activity chain model;

[0008] S2, define the state transition matrix to establish connections between state-selection pairs in consecutive stages;

[0009] S3, based on the rational travelers and the utility maximization decision criterion, derive the expected utility function of travelers in each stage;

[0010] S4, defining the distribution of the disturbance term, and obtaining a nested dynamic discrete choice model with closed form;

[0011] S5, defining the immediate utility function and calculating the option-specific utility function based on the selection characteristics of travel purpose, travel location, and travel mode;

[0012] S6, using known data to estimate the parameters of various immediate utility functions in step S5, and generating the travel activity chain of travelers based on the obtained expected utility function, and verifying the aggregate results of activity type time arrangement, travel mode selection, and location selection with observed data.

[0013] Furthermore, in step S1, the parameters of the dynamic discrete choice model include the time interval t, the current location l, the current activity type r, the historical activity record h, the historical travel mode record c, and the cumulative time period τ of the current activity; for the state variable x of the traveler at each stage k k , then x k =(t k , l k , r k ,h k , c k , τ k );

[0014] In each stage k, travelers need to make some travel activity choices to form a whole-day travel activity chain: t′ represents the binary choice of staying or traveling, r′ represents the travel purpose choice, l′ represents the travel location choice, and m′ represents the travel mode choice;

[0015] If the traveler chooses to stay, then r′ k =r k , l′ k = l k , m′ k =m stay ;

[0016] If the traveler chooses to travel, then r′k is the next activity to be executed, l′ k is the location of the next activity, m′ k Indicates the mode of travel;

[0017] If you use a k represents the choice made at stage k, then a k =(t′ k , r′ k , l′ k , m′ k ); Let the selection set be affected by the state variable and represented as C kj (x k ), then a k ∈C kj (x k );

[0018] Let y represent the traveler’s socioeconomic attributes and travel attribute set, including family attributes H, personal attributes I, travel distance TD, travel time TT and travel cost TC.

[0019] Furthermore, in step S2, the position and activity type of the k+1th stage correspond to the selection of the kth stage, and the connection between the state-selection pairs of the consecutive stages is expressed as: k+1 |x k , a k ) = l′ k and f(r k+1 |x k , a k ) = r′ k ;

[0020] When t′ k For traveling, and

[0021] Among them, when y does not belong to x, is equal to x∪y; otherwise, is equal to x;

[0022] The two time variables t and τ need to be updated separately according to the choice of staying or traveling. When staying, both t and τ change with time, so f(τ k+1 |x k , a k )=τ k +1 and f(t k+1 |x k , a k ) = t k +1;

[0023] When choosing to travel, the cumulative time period variable is reset to zero, f(τk+1 |x k , a k )=0, the time period variable needs to be added with the travel time.

[0024] Further, in step S3, ε k To represent the disturbance term, then ε k =(ε k (a k1 ), ε k (a k2 ), …, ε k (a kS )),in, Indicates that at stage k, it corresponds to option a kj unobservable influencing factors, S represents the number of options in this stage;

[0025] Assume that the traveler chooses a certain option in each stage, then the stage utility function V containing the disturbance term is k (x k , y, ε k ) is expressed as: The traveler’s current instantaneous utility function u k (x k , y, a k ), the sum of the discounted future utility function and the disturbance term;

[0026] Then the expected utility function V′ k (x k , y) is:

[0027]

[0028] Among them, V k Represents the stage utility function V k (x k , y, ε k ), represents the expected function of the disturbance term.

[0029] Further, in step S4, four nests are defined to cover all the choices, denoted by j;

[0030] Set to classify the nests to which it belongs only according to the travel mode in the option, including the stay nests stay , car nest auto , Public transportation nest transit and non-motorized nest non-motorized , the set of four nests is represented by Ω, then j∈Ω, and the selection set in nest j is C kj (x k ), a k ∈C kj (x k);

[0031] For option a in nest j k For example, the disturbance term is split according to the following formula:

[0032] ε k (a k )=ε kj +ε kj (a k )

[0033] Among them, ε kj For all options a in nest j k are equal, ε kj (a k ) is an option-specific disturbance term, and for all a k ∈C kj (x k ), ε kj (a k ) are independent of each other and follow the same Gumbel distribution;

[0034] For the expected utility function V′ k (x k , y), first select the largest item in nest j, and then compare between different nests, then we have:

[0035]

[0036] According to the properties of Gumbel distribution, we get:

[0037]

[0038] in,

[0039]

[0040]

[0041] v k (a k ) represents the option-specific utility function, which is obtained by adding the immediate utility function and the discounted future utility function; θ∈(0, 1) represents the discount coefficient for future utility.

[0042] Furthermore, in step S5, the instantaneous utility function u of stage k k (x k , y, a k ) consists of four functions, including the stay choice utility Purpose selection utility Location Utility and travel mode choice utility

[0043] When travelers choose to stay,

[0044] otherwise,

[0045] Finally, starting from the last stage K, the option-specific utility function is calculated, for all a K , v K (a K )=u K (a K ), and calculate the expected utility function V′ k (x k , y); for stage k∈{K-1, K-2, …, 2, 1}, calculate u k (a k ), and matches the possible state variables of the next stage according to the state transition matrix, so the option-specific utility function Until the first stage.

[0046] Compared with the prior art, the present invention has the following significant effects:

[0047] 1. A dynamic discrete choice model based on Markov process is introduced to consider the forward-looking behavior of travelers when making travel activity-related choices. Through a complete model framework, the travel mode, travel purpose, travel time and other decisions at different times of the day are linked, which solves the defects of the existing travel activity chain generation model, which is redundant and scattered, and difficult to connect in series.

[0048] 2. The introduction of disturbance term distribution based on the nested choice model can characterize the correlation of travel mode selection, so as to better predict the complete travel activity chain of travelers and meet the needs of relevant transportation departments for refined traffic demand management. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is the overall flow chart of the present invention;

[0050] Figure 2 To analyze the regional traffic zone division situation map;

[0051] Figure 3 (a) Comparison of observed and simulated distribution of departure time for commuting activities.

[0052] Figure 3(b) Comparison of observed and simulated distribution of departure time for school travel activities.

[0053] Figure 3(c) Comparison of the observed and simulated distribution of the departure time of the pick-up and drop-off activities.

[0054] Figure 3(d) Comparison of observed and simulated distribution of departure time for shopping trips.

[0055] Figure 3(e) Comparison of the observed and simulated distribution of the departure time of dining out activities.

[0056] Figure 3(f) Comparison of observed and simulated distribution of departure time of social travel activities.

[0057] Figure 3(g) Comparison of observed and simulated distribution of departure time for home travel activities.

[0058] Figure 3 (h) Comparison of observed and simulated distribution of departure times for other recreational travel activities;

[0059] Figure 4 This is a comparison result diagram of travel mode share;

[0060] Figure 5(a) shows the observation results of the location selection of each traffic zone.

[0061] Figure 5(b) shows the simulation results of the location selection of each traffic zone.

[0062] Figure 5(c) shows the simulation results and the absolute values ​​of the observation results for the selection of each traffic zone location. DETAILED DESCRIPTION

[0063] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0064] like Figure 1 As shown, the travel activity chain generation method of the present invention comprises the following steps:

[0065] S1, the traveler's full-day travel activity chain is abstracted into a mathematical modeling problem, and the properties of the Markov chain are used to define the following related concepts of the dynamic discrete choice model.

[0066] A day is divided into specific time periods, with 15 minutes as the time interval, represented by t, and a day is divided into 96 analysis units. The current location is represented by l. The current activity type is represented by r, including work, school, pick-up, shopping, dining, socializing, going home and other entertainment activities. The historical activity record is represented by h, and the historical travel mode record is represented by c. The accumulated time period of the current activity is represented by τ. All the above properties are the state variables of the traveler at each stage k, represented by x k If x k =(t k , l k , r k ,h k , c k , τ k ).

[0067] At each stage k, travelers need to make some travel activity choices to form a whole-day travel activity chain. First, travelers need to decide whether to continue the current activity or end the current activity and perform the next activity by traveling. t′ represents the two choices of staying or traveling. Further, r′ represents the choice of travel destination, l′ represents the choice of travel location, and m′ represents the choice of travel mode. If the traveler chooses to stay, then r′ k =r k , l′ k = l k , m′ k =m stay , if the traveler chooses to travel, then r′ k is the next activity to be executed, l′ k is the location of the next activity, m′ k Indicates the mode of travel, including walking, cycling, taking a car, taking a bus and taking the subway. k represents the choice made at stage k, then a k =(t′ k , r′ k , l′ k , m′ k ). Assume that the choice set is affected by the state variable and is represented by C kj (x k ), that is, a k ∈C kj (x k ).

[0068] Let y represent the traveler’s socioeconomic attributes and travel attribute set, including family attributes H, personal attributes I, travel distance TD, travel time TT and travel cost TC.

[0069] S2, define the state transfer matrix, connect the state-selection pairs of the consecutive stages, and use f(x k+1 |x k , a k ) indicates that the position and activity type of the k+1th stage correspond to the selection of the previous stage, that is, f(l k+1 |x k , a k ) = l′ k and f(r k+1 |x k , a k ) = r′ k The historical records of travel activities and travel modes need to store non-redundant information, so they are only updated when the traveler chooses to travel, that is, when t′ k For traveling, and When y does not belong to x, is equal to x∪y; otherwise, = x. The two time variables t and τ need to be updated separately according to the choice of staying or traveling. When staying, both t and τ change with time, that is, f(τ k+1 |x k , a k )=τ k +1 and f(t k+1 |x k , a k ) = t k +1; when choosing to travel, the cumulative time period variable is reset to zero, f(τ k+1 |x k , a k )=0, the time period variable needs to be added with the travel time, which is obtained through the Amap open platform API and needs to be converted into intervals of 15 minutes.

[0070] S3, assuming that travelers are rational and all selection processes adopt the utility maximization decision criterion, derive the expected utility function of travelers in each stage.

[0071] Since the above state variables x and attribute variables y cannot reflect all the influencing factors faced by travelers when making decisions, such as weather, mood, travel preferences, etc., ε k To represent the above unobservable influencing factors, that is, disturbance terms. Since unobservable factors are related to different options, ε k =(ε k (a k1 ), ε k (a k2 ), …, ε k (a kS )),in, Indicates that at stage k, it corresponds to option a kj The dynamic discrete choice model assumes that the traveler chooses an option at each stage to maximize the traveler's current instantaneous utility function u k (x k , y, a k ), the sum of the discounted future utility function and the disturbance term. This result is called the stage utility function with disturbance term, and is expressed as V k (x k , y, ε k ) is expressed. According to the above definition, we can get the following equation (in order to make the symbol expression clearer, we use u k (a k ) represents u k (xk , y, a k ), and V k Indicates V k (x k , y, ε k ), ε k+1 =(ε k+1，1 , …ε k+1，S )):

[0072]

[0073] Where θ∈(0,1) represents the discount coefficient for future utility, and E represents the expected function. Due to the existence of random disturbance terms, the calculation of the stage utility function is relatively complicated and requires the use of numerical simulation methods, which consumes a lot of time. In order to solve this problem, it is necessary to integrate the disturbance term to obtain the expected utility function V′ k (x k , y), and use V′ in the formula k Indicates that

[0074]

[0075] S4, defines the distribution of disturbance terms and obtains a nested dynamic discrete choice model with a closed form to simplify the calculation process, and further derives the calculation formulas of the expected utility function and conditional choice probability.

[0076] First, four nests are defined to cover all options, denoted by j. Assume that only the travel modes in the options are classified into the nests to which they belong, including the stay nest j stay , car nest auto , public transportation nest transit and non-motorized nest non-motorized The set of four nests is represented by Ω, that is, j∈Ω, and the selection set in nest j is C kj (x k ), that is, a k ∈C kj (x k ).

[0077] For option a in nest j k For example, the disturbance term can be decomposed according to the following formula:

[0078] ε k (a k )=ε kj +ε kj (a k ) (3)

[0079] where ε kj For all options a in nest j k are equal, εkj (a k ) is an option-specific disturbance term, and for all a k ∈C kj (x k ), ε kj (a k ) are independent of each other and follow the same Gumbel distribution, whose scale parameter is ρ j , the location parameter is 0. For the maximization function in the expected utility function formula, it is split into two parts, first selecting the largest item in nest j, and then comparing between different nests. It can be written as:

[0080]

[0081] Because for all a k ∈C kj (x k ), ε kj (a k ) are independent of each other and obey the same Gumbel distribution, so according to the properties of this distribution, we can get:

[0082]

[0083] in,

[0084]

[0085]

[0086] v k (a k ) is in the form of v k (x k , y, a k ), represents the option-specific utility function, which is obtained by adding the immediate utility function and the discounted future utility function; e represents a natural constant, ρ j represents the scale parameter of the Gumbel distribution specific to nest j. According to the linear transfer property of the Gumbel distribution (i.e., if the random variable A ≈ Gumbel(η, μ), then A+b ≈ Gumbel(η+b, μ), where b represents an arbitrary real number), we can define:

[0087]

[0088] in,

[0089] ε′ kj ~Gumbel(0,ρ j )

[0090] Therefore, the expected utility function can be simplified to

[0091]

[0092] Assume that the disturbance term ε for all nests j is kj Independent of each other, aggregate disturbance terms They are independent of each other and follow the same Gumbel distribution, with a scale parameter of 1 and a location parameter of 0. According to the properties of the distribution, we can further obtain:

[0093]

[0094] Where γ represents the Euler constant. According to the split form of the above maximization function, we can get the conditional selection probability formula, that is, the probability of selecting nest j among all nests Ω, P(j|Ω, x k , y), multiplied by the choice a in nest j k The probability P(a k |j,x k , y). Combining two independent and identically distributed disturbance terms (i.e. and ε kj (a k )) The closed logit calculation form obtained, the conditional selection probability formula is:

[0095]

[0096] j′ and a′ k and represents j and a k The meaning is the same (for example, if there are buses, subways and cars, and we want to calculate the probability of choosing a car, then a k represents a car, a′ k represents bus, subway and car). The conditional selection probability can be directly obtained through and v k (a k ) is calculated. Next, we need to define the immediate utility function u in the option-specific utility function k (x k , y, a k ), thus obtaining a complete computational framework.

[0097] S5, based on the selection characteristics of travel purpose, travel location, and travel mode, the immediate utility function is defined, and the process of calculating the option-specific utility function through the reverse derivation process is given.

[0098] The immediate utility function at stage k is composed of four functions, including and They correspond to the stop selection utility function, the destination selection utility function, the location selection utility function, and the travel mode selection utility function. otherwise

[0099] The stay choice utility is affected by the currently performed activity type r, the time variable t, and the cumulative period variable τ, that is,

[0100]

[0101] β end and β duration The utility parameter represents the end time of the activity and the duration of the activity. Here r, τ, t represent dummy variables with corresponding meanings. The travel purpose utility is affected by the family attribute H, personal attribute I, historical activity record h and time variable t, where and D h represents the dummy variables related to family attributes H, personal attributes I and historical activity records h, β H , β I , β h and β t The utility parameters representing family attributes, personal attributes, historical activity records, and time variables are as follows:

[0102]

[0103] The location utility of traveler n is given by:

[0104]

[0105] It is related to the current location, the selected location and the selected activity type. ηl′l Indicates the distance from the current location l to the selected location l' in segment η. The linear segments include 0-1km, 1-2km, 2-5km, 5-15km and above 15km. η Represents the parameter of the corresponding distance segment. r′l′ It represents the attraction of activity r′ at location l′, expressed by the number of POI (Point of Interest) of the corresponding type. r′l′ =0,ω r′l′ =1;δ r′l′ >0,ω r′l′ = 0. M represents a relatively large real number (set to 999 in the present invention) used to punish places that do not have corresponding activity attractions.

[0106] The utility of travel mode m′ in nest j is the sum of the following two parts:

[0107]

[0108]

[0109] TTm′ll′t and TC m′ll′t represents the time and cost of traveling from the current location l to the selected location l′ via travel mode m′ at time t, β TT，m′ and β TC，m′ Utility parameters representing time (Travel Time) and cost (Travel Cost). D c and D r′ Represents dummy variables related to household attributes H, personal attributes I, historical travel mode records c, and new activity types r′. H , β I , β c Utility parameters representing household attributes, personal attributes, and historical travel mode records.

[0110] Finally, computing the option-specific utility function requires starting from the last stage K and for all a K , v K (a K )=u K (a K ), and calculate the expected utility function V′ k For stage k∈{K-1, K-2, …, 2, 1}, calculate u k (a k ), and matches the possible state variables of the next stage according to the state transition matrix, so the option-specific utility function expression is:

[0111]

[0112] Until the first stage.

[0113] S6, using known data to estimate the parameters of various immediate utility functions in step S5, and generating the travel activity chain of travelers based on the obtained expected utility function, and verifying the aggregate results of activity type time arrangement, travel mode selection, and location selection with observed data.

[0114] Taking the travel questionnaire survey data of Chongqing residents as an example, this embodiment is further explained. The data includes 158,665 travel records of 30,422 households and 72,170 travelers, where the travel purposes include work, school, pick-up, shopping, dining, socializing, going home and other entertainment activities, and the travel modes are integrated into walking, cycling, taking a car, taking a bus and taking the subway. The locations include 467 traffic zones divided by the main urban area of ​​Chongqing, such as Figure 2As shown. Using the standard maximum likelihood estimation, we get the complete instantaneous utility functions, and then we can calculate the utility functions and expected utility functions of the options at each stage. And we can use this to calculate the utility functions of each state variable x at each stage k. k Next option a k The conditional choice probability is selected, and the largest option is selected as the simulation result. The calculation is performed from the first stage to the last stage, and finally the complete travel activity chain of all travelers under the given immediate utility is obtained. The various characteristics of the activity chain obtained by simulation, including activity type time arrangement, travel mode, and location aggregation, are aggregated to obtain the aggregated results and compared with the observed results. Figure 3 (a)-(h) shows the distribution of departure times for eight travel activities. Figure 4 The comparison results of travel mode share are shown in Figure 5 (a)-(c). The simulation results, observation results and the absolute value of the simulation results and observation results of the location selection of each traffic zone are shown. In general, the model based on nested dynamic discrete choice has a good performance in generating travel activity chains because it can capture the correlation between travelers' forward-looking choice behavior and travel mode selection.

Claims

1. A travel activity chain generation method based on nested dynamic discrete choice, characterized in that: The steps include: S1, according to the properties of Markov chain, define the parameters of dynamic discrete choice model and construct the traveler's full-day travel activity chain model; S2, define the state transition matrix to establish connections between state-selection pairs in consecutive stages; S3, based on the rational travelers and the utility maximization decision criterion, derive the expected utility function of travelers in each stage; S4, defining the distribution of the disturbance term, and obtaining a nested dynamic discrete choice model with closed form; S5, defining the immediate utility function and calculating the option-specific utility function based on the selection characteristics of travel purpose, travel location, and travel mode; S6, using known data to estimate the parameters of various immediate utility functions in step S5, and generating the travel activity chain of the travelers according to the obtained expected utility function, and verifying the aggregated results of activity type time arrangement, travel mode selection, and location selection with observed data; In step S1, the parameters of the dynamic discrete choice model include time interval t, current location l, current activity type r, historical activity record h, historical travel mode record c, and the cumulative period τ of the current activity; for the state variable x of the traveler at each stage k, k , then x k =(t k ,l k ,r k ,h k ,c k ,τ k ); In each stage k, travelers need to make some travel activity choices to form a whole-day travel activity chain: t′ represents the binary choice of staying or traveling, r′ represents the travel purpose choice, l′ represents the travel location choice, and m′ represents the travel mode choice; If the traveler chooses to stay, then r k ′=r k , l′ k = l k , m′ k =m stay ; If the traveler chooses to travel, then r k ' is the next activity to be executed, l' k is the location of the next activity, m′ k Indicates the mode of travel; If you use a k represents the choice made at stage k, then a k =(t′ k ,r k ′,l′ k ,m′ k ); Let the selection set be affected by the state variable and represented as C kj (x k ), then a k ∈C kj (x k ); Let y represent the socioeconomic attributes and travel attribute set of the traveler, including family attributes H, personal attributes I, travel distance TD, travel time TT and travel cost TC; In step S2, the position and activity type of the k+1th stage correspond to the selection of the kth stage, and the connection between the state-selection pairs of the consecutive stages is expressed as: k+1 |x k ,a k ) = l′ k and f(r k+1 |x k ,a k )=r k ′; When t′ k For traveling, and Among them, when y does not belong to x, is equal to x∪y; otherwise, is equal to x; The two time variables t and τ need to be updated separately according to the choice of staying or traveling. When staying, both t and τ change with time, so f(τ k+1 |x k ,a k )=τ k +1 and f(t k+1 |x k ,a k ) = t k +1; When choosing to travel, the cumulative time period variable is reset to zero, f(τ k+1 |x k ,a k )=0, the time period variable needs to be added with the travel time; In step S3, ε k To represent the disturbance term, then ε k =(ε k (a k1 ),ε k (a k2 ),…,ε k (a kS )),in, Indicates that at stage k, it corresponds to option a kj unobservable influencing factors, S represents the number of options in this stage; Assume that the traveler chooses a certain option in each stage, then the stage utility function V containing the disturbance term is k (x k ,y,ε k ) is expressed as: The traveler’s current instantaneous utility function u k (x k ,y,a k ), the sum of the discounted future utility function and the disturbance term; Then the expected utility function V k ′(x k ,y) is: Among them, V k Represents the stage utility function V k (x k ,y,ε k ), represents the expected function of the disturbance term; In step S4, four nests are defined to cover all the choices, denoted by j; Set to classify the nests to which it belongs only according to the travel mode in the option, including the stay nests stay , car nest auto , Public transportation nest transit and non-motorized nest non-motorized , the set of four nests is represented by Ω, then j∈Ω, and the selection set in nest j is C kj (x k ), a k ∈C kj (x k ); For option a in nest j k For example, the disturbance term is split according to the following formula: e k (a k )=e kj +e kj (a k ) Among them, ε kj For all options a in nest j k are equal, ε kj (a k ) is an option-specific disturbance term, and for all a k ∈C kj (x k ), ε kj (a k ) are independent of each other and follow the same Gumbel distribution; For the expected utility function V k ′(x k ,y), first select the largest item in nest j, and then compare between different nests, then we have: According to the properties of Gumbel distribution, we get: in, v k (a k ) represents the option-specific utility function, which is obtained by adding the immediate utility function and the discounted future utility function; θ∈(0,1) represents the discount coefficient for future utility; In step S5, the instantaneous utility function u of stage k k (x k ,y,a k ) consists of four functions, including the stay choice utility Purpose selection utility Location Utility and travel mode choice utility When travelers choose to stay, otherwise, Finally, starting from the last stage K, the option-specific utility function is calculated, for all a K , v K (a K )=u K (a K ), and calculate the expected utility function V k ′(x k ,y); for stage k∈{K-1,K-2,…,2,1}, calculate u k (a k ), option-specific utility function Until the first stage.

Images