Multi-mode travel structure optimization method and system for time value heterogeneous population

Abstract

The invention discloses a multi-mode travel structure optimization method and system for time value heterogeneous crowds, and the method comprises the steps: building an extended clock network to describe a travel behavior selection decision of a traveler in a period, and converting the travel behavior selection decision into a specifically represented path; a carbon point reward and punishment mechanism rule is adopted, and travelers are promoted to turn to more bus travel modes by combining reward and punishment measures; establishing a multi-mode travel behavior selection decision model and an online car-hailing market balance model; a carbon point reward and punishment mechanism model for leader decision making is established based on a multi-mode travel behavior selection decision model and an online car-hailing market balance model, is a double-layer optimization model composed of an upper-layer model and a lower-layer model, and is used for researching how leader designs a carbon point reward and punishment mechanism rule under different carbon emission targets. According to the invention, traffic management measures of reward and punishment are combined, heterogeneity of time value of travelers is considered, and public travel can be effectively promoted to be changed from driving travel to bus travel or online car-hailing travel.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic control and traffic management, and in particular to a method and system for optimizing the multi-mode travel structure of people with heterogeneous time values. Background Art

[0002] As the number of motor vehicles increases, the problems of traffic carbon emissions and road congestion are becoming more and more serious. There are usually three ways to solve the above problems: the first is to restrict and ban the use of motor vehicles from the source. Some cities have adopted the method of limiting the use of motor vehicles based on the last digit of the license plate, but this method can offset the impact of travelers by purchasing a second car over time; the second is to use punitive measures to encourage travelers to reduce the frequency of driving. Some cities have adopted congestion charges, but congestion charges lack social fairness, which may burden low-income families and increase traffic on surrounding roads, causing more serious congestion problems in surrounding areas; the third is to encourage people to switch to public transportation through rewards. Some cities have established a green life reward platform, but the current reward method has little effect in promoting travelers to use public transportation to reduce carbon emissions. Summary of the invention

[0003] The first purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and provide a multi-modal travel structure optimization method for people with heterogeneous time values. It combines traffic management measures with rewards and penalties in parallel, takes into account the heterogeneity of travelers' time values, and can effectively promote public travel from driving to public transportation or online car-hailing, thereby reducing traffic carbon emissions and road congestion.

[0004] The second object of the present invention is to provide a multi-modal travel structure optimization system for people with heterogeneous time values.

[0005] The first objective of the present invention is achieved by the following technical solution: A method for optimizing the multi-modal travel structure of a population with heterogeneous time value, comprising the following steps:

[0006] 1) Establish an extended clock network to describe the travel behavior selection decision of travelers in a cycle and convert it into a specific path; the structure of the extended clock network is 3 columns and Γ rows, Γ is the number of days in the cycle, each column corresponds to the three travel modes of driving, online car-hailing and public transportation, each row corresponds to each day in the cycle, and the rows and columns determine the travel mode selected by the traveler on a certain day; the extended clock network uses each day as a decision point. In order to ensure the effectiveness of the extended clock network path, virtual links are added between various travel modes every day. The purpose of establishing the extended clock network is to accurately represent the travel behavior selection decision of travelers every day;

[0007] 2) Establishing carbon credit reward and punishment mechanism rules, the purpose is to encourage travelers to turn to public transportation mode by combining reward and punishment measures; the carbon credit reward and punishment mechanism rules are that travelers can choose one of the travel modes of driving, public transportation, and online car-hailing every day during the cycle, and their carbon credit accounts will gain points or deduct corresponding points. At the end of the cycle, travelers with positive carbon credit accounts will be rewarded, and travelers with negative carbon credit accounts will pay fines; according to the path of the travelers in the expanded clock network obtained in step 1), the path is calculated according to the daily travel mode. The surplus of the carbon credit account at the end of the cycle is calculated to obtain the carbon credit cost of the cycle;

[0008] 3) Establish a multi-modal travel behavior selection decision model and an online car-hailing market equilibrium model; the multi-modal travel behavior selection decision model first discretizes travelers into different types of travelers according to their time value. Time value measures the value that different travelers can create per hour, which is related to wages. After discretizing the population according to time value, the number of travelers of each type is obtained, and then the decision of each type of traveler, that is, the travel mode of each day in the cycle is described. The model goal is to minimize the total travel cost of all travelers in a cycle according to the decision of the travelers, including time cost, monetary cost and carbon credit cost calculated based on the carbon credit reward and punishment mechanism rules in step 2). Since the decisions of travelers are independent, the objective function of the multi-modal travel behavior selection decision model is obtained by linearly superimposing the objective function of each traveler, and the decision of each traveler's travel mode every day in a cycle is output; the online car-hailing market equilibrium model describes the number of online car-hailing drivers and passenger waiting time when the online car-hailing market is in equilibrium under given online car-hailing market parameters, and then calculates the time cost and monetary cost of online car-hailing travel every day;

[0009] 4) Establish a carbon credit reward and punishment mechanism model for leader decision-making. The carbon credit reward and punishment mechanism model for leader decision-making is a two-layer optimization model composed of an upper and lower model, which is used to study how leaders design carbon credit reward and punishment mechanism rules under different carbon emission targets; the upper model takes different carbon emission targets as the objective function, and considers Pareto improvement and income neutrality to limit the output of carbon credit reward and punishment mechanism rules. Pareto improvement means that after the implementation of the carbon credit reward and punishment mechanism rules, the total travel cost of a traveler in one cycle is less than that without the implementation of the carbon credit reward and punishment mechanism rules; the lower model is the multi-modal travel behavior selection decision model and the online car-hailing market equilibrium model proposed in step 3), and an algorithm is designed to calculate and initialize the lower model to output the traveler's periodic travel behavior selection decision and the online car-hailing market operation status, and use it as the input of the upper model to solve the upper model output as the carbon credit reward and punishment mechanism rules, and then input the lower model to cycle until the objective function values ​​of the upper models do not change twice before terminating the algorithm, and finally output the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision and the online car-hailing market operation status.

[0010] Furthermore, in step 1), an extended clock network is established to describe the traveler's periodic travel behavior selection decision and converted into a concretely representable path, as follows:

[0011] For each traveler, he needs to complete a trip from the starting point set O to the destination point set D every day. The only transportation modes available to travelers are driving, online car-hailing and bus. In the extended clock network, there are 3 columns and Γ rows, Γ is the number of periodic days, and the 1st, 2nd and 3rd columns correspond to the travel modes of driving, online car-hailing and bus respectively; the i-th row corresponds to the i-th day, and the travel mode of the day is determined by the rows and columns together. All selected nodes are connected to obtain the periodic travel path; let a be the link selected by the travel mode, a∈A1 represents the link set for choosing driving, which is represented by the link connecting any two nodes in the first column in the extended clock network; a∈A2 represents the link set for choosing online car-hailing, which is represented by the link connecting any two nodes in the second column in the extended clock network; a∈A3 represents the link set for choosing bus travel, which is represented by the link connecting any two nodes in the third column in the extended clock network; a∈A4 represents the link set for converting between various travel modes, which is a virtual link that does not require a travel cost, and is represented by the link connecting different travel modes in the same row in the extended clock network.

[0012] Further, in step 2), the carbon credit reward and penalty mechanism is implemented with a period of Γ days, where I is a set of days, I = {1, 2, ..., Γ}. At the beginning of the period, the amount of points added or deducted for the three travel modes per day is predetermined (κ a , κ b ,κ c ) and the amount of reward or punishment the traveler receives after the cycle ends (p c ,p r ), as follows:

[0013] All travelers will get a carbon credit account with an initial balance of zero at the beginning of the cycle, and a κ a Every day when you take a ride-hailing car, your account will increase or deduct k b For the online car-hailing mode, the goal is to encourage people to switch to online car-hailing or to switch to public transportation. Therefore, κ b Can be positive or negative, the account increases by κ every day by taking public transportation c At the end of the period, if the traveler's carbon credit account is positive, the traveler will receive p r If the carbon credit account is negative, the traveler will be rewarded at p c The penalty rate is paid, for (κ a ,κ b , κ c) The size relationship between the three is unlimited, and they can be equal or unequal. For (p c ,p r )Requirement|p c |>|p r |;

[0014] Since the positive and negative relationship of the carbon credit account is uncertain at the end of the period, it cannot be directly expressed by a constraint equation, so an auxiliary function is introduced and The carbon integral cost CCR corresponding to the path r of the extended clock network in one cycle of the traveler r , if the balance of the traveler's carbon credit account at the end of the period is equal to If the balance of the traveler's carbon credit account is negative at the end of the period, then equation (2) is equal to

[0015]

[0016] In the formula, δ ar Indicates whether the driving route is selected, the value is 1 if selected, and 0 if not selected; δ br Indicates whether the route of the online car-hailing trip is selected, the value is 1 if selected, and the value is 0 if not selected; δ cr Indicates whether the bus travel route is selected, the value is 1 if selected, and 0 if not selected.

[0017] Furthermore, in step 3), the specific situation of establishing a multi-modal travel behavior selection decision model is as follows:

[0018] The crowd is divided into car owners and car-free people. Car owners can choose to travel by car, online car-hailing or public transportation, while car-free people can only choose online car-hailing or public transportation. The car owners and car-free people are discretized separately, and the set M is used to represent the car owners and car-free people. 1 ={1,2,...,m 1} and M 2 ={1,2,...,m 2} indicates that m 1 and m 2 They represent the discretized population categories of car-owning and car-free populations, and are discretized with τ as the length of each interval according to the log-normal distribution curve. The average value of each interval is taken as the VOT of the discrete interval, denoted as β n,m , where n = 1 represents the car-owning population, n = 2 represents the car-free population, m represents the VOT category to which each sub-population belongs, and the probability density function describing VOT is recorded as f(x), x∈[a0,b0], x represents the time value, a0, b0 represent the left endpoint and right endpoint of the above log-normal distribution curve respectively; the total number of car-owning and car-free populations N1 、N 2 is given, record N n,m is the number of people corresponding to the (n,m) group of people;

[0019]

[0020] β n,m =a0+τm-τ / 2 (5)

[0021] After defining the expression of discrete population, the number of people in the (n, m)th category who choose the travel mode on the i-th day is defined as a=1 means driving, a=2 means online car-hailing, and a=3 means public transportation;

[0022] For time cost and monetary cost, remember is the sum of the time cost and monetary cost of the travel mode selected by the (n,m)th person on the i-th day; in the extended clock network, the time cost of the neighboring node is calculated by multiplying the time value by the time spent, and the monetary cost is given directly; the time spent on the public transportation mode is t transit Due to the high punctuality and consistency of bus schedules, regardless of road traffic, the monetary cost of bus travel c transit Fixed, so the sum of the time cost and monetary cost of the (n,m)th person taking public transportation on the i-th day is expressed as:

[0023]

[0024] For the driving mode, the time spent is related to the road traffic, using the function Indicates that the road flow on the i-th day includes the total number of people traveling by car And the number of drivers entering the online car-hailing market u i , ph is defined as the peak hour number and takes 2 hours, and the time spent on driving travel mode on the i-th day is obtained as t 1,i The calculation formula is:

[0025]

[0026] Monetary cost of driving car is fixed, and the sum of the time cost and monetary cost of the (n,m)th person driving on the i-th day is expressed as:

[0027]

[0028] The time spent on online car-hailing travel includes driving time and waiting time. The driving time is equal to the driving time on the same day. represents the waiting time of passengers on the i-th day, u i represents the number of drivers entering the online car-hailing market, cride Denotes the monetary cost of taking an online ride-hailing car every day, and the expression for the sum of the time cost and monetary cost of the travel mode chosen by the (n,m)th person on the i-th day is:

[0029]

[0030] For all travelers, each person's goal is to minimize the generalized travel cost within a cycle, without considering that the change of their own travel behavior choice decision will increase the travel cost of other travelers. Therefore, the objective function of the multi-modal travel behavior choice decision model is to minimize the periodic generalized total travel cost of all travelers. The decisions between each person do not affect each other, so the travel costs of travelers of different (n, m) types are linearly superimposed; the objective functions of the car group and the unmanned car group in the multi-modal travel behavior choice decision model are given below:

[0031] For car owners, the objective function is expressed as formula (10), where R 1 It is defined as the set of all cycle paths selected by the car-owning group in the extended clock network. Defined as the (1,m)th type of traffic flow that chooses path r:

[0032]

[0033] For people without cars, the objective function is expressed as formula (11), where R 2 It is defined as the set of all periodic paths chosen by the car-free population in the extended clock network. Defined as the (2,m)th type of traffic that chooses path r:

[0034]

[0035] Therefore, for the entire multi-modal travel behavior selection decision model, the objective function is to minimize the sum of equations (10) and (11):

[0036]

[0037] The constraints are the constraints on the flow, i.e., formulas (13) to (17), and the constraints on the rules of the carbon credit reward and punishment mechanism in step 2), i.e., formula (2):

[0038]

[0039]

[0040] Furthermore, in step 3), the specific situation of establishing the online car-hailing market equilibrium model is as follows:

[0041] Considering the static online car-hailing market model, the number of waiting passengers on the i-th day is The passenger arrival rate Q on day i i The average waiting time of passengers on day i The number of empty vehicles in the ith Arrival rate of vehicles in the sky through the i-th The average idle time of drivers per hour on day i The product of is approximated, assuming that the online ride-hailing platform matches waiting passengers with the nearest available vehicles and the online ride-hailing market is in a steady state. and the number of empty vehicles Remains unchanged within a day; the platform's hourly successful matching rate is recorded as m c-vt , which is determined by the number of waiting passengers and the number of empty vehicles on the i-th node, the matching rate is described by formula (18):

[0042]

[0043] Where A is a proportional parameter used to capture the external features that affect the matching results, which is the market area S divided by the running speed v of the empty vehicle. vt Indicates A;

[0044] When the online car-hailing market is in equilibrium, the hourly passenger arrival rate, hourly empty vehicle arrival rate, and hourly online car-hailing platform successful matching rate are equal, and the number of online car-hailing vehicles occupied per hour is The hourly passenger arrival rate Q i The average driving time per order of online car-hailing drivers t a,i , The product of is given by , therefore, the number of drivers entering the online car-hailing market per hour on the i-th day is u i It is the sum of the number of occupied online-hailing vehicles and the number of empty vehicles, expressed as:

[0045]

[0046] On the labor supply side, the number of drivers entering the online car-hailing market per hour on day i is u i It is not fixed. Whether a driver is willing to enter the online car-hailing market and provide services depends on his / her hourly reservation value r′, that is, the minimum hourly wage that potential online car-hailing owners are willing to provide services. Due to the heterogeneity of drivers, their reservation values ​​r′ are different. We use uniform distribution to describe the reservation value of a group of drivers, and assume that r′~U(0,μ m ), U(0,μ m ) means obeying (0,μ m ), μ m is the maximum retention value of the driver group; γ is defined as the commission of the online car-hailing platform, and the net profit η per driver per hour is expressed as follows:

[0047]

[0048] Assume that the maximum number of drivers entering the online car-hailing market per hour is u max For each driver, only when the net profit per hour η is greater than the reserved value r′ will the driver enter the market, that is, η ≥ r′. Define g(·) as the uniform distribution probability density function. The following formula is used to express the number of drivers entering the online car-hailing market every hour per day u i :

[0049]

[0050] On the demand side, the demand for online ride-hailing vehicles on day i is Q i ′ is the result of the traveler's choice based on the minimization of the periodic generalized travel cost:

[0051]

[0052] Furthermore, in step 4), a carbon credit reward and punishment mechanism model for leader decision-making is established, and the carbon credit reward and punishment mechanism rules (κ a , κ b , κ c ,p c ,p r ) and the corresponding traveler’s periodic travel behavior selection decision;

[0053] The carbon credit reward and punishment mechanism rules are used to achieve two goals. On the one hand, it is hoped that road congestion can be reduced by reducing the volume of vehicles on the road. On the other hand, it is hoped that the carbon emissions from road traffic can be reduced by reducing the number of travelers driving and promoting the shift to travel modes including online car-hailing and public transportation. In order to comprehensively consider the impact of the above two goals on the optimization of the carbon credit reward and punishment mechanism rules, the objective function is described by a multi-objective function and the objective weight factor ρ is introduced. ρ∈(0,1) is used to express the importance of the goal of reducing congestion, and 1-ρ∈(0,1) is used to express the importance of the goal of reducing traffic carbon emissions.

[0054] The congestion reduction target is measured by a road traffic congestion index, which is calculated by dividing the actual traffic volume of a road by its maximum capacity:

[0055]

[0056] For the reduction of transportation carbon emissions, the continuous function E of the COPERT emission function is used. a (·) calculation, E a (·) is a function of the average travel time t a,i The carbon emission E on the i-th day is obtained by the following formula: i , where l is defineda is the actual road distance in km:

[0057]

[0058] The objective function of the carbon credit reward and punishment mechanism model for leader decision-making is:

[0059]

[0060] The constraints of the carbon credit reward and punishment mechanism model for leader decision-making include income neutrality constraints and Pareto improvement constraints. The income neutrality constraint ensures that the income or subsidy obtained by the leader in each cycle remains within a controlled range; since the carbon credit reward and punishment mechanism rules are not intended to generate profits, limited subsidies are allowed; the income neutrality constraint strictly limits the income or subsidy generated by the carbon credit reward and punishment mechanism rules to B = 0, but in actual situations, the threshold is relaxed and defined as: -ε≤B≤ε, where B represents the net income or subsidy of the carbon credit reward and punishment mechanism rules in one cycle, and f r n,m represents the number of (n,m)-th type travelers who choose path r, ε is a constant, and the income neutrality constraint is expressed as:

[0061]

[0062] When implementing the carbon credit reward and punishment mechanism rules, the welfare of travelers needs to be improved, that is, the total travel cost of the cycle after the implementation of the carbon credit reward and punishment mechanism rules for (n, m) type travelers must be lower than the cost when the carbon credit reward and punishment mechanism rules are not implemented. The carbon credit reward and punishment mechanism rules that achieve this constraint are called Pareto improvements, which are expressed as formula (27), where It represents the sum of the time cost and monetary cost of the (n,m)th person who chooses the appearance mode on the i-th day without implementing the carbon credit reward and punishment mechanism rules:

[0063]

[0064] Design algorithm calculation, initialize the lower model to output travelers' periodic travel behavior choice decisions and online car-hailing market operations, and use them as the input of the upper model. Solve the upper model to obtain the output of the carbon credit reward and punishment mechanism rule setting, and then input the lower model cycle until the objective function value of the upper model twice does not change, then terminate the algorithm, and finally output the carbon credit reward and punishment mechanism rule setting, travelers' periodic travel behavior choice decisions, and online car-hailing market operations.

[0065] Furthermore, in order to solve the multi-modal travel behavior selection decision model in step 3), the Lagrange multiplier is introduced to transform it into a complementary relaxation problem, where Z(f) is the objective function of the multi-modal travel behavior selection decision model established in step 3), f and λ are vector groups composed of flow and Lagrange multipliers, respectively, and the vector λ 1,m , 2,m denote the Lagrange multiplier corresponding to the flow of the mth group of people with cars and the Lagrange multiplier corresponding to the flow of the mth group of people without cars, respectively, and denote the Lagrange multiplier corresponding to equations (16) and (17); from an economic point of view, λ 1 and λ 2 They represent the minimum travel cost of the car group and the car-free group in the equilibrium state, λ 1 and λ 2 are all positive numbers strictly greater than 0, L(f,λ) represents the Lagrangian function related to f and λ;

[0066]

[0067] When the above equation (28) satisfies the KKT condition, it is converted into the following equivalent problem:

[0068]

[0069]

[0070] In the formula, represents the total cycle cost of the path currently selected by the mth group of people with cars, It represents the total cycle cost of the path currently selected by the mth group of people without cars.

[0071] Furthermore, the steps of the algorithm for solving the multi-modal travel behavior choice decision model are as follows:

[0072] a. Initialize the parameters of each variable and construct the initial solution by finding the shortest path and performing all-or-nothing allocation. All-or-nothing allocation means loading the traffic of each initial group into the extended clock network according to the shortest path obtained above. The main iteration number j = 0;

[0073] b. Based on the initial solution, generate a new path through the column generation method, iterate it, and evaluate the convergence, j = j + 1;

[0074] c. Allocate traffic on the new path until the multi-modal travel behavior choice decision model reaches an equilibrium state;

[0075] For the column generation method in step b, a new shortest path for each group of people needs to be obtained. The carbon credit cost as a cycle cost cannot be directly expressed as a point-to-point cost. It is processed by using the k shortest path method on the extended clock network. Since the carbon credit cost will also affect the cost between nodes, directly using the extended clock network that only contains the time and monetary costs between directly connected nodes will lead to inaccurate results. Therefore, the input extended clock network is modified as follows: For the driving node, in addition to the time cost and monetary cost, a virtual daily carbon credit cost κ is added a ×p c ; For bus nodes, subtract a virtual daily carbon credit cost κ c ×p r ; For online car-hailing nodes, the virtual carbon credit cost is adjusted according to the carbon credit reward and punishment mechanism rules. If κ b >0, subtract κ b ×p r , if κ b If <0, add κ b ×p c ;

[0076] For each group of people, the shortest path is obtained according to the above-mentioned extended clock network and Dijkstra algorithm is applied. Each node on the generated shortest path is regarded as a deviation node. The path from the starting point to the deviation node is fixed, and the subsequent path is recalculated to generate a new candidate path until k paths are generated;

[0077] For step c, define a function G(f,λ) related to (f,λ), where Ω(a,b) is the constructed auxiliary function, which is specifically expressed as:

[0078]

[0079] There is a mathematical property for the function G(f,λ): there is always a set of (f,λ) such that G(f,λ)=0, and when G(f,λ)=0 is the global function minimum, so the algorithm aims to make G(f,λ)=0, which is achieved by gradient descent method and search step size, but due to the non-convexity of the objective function, it will face local optimality during the descent process, making it difficult for the function to continue to descend. Therefore, when the function falls into the local optimal solution, the momentum method and random perturbation are introduced to make the function jump out of the local optimal solution and continue to descend; for the judgment of the local optimal solution, the judgment method used is: if the absolute value of the difference of G(f,λ) in five consecutive times is less than ε3, ε3 is a number close to 0, then it is considered that it is trapped in the local optimal solution at this time. The mathematical expression is as follows, where f (j,k+n) , (j,k+n) They are the flow vector and Lagrange multiplier vector corresponding to the sub-iteration number k+n under the main iteration number j:

[0080] |G(f(j,k+n) ,λ (j,k+n) )-G(f (j,k) ,λ (j,k) )|≤ε3 (34)

[0081] The gradient d of the function G(f,λ) at the kth sub-iteration k The expression is as follows (35). Note that p and q are used to distinguish the expressions of r and m in calculating the gradient. In fact, they still represent the corresponding paths and populations. (j) , (j) ,j=0,1,2,... represents the flow vector of the path selected by each group at the jth main iteration and the Lagrange multiplier vector f and λ corresponding to the flow; f (j,k) , (j,k) ,j=0,1,2,...,k=0,1,2,... represents the flow vector of the path selected by each group in the kth sub-iteration under the jth main iteration and the Lagrange multiplier vector f and λ corresponding to the flow, They represent the number of people in the qth category who choose path p in the kth sub-iteration under the jth main iteration, and the number of people in the qth category who choose path p in the car-carrying and car-free groups, respectively. They represent the Lagrange multipliers of the qth type of people in the car-owning and car-free populations corresponding to the kth sub-iteration under the jth main iteration:

[0082]

[0083] The function G(f,λ) decreases by a step size ρ at the kth sub-iteration k is the maximum value that satisfies the following inequality, c is a constant approaching 0, Represents the transpose of the gradient matrix of the function G for the flow f and the Lagrange multiplier λ at the kth sub-iteration under the jth main iteration:

[0084]

[0085] When there is no local optimal descent, the descent rule of G(f,λ) for the kth sub-iteration is:

[0086] f (j,k+1) =f (j,k) +ρ k d k ; (j,k+1) =λ (j,k) +ρ k d k (37)

[0087] In the formula, f (j,k-1) , (j,k-1) represents the flow f and Lagrange multiplier of the k+1th sub-iteration under the jth main iteration;

[0088] When falling into a local optimum, the gradient obtained by the momentum method is used as the function descent search direction. The momentum method accumulates the historical gradient data since falling into the local optimum and the current gradient synthesis as the search direction. When the function jumps out of the local optimum, the ordinary gradient is used as the search direction again. The gradient expression of the momentum method in the kth sub-iteration is:

[0089]

[0090] Where, d k-1 is the gradient at the k-1 sub-iteration;

[0091] Definition rules for random perturbations, including the perturbation term ε on the flow f and the perturbation term ε to the minimum cost λ , use a constant value of 1 to introduce random perturbations near the local optimal solution; not all types m and paths p need to be perturbed when they fall into the local optimal solution; some people are already in equilibrium, and the gradient value is checked to determine whether adjustment is needed: if the gradient of type m and path p is significantly non-zero, it indicates imbalance, and a threshold ε4 is used to determine when the absolute value of the gradient exceeds ε4 to apply random perturbations, and for ε f The perturbation in, when a steeper local optimal solution requires a larger perturbation to cross the local optimal point, the perturbation size is adjusted according to the frequency of encountering the local optimal point, while ensuring that the flow is always positive and does not exceed the given number of people;

[0092] Therefore, when falling into the local optimum, the descent rule of G(f,λ) of the kth sub-iteration is:

[0093] f (j,k+1) =f (j,k) +ρ k d k +ε f ; (j,k+1) =λ (j,k) +ρ k d k +ε λ (39)

[0094] When the traveler's periodic travel behavior selection decision is obtained for a given carbon credit reward and punishment mechanism rule, it is input into the upper model, the objective function of the carbon credit reward and punishment mechanism model decided by the leader is calculated, and a new set of carbon credit reward and punishment mechanism rules are generated according to the constraints. If the carbon credit reward and punishment mechanism rules generated twice are the same, the algorithm terminates, otherwise the newly generated carbon credit reward and punishment mechanism rules are returned as input to the lower model to continue the iterative calculation.

[0095] Further, the steps of the algorithm are summarized as follows:

[0096] a. Initialization: Use the k shortest path algorithm to obtain the initial path set R under zero flow conditions, the number of main iterations j = 0, and use all-or-nothing allocation to obtain f (0) , (0) , f (0) , (0) The flow vector and the Lagrange multiplier vector corresponding to the flow when the main iteration number is 0 are given the main program convergence value ε1 and the equilibrium convergence value ε2;

[0097] b. Column generation: j=j+1; based on f (j) , (j) , f (j) , (j) is the flow vector and the Lagrange multiplier vector corresponding to the flow at the jth main iteration, and the new path candidate set R is obtained using the k shortest path method (j) ;

[0098] c. Determine whether the main program converges: For each m∈M 1 ∪M 2 , the cost of the corresponding shortest path is obtained by solving the shortest path if represents the Lagrange multiplier of the (n,m)th group of people flow in the jth main iteration, and adds the path r to the path set R; if for any m∈M 1 ∪M 2 , all satisfied Then terminate the program, otherwise enter the equilibrium step;

[0099] d. Balance: For each iteration k, if the function value G in the kth sub-iteration under the jth main iteration is (j,k) (f,λ) satisfies equation (37), then the momentum method is used to search the direction, and random perturbations are used to update f (j,k) , (j,k) Otherwise, use the gradient method to search for the direction and update f using the Armijo step size (j,k) , (j,k) ;

[0100] e. Determine whether the equalization procedure has converged: For each iteration k, determine whether G (j,k) (f,λ)≤ε2, if the condition is met, return to the column generation step, otherwise re-enter the balancing step;

[0101] f. Solving the model of carbon credit reward and punishment mechanism for leader decision-making: Input the flow vector f of travelers at this time, calculate the multi-objective function of reducing emissions and congestion, and solve the new carbon credit reward and punishment mechanism rules in combination with the constraints;

[0102] g. Determine whether the newly generated carbon credit reward and punishment mechanism rules are the same as those in the previous round. If they are the same, the entire algorithm terminates and outputs the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision; otherwise, return to the column generation step.

[0103] The second object of the present invention is achieved by the following technical solution: a multi-modal travel structure optimization system for people with heterogeneous time values, used to implement the multi-modal travel structure optimization method for people with heterogeneous time values, comprising:

[0104] The extended clock network is used to describe the travel behavior selection decision of travelers in a cycle and convert it into a specific path. The structure of the extended clock network is 3 columns and Γ rows, where Γ is the number of days in the cycle, each column corresponds to the three travel modes of driving, online car-hailing and public transportation, and each row corresponds to each day in the cycle. The rows and columns determine the travel mode selected by the traveler on a certain day. The extended clock network uses each day as a decision point. In order to ensure the effectiveness of the extended clock network path, virtual links are added between various travel modes every day. The purpose of establishing the extended clock network is to accurately represent the travel behavior selection decision of travelers every day.

[0105] The carbon credit reward and punishment mechanism rule module is used to establish the carbon credit reward and punishment mechanism rules, the purpose of which is to encourage travelers to turn to the public transportation mode by combining reward and punishment measures; the carbon credit reward and punishment mechanism rules are that travelers can choose one of the travel modes of driving, public transportation, and online car-hailing every day during the cycle, and their carbon credit accounts will gain points or deduct corresponding points. At the end of the cycle, travelers with positive carbon credit accounts will be rewarded, and travelers with negative carbon credit accounts will pay fines; according to the obtained path of the travelers in the extended clock network, the path is calculated according to the daily travel mode to calculate the carbon credit account surplus at the end of the cycle, and the carbon credit cost of the cycle is obtained;

[0106] The multi-modal travel behavior selection decision model discretizes travelers into different types of travelers according to their time value. Time value measures the value that different travelers can create per hour, which is related to wages. After discretizing the population according to time value, the number of travelers of each type is obtained, and then the decision of each type of traveler is described, that is, the travel mode of each day in the cycle. The model goal is to minimize the total travel cost of all travelers in a cycle according to the decision of travelers, including time cost, monetary cost and carbon credit cost calculated based on the rules of carbon credit reward and punishment mechanism. Since the decisions of travelers are independent, the objective function of the multi-modal travel behavior selection decision model is obtained by linearly superimposing the objective function of each traveler, and the decision of each traveler on the travel mode of each day in a cycle is output;

[0107] The online car-hailing market equilibrium model is used to describe the number of online car-hailing drivers and passenger waiting time when the online car-hailing market is in equilibrium under given online car-hailing market parameters, and then calculate the time cost and monetary cost of online car-hailing travel every day;

[0108] The carbon credit reward and punishment mechanism model for leader decision-making is a two-level optimization model consisting of an upper and lower model, which is used to study how leaders design carbon credit reward and punishment mechanism rules under different carbon emission targets; the upper model takes different carbon emission targets as the objective function, and considers Pareto improvement and income neutrality to restrict the output of carbon credit reward and punishment mechanism rules. Pareto improvement means that after the implementation of the carbon credit reward and punishment mechanism rules, the total travel cost of a traveler in one cycle is less than that without the implementation of the carbon credit reward and punishment mechanism rules; the lower model is a multi-modal travel behavior selection decision model and an online car-hailing market equilibrium model, and an algorithm calculation is designed to initialize the lower model to output the traveler's periodic travel behavior selection decision and the online car-hailing market operation status, and use it as the input of the upper model to solve the upper model output as the carbon credit reward and punishment mechanism rules, and then input the lower model to cycle until the objective function values ​​of the upper models do not change twice before terminating the algorithm, and finally output the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision and the online car-hailing market operation status.

[0109] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0110] 1. The present invention adopts an extended clock network to describe the periodic behavior decisions of travelers, and accurately obtains the travel modes chosen by various groups of people every day.

[0111] 2. The present invention takes into account the impact of the online car-hailing market equilibrium on the carbon credit reward and punishment mechanism, and works with leaders in the online car-hailing market to regulate and optimize travelers' travel mode choices to reduce transportation carbon emissions.

[0112] 3. The present invention designs a corresponding k shortest path method for solving the shortest path problem in an extended clock network, which effectively solves the shortest path, that is, the periodic travel behavior decision of the traveler.

[0113] 4. The present invention uses momentum method and random perturbation to overcome the problem of local optimality, and has a good effect in solving the algorithm of expanding the clock network.

[0114] 5. The present invention establishes a two-layer optimization model and a multi-objective function to solve the travel market in which leaders and travelers work together. Leaders flexibly adjust the rules of the carbon credit reward and punishment mechanism based on the carbon emissions and congestion index that need to be controlled at present. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] Figure 1 Schematic diagram of the expanded clock network.

[0116] Figure 2 Schematic diagram of the system of the present invention. DETAILED DESCRIPTION

[0117] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0118] Example 1

[0119] This embodiment discloses a multi-mode travel structure optimization method for people with heterogeneous time values, including the following steps:

[0120] 1) Establish an extended clock network to describe the travel behavior selection decision of travelers in a cycle and convert it into a specific path; the structure of the extended clock network is 3 columns and Γ rows, Γ is the number of days in the cycle, each column corresponds to the three travel modes of driving, online car-hailing and public transportation, each row corresponds to each day in the cycle, and the rows and columns determine the travel mode selected by the traveler on a certain day; the extended clock network uses each day as a decision point. In order to ensure the effectiveness of the extended clock network path, virtual links are added between various travel modes every day. The purpose of establishing the extended clock network is to accurately represent the travel behavior selection decision of travelers every day;

[0121] like Figure 1 As shown in the figure, for each traveler, he needs to complete a trip from the starting point set O to the destination point set D every day. The only transportation modes available to travelers are driving, online car-hailing and bus. There are 3 columns and Γ rows in the extended clock network, Γ is the number of periodic days, and the 1st, 2nd and 3rd columns correspond to the travel modes of driving, online car-hailing and bus respectively; the i-th row corresponds to the i-th day, and the travel mode of the day is determined by the rows and columns together. All selected nodes are connected to obtain the periodic travel path; let a be the link selected by the travel mode, a∈A0 represents the set of travelers entering the extended clock network, and a∈A1 represents the choice of driving travel The set of links is represented by the link connecting any two nodes in the first column in the extended clock network; a∈A2 represents the set of links for choosing online car-hailing travel, which is represented by the link connecting any two nodes in the second column in the extended clock network; a∈A3 represents the set of links for choosing public transportation travel, which is represented by the link connecting any two nodes in the third column in the extended clock network; a∈A4 represents the set of links for switching between various travel modes, which is a virtual link that does not require a travel cost, and is represented by the link connecting different travel modes in the same row in the extended clock network; a∈A5 represents the set of travelers leaving the extended clock network.

[0122] 2) Establishing carbon credit reward and punishment mechanism rules, the purpose is to encourage travelers to turn to public transportation mode by combining reward and punishment measures; the carbon credit reward and punishment mechanism rules are that travelers can choose one of the travel modes of driving, public transportation, and online car-hailing every day during the cycle, and their carbon credit accounts will gain points or deduct corresponding points. At the end of the cycle, travelers with positive carbon credit accounts will be rewarded, and travelers with negative carbon credit accounts will pay fines; according to the path of the travelers in the expanded clock network obtained in step 1), the path is calculated according to the daily travel mode. The surplus of the carbon credit account at the end of the cycle is calculated to obtain the carbon credit cost of the cycle;

[0123] The carbon credit reward and penalty mechanism is implemented on a Γ-day cycle, where I is a set of days, I = {1, 2, ..., Γ}. At the beginning of the cycle, the amount of points added or deducted for the three travel modes per day is predetermined (κ a , κ b , κ c ) and the amount of reward or punishment the traveler receives after the cycle ends (p c ,p r ), as follows:

[0124] All travelers will get a carbon credit account with an initial balance of zero at the beginning of the cycle, and a κ a Every day when you take a ride-hailing car, your account will increase or deduct k b For the online car-hailing mode, the goal is to encourage people to switch to online car-hailing or to switch to public transportation. Therefore, κ b Can be positive or negative, the account increases by κ every day by taking public transportation c At the end of the period, if the traveler's carbon credit account is positive, the traveler will receive p r If the carbon credit account is negative, the traveler will be rewarded at p c The penalty rate is paid, for (κ a , κ b , κ c ) The size relationship between the three is unlimited, and they can be equal or unequal. For (p c ,p r )Requirement|p c |>|p r |;

[0125] Since the positive and negative relationship of the carbon credit account is uncertain at the end of the period, it cannot be directly expressed by a constraint equation, so an auxiliary function is introduced and The carbon integral cost CCR corresponding to the path r of the extended clock network in one cycle of the traveler r , if the balance of the traveler's carbon credit account at the end of the period is equal to If the balance of the traveler's carbon credit account is negative at the end of the period, then equation (2) is equal to

[0126]

[0127] In the formula, δ ar Indicates whether the driving route is selected, the value is 1 if selected, and 0 if not selected; δ br Indicates whether the route of the online car-hailing trip is selected, the value is 1 if selected, and the value is 0 if not selected; δ cr Indicates whether the bus travel route is selected, the value is 1 if selected, and 0 if not selected.

[0128] 3) Establish a multi-modal travel behavior selection decision model and an online car-hailing market equilibrium model; the multi-modal travel behavior selection decision model first discretizes travelers into different types of travelers according to their time value. Time value measures the value that different travelers can create per hour, which is related to wages. After discretizing the population according to time value, the number of travelers of each type is obtained, and then the decision of each type of traveler, that is, the travel mode of each day in the cycle is described. The model goal is to minimize the total travel cost of all travelers in a cycle according to the decision of the travelers, including time cost, monetary cost and carbon credit cost calculated based on the carbon credit reward and punishment mechanism rules in step 2). Since the decisions of travelers are independent, the objective function of the multi-modal travel behavior selection decision model is obtained by linearly superimposing the objective function of each traveler, and the decision of each traveler's travel mode every day in a cycle is output; the online car-hailing market equilibrium model describes the number of online car-hailing drivers and passenger waiting time when the online car-hailing market is in equilibrium under given online car-hailing market parameters, and then calculates the time cost and monetary cost of online car-hailing travel every day;

[0129] The specific situation of establishing a multi-modal travel behavior choice decision model is as follows:

[0130] The crowd is divided into car owners and car-free people. Car owners can choose to travel by car, online car-hailing or public transportation, while car-free people can only choose online car-hailing or public transportation. The car owners and car-free people are discretized separately, and the set M is used to represent the car owners and car-free people. 1 ={1,2,...,m 1} and M 2 ={1,2,...,m 2} indicates that m 1 and m 2 They represent the discretized population categories of car-owning and car-free populations, and are discretized with τ as the length of each interval according to the log-normal distribution curve. The average value of each interval is taken as the VOT of the discrete interval, denoted as β n,m, where n = 1 represents the car-owning population, n = 2 represents the car-free population, m represents the VOT category to which each sub-population belongs, and the probability density function describing VOT is recorded as f(x), x∈[a0,b0], x represents the time value, a0, b0 represent the left endpoint and right endpoint of the above log-normal distribution curve respectively; the total number of car-owning and car-free populations N 1 、N 2 is given, record N n,m is the number of people corresponding to the (n,m) group of people;

[0131]

[0132] β n,m =a0+τm-τ / 2 (5)

[0133] After defining the expression of discrete population, the number of people in the (n, m)th category who choose the travel mode on the i-th day is defined as a=1 means driving, a=2 means online car-hailing, and a=3 means public transportation;

[0134] For time cost and monetary cost, remember is the sum of the time cost and monetary cost of the travel mode selected by the (n,m)th person on the i-th day; in the extended clock network, the time cost of the neighboring node is calculated by multiplying the time value by the time spent, and the monetary cost is given directly; the time spent on the public transportation mode is t transit Due to the high punctuality and consistency of bus schedules, regardless of road traffic, the monetary cost of bus travel c transit Fixed, so the sum of the time cost and monetary cost of the (n,m)th person taking public transportation on the i-th day is expressed as:

[0135]

[0136] For the driving mode, the time spent is related to the road traffic, using the function Indicates that the road flow on the i-th day includes the total number of people traveling by car And the number of drivers entering the online car-hailing market u i , ph is defined as the peak hour number and takes 2 hours, and the time spent on driving travel mode on the i-th day is obtained as t 1,i The calculation formula is:

[0137]

[0138] Monetary cost of driving car is fixed, and the sum of the time cost and monetary cost of the (n,m)th person driving on the i-th day is expressed as:

[0139]

[0140] The time spent on online car-hailing travel includes driving time and waiting time. The driving time is equal to the driving time on the same day. represents the waiting time of passengers on the i-th day, u i represents the number of drivers entering the online car-hailing market, c ride Denotes the monetary cost of taking an online ride-hailing car every day, and the expression for the sum of the time cost and monetary cost of the travel mode chosen by the (n,m)th person on the i-th day is:

[0141]

[0142] For all travelers, each person's goal is to minimize the generalized travel cost within a cycle, without considering that the change of their own travel behavior choice decision will increase the travel cost of other travelers. Therefore, the objective function of the multi-modal travel behavior choice decision model is to minimize the periodic generalized total travel cost of all travelers. The decisions between each person do not affect each other, so the travel costs of travelers of different (n, m) types are linearly superimposed; the objective functions of the car group and the unmanned car group in the multi-modal travel behavior choice decision model are given below:

[0143] For car owners, the objective function is expressed as formula (10), where R 1 It is defined as the set of all cycle paths selected by the car-owning group in the extended clock network. Defined as the (1,m)th type of traffic flow that chooses path r:

[0144]

[0145] For people without cars, the objective function is expressed as formula (11), where R 2 It is defined as the set of all periodic paths chosen by the car-free population in the extended clock network. Defined as the (2,m)th type of traffic that chooses path r:

[0146]

[0147] Therefore, for the entire multi-modal travel behavior selection decision model, the objective function is to minimize the sum of equations (10) and (11):

[0148]

[0149] The constraints are the constraints on the flow, i.e., formulas (13) to (17), and the constraints on the rules of the carbon credit reward and punishment mechanism in step 2), i.e., formula (2):

[0150]

[0151] The specific situation of establishing the equilibrium model of the online car-hailing market is as follows:

[0152] Considering the static online car-hailing market model, the number of waiting passengers N on the i-th day is i c The passenger arrival rate Q on day i i The average waiting time of passengers on day i The number of empty vehicles N in the i-th i vt Arrival rate of vehicles in the sky through the i-th The average idle time of drivers per hour on day i The product of is approximated, assuming that the online ride-hailing platform matches waiting passengers with the nearest available vehicles and the online ride-hailing market is in a steady state. and the number of empty vehicles Remains unchanged within a day; the platform's hourly successful matching rate is recorded as m c-vt , which is determined by the number of waiting passengers and the number of empty vehicles on the i-th node, the matching rate is described by formula (18):

[0153]

[0154] Where A is a proportional parameter used to capture the external features that affect the matching results, which is the market area S divided by the running speed v of the empty vehicle. vt Indicates A;

[0155] When the online car-hailing market is in equilibrium, the hourly passenger arrival rate, hourly empty vehicle arrival rate, and hourly online car-hailing platform successful matching rate are equal, and the number of online car-hailing vehicles occupied per hour is The hourly passenger arrival rate Q i The average driving time per order of online car-hailing drivers t a,i , The product of i∈I gives: Therefore, the number of drivers entering the online car-hailing market per hour on the i-th day is u i It is the sum of the number of occupied online-hailing vehicles and the number of empty vehicles, expressed as:

[0156]

[0157] On the labor supply side, the number of drivers entering the online car-hailing market per hour on day i is u i It is not fixed. Whether a driver is willing to enter the online car-hailing market and provide services depends on his / her hourly reservation value r′, that is, the minimum hourly wage that potential online car-hailing owners are willing to provide services. Due to the heterogeneity of drivers, their reservation values ​​r′ are different. We use uniform distribution to describe the reservation value of a group of drivers, and assume that r′~U(0,μ m), U(0,μ m ) means obeying (0,μ m ), μ m is the maximum retention value of the driver group; γ is defined as the commission of the online car-hailing platform, and the net profit η per driver per hour is expressed as follows:

[0158]

[0159] Assume that the maximum number of drivers entering the online car-hailing market per hour is u max For each driver, only when the net profit per hour η is greater than the reserved value r′ will the driver enter the market, that is, η ≥ r′. Define g(·) as the uniform distribution probability density function, and use the following formula to express the number of drivers entering the online car-hailing market every hour per day u i :

[0160]

[0161] On the demand side, the demand for online ride-hailing vehicles on day i is Q i ′ is the result of the traveler's choice based on the minimization of the periodic generalized travel cost:

[0162]

[0163] 4) Establish a carbon credit reward and punishment mechanism model for leader decision-making. The carbon credit reward and punishment mechanism model for leader decision-making is a two-layer optimization model composed of an upper and lower model, which is used to study how leaders design carbon credit reward and punishment mechanism rules under different carbon emission targets; the upper model takes different carbon emission targets as the objective function, and considers Pareto improvement and income neutrality to limit the output of carbon credit reward and punishment mechanism rules. Pareto improvement means that after the implementation of the carbon credit reward and punishment mechanism rules, the total travel cost of a traveler in one cycle is less than that without the implementation of the carbon credit reward and punishment mechanism rules; the lower model is the multi-modal travel behavior selection decision model and the online car-hailing market equilibrium model proposed in step 3), and an algorithm is designed to calculate and initialize the lower model to output the traveler's periodic travel behavior selection decision and the online car-hailing market operation status, and use it as the input of the upper model to solve the upper model output as the carbon credit reward and punishment mechanism rules, and then input the lower model to cycle until the objective function values ​​of the upper models twice before and after do not change, then terminate the algorithm, and finally output the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision and the online car-hailing market operation status.

[0164] Establish a carbon credit reward and punishment mechanism model for leader decision-making, and obtain the carbon credit reward and punishment mechanism rules (κ a , κ b , κ c ,p c ,p r) and the corresponding traveler’s periodic travel behavior selection decision;

[0165] The carbon credit reward and punishment mechanism rules are used to achieve two goals. On the one hand, it is hoped that road congestion can be reduced by reducing the volume of vehicles on the road. On the other hand, it is hoped that the carbon emissions from road traffic can be reduced by reducing the number of travelers driving and promoting the shift to travel modes including online car-hailing and public transportation. In order to comprehensively consider the impact of the above two goals on the optimization of the carbon credit reward and punishment mechanism rules, the objective function is described by a multi-objective function and the objective weight factor ρ is introduced. ρ∈(0,1) is used to express the importance of the goal of reducing congestion, and 1-ρ∈(0,1) is used to express the importance of the goal of reducing traffic carbon emissions.

[0166] The congestion reduction target is measured by a road traffic congestion index, which is calculated by dividing the actual traffic volume of a road by its maximum capacity:

[0167]

[0168] For the reduction of transportation carbon emissions, the continuous function E of the COPERT emission function is used. a (·) calculation, E a (·) is a function of the average travel time t a,i The carbon emission E on the i-th day is obtained by the following formula: i , where l is defined a is the actual road distance in km:

[0169]

[0170] The objective function of the carbon credit reward and punishment mechanism model for leader decision-making is:

[0171]

[0172] The constraints of the carbon credit reward and punishment mechanism model for leader decision-making include revenue neutrality constraints and Pareto improvement constraints. The revenue neutrality constraint ensures that the income or subsidy obtained by the leader in each cycle remains within a controlled range; since the carbon credit reward and punishment mechanism rules are not intended to generate profits, limited subsidies are allowed; the revenue neutrality constraint strictly limits the income or subsidy generated by the carbon credit reward and punishment mechanism rules to B = 0, but in actual situations, the threshold is relaxed and defined as: -ε≤B≤ε, where B represents the net income or subsidy of the carbon credit reward and punishment mechanism rules in one cycle, represents the number of (n,m)-th type travelers who choose path r, ε is a constant, and the income neutrality constraint is expressed as:

[0173]

[0174] When implementing the carbon credit reward and punishment mechanism rules, the welfare of travelers needs to be improved, that is, the total travel cost of the cycle after the implementation of the carbon credit reward and punishment mechanism rules for (n, m) type travelers must be lower than the cost when the carbon credit reward and punishment mechanism rules are not implemented. The carbon credit reward and punishment mechanism rules that achieve this constraint are called Pareto improvements, which are expressed as formula (27), where It represents the sum of the time cost and monetary cost of the (n,m)th person who chooses the appearance mode on the i-th day without implementing the carbon credit reward and punishment mechanism rules:

[0175]

[0176] Design algorithm calculation, initialize the lower model to output travelers' periodic travel behavior choice decisions and online car-hailing market operations, and use them as the input of the upper model. Solve the upper model to obtain the output of the carbon credit reward and punishment mechanism rule setting, and then input the lower model cycle until the objective function value of the upper model twice does not change, then terminate the algorithm, and finally output the carbon credit reward and punishment mechanism rule setting, travelers' periodic travel behavior choice decisions, and online car-hailing market operations.

[0177] In order to solve the multi-modal travel behavior selection decision model in step 3), the Lagrange multiplier is introduced to transform it into a complementary relaxation problem, where Z(f) is the objective function of the multi-modal travel behavior selection decision model established in step 3), f and λ are vector groups composed of flow and Lagrange multipliers respectively, and the vector λ 1,m , 2,m denote the Lagrange multiplier corresponding to the flow of the mth group of people with cars and the Lagrange multiplier corresponding to the flow of the mth group of people without cars, respectively, and denote the Lagrange multiplier corresponding to equations (16) and (17); from an economic point of view, λ 1 and λ 2 They represent the minimum travel cost of the car group and the car-free group in the equilibrium state, λ 1 and λ 2 are all positive numbers strictly greater than 0, L(f,λ) represents the Lagrangian function related to f and λ;

[0178]

[0179] When the above equation (28) satisfies the KKT condition, it is converted into the following equivalent problem:

[0180]

[0181] In the formula, represents the total cycle cost of the path currently selected by the mth group of people with cars, It represents the total cycle cost of the path currently selected by the mth group of people without cars.

[0182] The main steps of the algorithm for solving the multi-modal travel behavior choice decision model are as follows:

[0183] a. Initialize the parameters of each variable and construct the initial solution by finding the shortest path and performing all-or-nothing allocation. All-or-nothing allocation means loading the traffic of each initial group into the extended clock network according to the shortest path obtained above. The main iteration number j = 0;

[0184] b. Based on the initial solution, generate a new path through the column generation method, iterate it, and evaluate the convergence, j = j + 1;

[0185] c. Allocate traffic on the new path until the multi-modal travel behavior choice decision model reaches an equilibrium state;

[0186] For the column generation method in step b, a new shortest path for each group of people needs to be obtained. The carbon credit cost as a cycle cost cannot be directly expressed as a point-to-point cost. It is processed by using the k shortest path method on the extended clock network. Since the carbon credit cost will also affect the cost between nodes, directly using the extended clock network that only contains the time and monetary costs between directly connected nodes will lead to inaccurate results. Therefore, the input extended clock network is modified as follows: For the driving node, in addition to the time cost and monetary cost, a virtual daily carbon credit cost κ is added a ×p c ; For bus nodes, subtract a virtual daily carbon credit cost κ c ×p r ; For online car-hailing nodes, the virtual carbon credit cost is adjusted according to the carbon credit reward and punishment mechanism rules. If κ b >0, subtract κ b ×p r , if κ b If <0, add κ b ×p c ;

[0187] For each group of people, the shortest path is obtained according to the above-mentioned extended clock network and Dijkstra algorithm is applied. Each node on the generated shortest path is regarded as a deviation node. The path from the starting point to the deviation node is fixed, and the subsequent path is recalculated to generate a new candidate path until k paths are generated;

[0188] For step c, define a function G(f,λ) related to (f,λ), where Ω(a,b) is the constructed auxiliary function, which is specifically expressed as:

[0189]

[0190] There is a mathematical property for the function G(f,λ): there is always a set of (f,λ) such that G(f,λ)=0, and when G(f,λ)=0 is the global function minimum, so the algorithm aims to make G(f,λ)=0, which is achieved by gradient descent method and search step size, but due to the non-convexity of the objective function, it will face local optimality during the descent process, making it difficult for the function to continue to descend. Therefore, when the function falls into the local optimal solution, the momentum method and random perturbation are introduced to make the function jump out of the local optimal solution and continue to descend; for the judgment of the local optimal solution, the judgment method used is: if the absolute value of the difference of G(f,λ) in five consecutive times is less than ε3, ε3 is a number close to 0, then it is considered that it is trapped in the local optimal solution at this time. The mathematical expression is as follows, where f (j,k+n) , (j,k+n) They are the flow vector and Lagrange multiplier vector corresponding to the sub-iteration number k+n under the main iteration number j:

[0191] |G(f (j,k+n) ,λ (j,k+n) )-G(f (j,k) ,λ (j,k) )|≤ε3 (34)

[0192] The gradient d of the function G(f,λ) at the kth sub-iteration k The expression is as follows (35). Note that p and q are used to distinguish the expressions of r and m in calculating the gradient. In fact, they still represent the corresponding paths and populations. (j) , (j) ,j=0,1,2,... represents the flow vector of the path selected by each group at the jth main iteration and the Lagrange multiplier vector f and λ corresponding to the flow; f (j,k) , (j,k) ,j=0,1,2,...,k=0,1,2,... represents the flow vector of the path selected by each group in the kth sub-iteration under the jth main iteration and the Lagrange multiplier vector f and λ corresponding to the flow, They represent the number of people in the qth category who choose path p in the kth sub-iteration under the jth main iteration, and the number of people in the qth category who choose path p in the car-carrying and car-free groups, respectively. They represent the Lagrange multipliers of the qth type of people in the car-owning and car-free populations corresponding to the kth sub-iteration under the jth main iteration:

[0193]

[0194] The function G(f,λ) decreases by a step size ρ at the kth sub-iteration k is the maximum value that satisfies the following inequality, c is a constant approaching 0, Represents the transpose of the gradient matrix of the function G for the flow f and the Lagrange multiplier λ at the kth sub-iteration under the jth main iteration:

[0195]

[0196] When there is no local optimal descent, the descent rule of G(f,λ) for the kth sub-iteration is:

[0197] f (j,k+1) =f (j,k) +ρ k d k ; (j,k+1) =λ (j,k) +ρ k d k (37)

[0198] In the formula, f (j,k+1) , (j,k+1) represents the flow f and Lagrange multiplier of the k+1th sub-iteration under the jth main iteration;

[0199] When falling into a local optimum, the gradient obtained by the momentum method is used as the function descent search direction. The momentum method accumulates the historical gradient data since falling into the local optimum and the current gradient synthesis as the search direction. When the function jumps out of the local optimum, the ordinary gradient is used as the search direction again. The gradient expression of the momentum method in the kth sub-iteration is:

[0200]

[0201] Where, d k-1 is the gradient at the k-1 sub-iteration;

[0202] Definition rules for random perturbations, including the perturbation term ε on the flow f and the perturbation term ε to the minimum cost λ , use a constant value of 1 to introduce random perturbations near the local optimal solution; not all types m and paths p need to be perturbed when they fall into the local optimal solution; some people are already in equilibrium, and the gradient value is checked to determine whether adjustment is needed: if the gradient of type m and path p is significantly non-zero, it indicates imbalance, and a threshold ε4 is used to determine when the absolute value of the gradient exceeds ε4 to apply random perturbations, and for ε f The perturbation in, when a steeper local optimal solution requires a larger perturbation to cross the local optimal point, the perturbation size is adjusted according to the frequency of encountering the local optimal point, while ensuring that the flow is always positive and does not exceed the given number of people;

[0203] Therefore, when falling into the local optimum, the descent rule of G(f,λ) of the kth sub-iteration is:

[0204] f (j,k+1) =f (j,k) +ρ k dk +ε f ;λ( (j,k+1) =λ (j,k) +ρ k d k +ε λ (39)

[0205] When the traveler's periodic travel behavior selection decision is obtained for a given carbon credit reward and punishment mechanism rule, it is input into the upper model, the objective function of the carbon credit reward and punishment mechanism model decided by the leader is calculated, and a new set of carbon credit reward and punishment mechanism rules are generated according to the constraints. If the carbon credit reward and punishment mechanism rules generated twice are the same, the algorithm terminates, otherwise the newly generated carbon credit reward and punishment mechanism rules are returned as input to the lower model to continue the iterative calculation.

[0206] Finally, the main steps of the algorithm are summarized as follows:

[0207] a. Initialization: Use the k shortest path algorithm to obtain the initial path set R under zero flow conditions, the number of main iterations j = 0, and use all-or-nothing allocation to obtain f (0) , (0) , f (0) , (0) The flow vector and the Lagrange multiplier vector corresponding to the flow when the main iteration number is 0 are given the main program convergence value ε1 and the equilibrium convergence value ε2;

[0208] b. Column generation: j=j+1; based on f (j) , (j) , f (j) , (j) is the flow vector and the Lagrange multiplier vector corresponding to the flow at the jth main iteration, and the new path candidate set R is obtained using the k shortest path method (j) ;

[0209] c. Determine whether the main program converges: For each m∈M 1 ∪M 2 , the cost of the corresponding shortest path is obtained by solving the shortest path if represents the Lagrange multiplier of the (n,m)th group of people flow in the jth main iteration, and adds the path r to the path set R; if for any m∈M 1 ∪M 2 , all satisfied Then terminate the program, otherwise enter the equilibrium step;

[0210] d. Balance: For each iteration k, if the function value G in the kth sub-iteration under the jth main iteration is (j,k)(f,λ) satisfies equation (37), then the momentum method is used to search the direction, and random perturbations are used to update f (j,k) , (j,k) Otherwise, use the gradient method to search for the direction and update f using the Armijo step size (j,k) , (j,k) ;

[0211] e. Determine whether the equalization procedure has converged: For each iteration k, determine whether G (j,k) (f,λ)≤ε2, if the condition is met, return to the column generation step, otherwise re-enter the balancing step;

[0212] f. Solving the model of carbon credit reward and punishment mechanism for leader decision-making: Input the flow vector f of travelers at this time, calculate the multi-objective function of reducing emissions and congestion, and solve the new carbon credit reward and punishment mechanism rules in combination with the constraints;

[0213] g. Determine whether the newly generated carbon credit reward and punishment mechanism rules are the same as those in the previous round. If they are the same, the entire algorithm terminates and outputs the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision; otherwise, return to the column generation step.

[0214] Example 2

[0215] This embodiment discloses a multi-mode travel structure optimization system for people with heterogeneous time values, which is used to implement the multi-mode travel structure optimization method for people with heterogeneous time values ​​described in Example 1. Figure 2 As shown, including:

[0216] The extended clock network is used to describe the travel behavior selection decision of travelers in a cycle and convert it into a specific path. The structure of the extended clock network is 3 columns and Γ rows, where Γ is the number of days in the cycle, each column corresponds to the three travel modes of driving, online car-hailing and public transportation, and each row corresponds to each day in the cycle. The rows and columns determine the travel mode selected by the traveler on a certain day. The extended clock network uses each day as a decision point. In order to ensure the effectiveness of the extended clock network path, virtual links are added between various travel modes every day. The purpose of establishing the extended clock network is to accurately represent the travel behavior selection decision of travelers every day.

[0217] The carbon credit reward and punishment mechanism rule module is used to establish the carbon credit reward and punishment mechanism rules, the purpose of which is to encourage travelers to turn to the public transportation mode by combining reward and punishment measures; the carbon credit reward and punishment mechanism rules are that travelers can choose one of the travel modes of driving, public transportation, and online car-hailing every day during the cycle, and their carbon credit accounts will gain points or deduct corresponding points. At the end of the cycle, travelers with positive carbon credit accounts will be rewarded, and travelers with negative carbon credit accounts will pay fines; according to the obtained path of the travelers in the extended clock network, the path is calculated according to the daily travel mode to calculate the carbon credit account surplus at the end of the cycle, and the carbon credit cost of the cycle is obtained;

[0218] The multi-modal travel behavior selection decision model discretizes travelers into different types of travelers according to their time value. Time value measures the value that different travelers can create per hour, which is related to wages. After discretizing the population according to time value, the number of travelers of each type is obtained, and then the decision of each type of traveler is described, that is, the travel mode of each day in the cycle. The model goal is to minimize the total travel cost of all travelers in a cycle according to the decision of travelers, including time cost, monetary cost and carbon credit cost calculated based on the rules of carbon credit reward and punishment mechanism. Since the decisions of travelers are independent, the objective function of the multi-modal travel behavior selection decision model is obtained by linearly superimposing the objective function of each traveler, and the decision of each traveler on the travel mode of each day in a cycle is output;

[0219] The online car-hailing market equilibrium model is used to describe the number of online car-hailing drivers and passenger waiting time when the online car-hailing market is in equilibrium under given online car-hailing market parameters, and then calculate the time cost and monetary cost of online car-hailing travel every day;

[0220] The carbon credit reward and punishment mechanism model for leader decision-making is a two-level optimization model consisting of an upper and lower model, which is used to study how leaders design carbon credit reward and punishment mechanism rules under different carbon emission targets; the upper model takes different carbon emission targets as the objective function, and considers Pareto improvement and income neutrality to restrict the output of carbon credit reward and punishment mechanism rules. Pareto improvement means that after the implementation of the carbon credit reward and punishment mechanism rules, the total travel cost of a traveler in one cycle is less than that without the implementation of the carbon credit reward and punishment mechanism rules; the lower model is a multi-modal travel behavior selection decision model and an online car-hailing market equilibrium model, and an algorithm calculation is designed to initialize the lower model to output the traveler's periodic travel behavior selection decision and the online car-hailing market operation status, and use it as the input of the upper model to solve the upper model output as the carbon credit reward and punishment mechanism rules, and then input the lower model to cycle until the objective function values ​​of the upper models do not change twice before terminating the algorithm, and finally output the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision and the online car-hailing market operation status.

[0221] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A multi-modal travel structure optimization method for people with heterogeneous time values, characterized in that: The following steps are involved: 1) Establish an extended clock network to describe the travel behavior selection decision of travelers in a cycle and convert it into a specific path; the structure of the extended clock network is 3 columns and Γ rows, Γ is the number of days in the cycle, each column corresponds to the three travel modes of driving, online car-hailing and public transportation, each row corresponds to each day in the cycle, and the rows and columns determine the travel mode selected by the traveler on a certain day; the extended clock network uses each day as a decision point. In order to ensure the effectiveness of the extended clock network path, virtual links are added between various travel modes every day. The purpose of establishing the extended clock network is to accurately represent the travel behavior selection decision of travelers every day; 2) Establishing carbon credit reward and punishment mechanism rules, the purpose is to encourage travelers to turn to public transportation mode by combining reward and punishment measures; the carbon credit reward and punishment mechanism rules are that travelers can choose one of the travel modes of driving, public transportation, and online car-hailing every day during the cycle, and their carbon credit accounts will gain points or deduct corresponding points. At the end of the cycle, travelers with positive carbon credit accounts will be rewarded, and travelers with negative carbon credit accounts will pay fines; according to the path of the travelers in the expanded clock network obtained in step 1), the path is calculated according to the daily travel mode. The surplus of the carbon credit account at the end of the cycle is calculated to obtain the carbon credit cost of the cycle; 3) Establish a multi-modal travel behavior selection decision model and an online car-hailing market equilibrium model; the multi-modal travel behavior selection decision model first discretizes travelers into different types of travelers according to their time value. Time value measures the value that different travelers can create per hour, which is related to wages. After discretizing the population according to time value, the number of travelers of each type is obtained, and then the decision of each type of traveler, that is, the travel mode of each day in the cycle is described. The model goal is to minimize the total travel cost of all travelers in a cycle according to the decision of the travelers, including time cost, monetary cost and carbon credit cost calculated based on the carbon credit reward and punishment mechanism rules in step 2). Since the decisions of travelers are independent, the objective function of the multi-modal travel behavior selection decision model is obtained by linearly superimposing the objective function of each traveler, and the decision of each traveler's travel mode every day in a cycle is output; the online car-hailing market equilibrium model describes the number of online car-hailing drivers and passenger waiting time when the online car-hailing market is in equilibrium under given online car-hailing market parameters, and then calculates the time cost and monetary cost of online car-hailing travel every day; 4) Establish a carbon credit reward and punishment mechanism model for leader decision-making. The carbon credit reward and punishment mechanism model for leader decision-making is a two-layer optimization model composed of an upper and lower model, which is used to study how leaders design carbon credit reward and punishment mechanism rules under different carbon emission targets; the upper model takes different carbon emission targets as the objective function, and considers Pareto improvement and income neutrality to limit the output of carbon credit reward and punishment mechanism rules. Pareto improvement means that after the implementation of the carbon credit reward and punishment mechanism rules, the total travel cost of a traveler in one cycle is less than that without the implementation of the carbon credit reward and punishment mechanism rules; the lower model is the multi-modal travel behavior selection decision model and the online car-hailing market equilibrium model proposed in step 3), and an algorithm is designed to calculate and initialize the lower model to output the traveler's periodic travel behavior selection decision and the online car-hailing market operation status, and use it as the input of the upper model to solve the upper model output as the carbon credit reward and punishment mechanism rules, and then input the lower model to cycle until the objective function values ​​of the upper models do not change twice before terminating the algorithm, and finally output the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision and the online car-hailing market operation status.

2. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 1 is characterized in that: In step 1), an extended clock network is established to describe the traveler's periodic travel behavior selection decision and converted into a concretely representable path, as follows: For each traveler, he needs to complete a trip from the starting point set O to the destination point set D every day. The only transportation modes available to travelers are driving, online car-hailing and bus. In the extended clock network, there are 3 columns and Γ rows, Γ is the number of periodic days, and the 1st, 2nd and 3rd columns correspond to the travel modes of driving, online car-hailing and bus respectively; the i-th row corresponds to the i-th day, and the travel mode of the day is determined by the rows and columns together. All selected nodes are connected to obtain the periodic travel path; let a be the link selected by the travel mode, a∈A1 represents the link set for choosing driving, which is represented by the link connecting any two nodes in the first column in the extended clock network; a∈A2 represents the link set for choosing online car-hailing, which is represented by the link connecting any two nodes in the second column in the extended clock network; a∈A3 represents the link set for choosing bus travel, which is represented by the link connecting any two nodes in the third column in the extended clock network; a∈A4 represents the link set for converting between various travel modes, which is a virtual link that does not require a travel cost, and is represented by the link connecting different travel modes in the same row in the extended clock network.

3. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 2 is characterized in that: In step 2), the carbon credit reward and penalty mechanism is implemented with a period of Γ days, where I is a set of days, I = {1, 2, ..., Γ}. At the beginning of the period, the amount of points added or deducted for the three travel modes per day is predetermined (κ a ,κ b ,κ c ) and the amount of reward or punishment the traveler receives after the cycle ends (p c ,p r ), as follows: All travelers will get a carbon credit account with an initial balance of zero at the beginning of the cycle, and a κ a Every day when you take a ride-hailing car, your account will increase or deduct k b For the online car-hailing mode, the goal is to encourage people to switch to online car-hailing or to switch to public transportation. Therefore, κ b Can be positive or negative, the account increases by κ every day by taking public transportation c At the end of the period, if the traveler's carbon credit account is positive, the traveler will receive p r If the carbon credit account is negative, the traveler will be rewarded at p c The penalty rate is paid, for (κ a , κ b , κ c ) The size relationship between the three is unlimited, and they can be equal or unequal. For (p c ,p r )Requirement|p c |>|p r |; Since the positive and negative relationship of the carbon credit account is uncertain at the end of the period, it cannot be directly expressed by a constraint equation, so an auxiliary function is introduced and The carbon integral cost CCR corresponding to the path r of the extended clock network in one cycle of the traveler r , if the balance of the traveler's carbon credit account at the end of the period is equal to If the balance of the traveler's carbon credit account is negative at the end of the period, then equation (2) is equal to In the formula, δ ar Indicates whether the driving route is selected, the value is 1 if selected, and 0 if not selected; δ br Indicates whether the route of the online car-hailing trip is selected, the value is 1 if selected, and the value is 0 if not selected; δ cr Indicates whether the bus travel route is selected, the value is 1 if selected, and 0 if not selected.

4. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 3 is characterized in that: In step 3), the specific situation of establishing the multi-modal travel behavior selection decision model is as follows: The crowd is divided into car owners and car-free people. Car owners can choose to travel by car, online car-hailing or public transportation, while car-free people can only choose online car-hailing or public transportation. The car owners and car-free people are discretized separately, and the set M is used to represent the car owners and car-free people. 1 ={1,2,...,m 1 } and M 2 ={1,2,...,m 2 } indicates that m 1 and m 2 They represent the discretized population categories of car-owning and car-free populations, and are discretized with τ as the length of each interval according to the log-normal distribution curve. The average value of each interval is taken as the VOT of the discrete interval, denoted as β n,m , where n = 1 represents the car-owning population, n = 2 represents the car-free population, m represents the VOT category to which each sub-population belongs, and the probability density function describing VOT is recorded as f(x), x∈[a0,b0], x represents the time value, a0, b0 represent the left endpoint and right endpoint of the above log-normal distribution curve respectively; the total number of car-owning and car-free populations N 1 、N 2 is given, record N n,m is the number of people corresponding to the (n,m) group of people; b n,m =a0+τm-τ / 2 (5) After defining the expression of discrete population, the number of people in the (n, m)th category who choose the travel mode on the i-th day is defined as a=1 means driving, a=2 means online car-hailing, and a=3 means public transportation; For time cost and monetary cost, remember is the sum of the time cost and monetary cost of the travel mode chosen by the (n,m)th person on the i-th day; In the extended clock network, the time cost of the neighboring nodes is calculated by multiplying the time value by the time spent, and the monetary cost is given directly; the time spent in the public transportation mode is t transit Due to the high punctuality and consistency of bus schedules, regardless of road traffic, the monetary cost of bus travel c transit Fixed, so the sum of the time cost and monetary cost of the (n,m)th person taking public transportation on the i-th day is expressed as: For the driving mode, the time spent is related to the road traffic, using the function Indicates that the road flow on the i-th day includes the total number of people traveling by car And the number of drivers entering the online car-hailing market u i , ph is defined as the peak hour number and takes 2 hours, and the time spent on driving travel mode on the i-th day is obtained as t 1,i The calculation formula is: Monetary cost of driving car is fixed, and the sum of the time cost and monetary cost of the (n,m)th person driving on the i-th day is expressed as: The time spent on online car-hailing travel includes driving time and waiting time. The driving time is equal to the driving time on the same day. represents the waiting time of passengers on the i-th day, u i represents the number of drivers entering the online car-hailing market, c ride Denotes the monetary cost of taking an online ride-hailing car every day, and the expression for the sum of the time cost and monetary cost of the travel mode chosen by the (n,m)th person on the i-th day is: For all travelers, each person's goal is to minimize the generalized travel cost within a cycle, without considering that the change of their own travel behavior choice decision will increase the travel cost of other travelers. Therefore, the objective function of the multi-modal travel behavior choice decision model is to minimize the periodic generalized total travel cost of all travelers. The decisions between each person do not affect each other, so the travel costs of travelers of different (n, m) types are linearly superimposed; the objective functions of the car group and the unmanned car group in the multi-modal travel behavior choice decision model are given below: For car owners, the objective function is expressed as formula (10), where R 1 It is defined as the set of all cycle paths selected by the car-owning group in the extended clock network. Defined as the (1,m)th type of traffic flow that chooses path r: For people without cars, the objective function is expressed as formula (11), where R 2 It is defined as the set of all periodic paths chosen by the car-free population in the extended clock network. Defined as the (2,m)th type of traffic that chooses path r: Therefore, for the entire multi-modal travel behavior selection decision model, the objective function is to minimize the sum of equations (10) and (11): The constraints are the constraints on the flow, i.e., formulas (13) to (17), and the constraints on the rules of the carbon credit reward and punishment mechanism in step 2), i.e., formula (2):

5. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 4 is characterized by: In step 3), the specific situation of establishing the online car-hailing market equilibrium model is as follows: Considering the static online car-hailing market model, the number of waiting passengers on the i-th day is The passenger arrival rate Q on day i i The average waiting time of passengers on day i The number of empty vehicles in the ith Arrival rate of vehicles in the sky through the i-th The average idle time of drivers per hour on day i The product of is approximated, assuming that the online ride-hailing platform matches waiting passengers with the nearest available vehicles and the online ride-hailing market is in a steady state. and the number of empty vehicles Remains unchanged within a day; the platform's hourly successful matching rate is recorded as m c-vt , which is determined by the number of waiting passengers and the number of empty vehicles on the i-th node, the matching rate is described by formula (18): Where A is a proportional parameter used to capture the external features that affect the matching results, which is the market area S divided by the running speed v of the empty vehicle. vt Indicates A; When the online car-hailing market is in equilibrium, the hourly passenger arrival rate, hourly empty vehicle arrival rate, and hourly online car-hailing platform successful matching rate are equal, and the number of online car-hailing vehicles occupied per hour is The hourly passenger arrival rate Q i Average driving time per order with online ride-hailing drivers The product of is given by , therefore, the number of drivers entering the online car-hailing market per hour on the i-th day is u i It is the sum of the number of occupied online-hailing vehicles and the number of empty vehicles, expressed as: On the labor supply side, the number of drivers entering the online car-hailing market per hour on day i is u i It is not fixed. Whether a driver is willing to enter the online car-hailing market and provide services depends on his / her hourly reservation value r′, that is, the minimum hourly wage that potential online car-hailing owners are willing to provide services. Due to the heterogeneity of drivers, their reservation values ​​r′ are different. We use uniform distribution to describe the reservation value of a group of drivers, and assume that r′~U(0,μ m ), U(0,μ m ) means obeying (0,μ m ), μ m is the maximum retention value of the driver group; γ is defined as the commission of the online car-hailing platform, and the net profit η per driver per hour is expressed as follows: Assume that the maximum number of drivers entering the online car-hailing market per hour is u max For each driver, only when the net profit per hour η is greater than the reserved value r′ will the driver enter the market, that is, η ≥ r′. Define g(·) as the uniform distribution probability density function, and use the following formula to express the number of drivers entering the online car-hailing market every hour per day u i : On the demand side, the demand for online ride-hailing vehicles on day i is Q i ′ is the result of the traveler's choice based on the minimization of the periodic generalized travel cost:

6. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 5 is characterized in that: In step 4), a carbon credit reward and punishment mechanism model for leader decision-making is established, and the carbon credit reward and punishment mechanism rules (κ a , κ b , κ c ,p c ,p r ) and the corresponding traveler’s periodic travel behavior selection decision; The carbon credit reward and punishment mechanism rules are used to achieve two goals. On the one hand, it is hoped that road congestion can be reduced by reducing the volume of vehicles on the road. On the other hand, it is hoped that the carbon emissions from road traffic can be reduced by reducing the number of travelers driving and promoting the shift to travel modes including online car-hailing and public transportation. In order to comprehensively consider the impact of the above two goals on the optimization of the carbon credit reward and punishment mechanism rules, the objective function is described by a multi-objective function and the objective weight factor ρ is introduced. ρ∈(0,1) is used to express the importance of the goal of reducing congestion, and 1-ρ∈(0,1) is used to express the importance of the goal of reducing traffic carbon emissions. The congestion reduction target is measured by a road traffic congestion index, which is calculated by dividing the actual traffic volume of a road by its maximum capacity: For the reduction of transportation carbon emissions, the continuous function E of the COPERT emission function is used. a (·) calculation, E a (·) is a function of the average travel time t a,i The carbon emission E on the i-th day is obtained by the following formula: i , where l is defined a is the actual road distance in km: The objective function of the carbon credit reward and punishment mechanism model for leader decision-making is: The constraints of the carbon credit reward and punishment mechanism model for leader decision-making include revenue neutrality constraints and Pareto improvement constraints. The revenue neutrality constraint ensures that the income or subsidy obtained by the leader in each cycle remains within a controlled range; since the carbon credit reward and punishment mechanism rules are not intended to generate profits, limited subsidies are allowed; the revenue neutrality constraint strictly limits the income or subsidy generated by the carbon credit reward and punishment mechanism rules to B = 0, but in actual situations, the threshold is relaxed and defined as: -ε≤B≤ε, where B represents the net income or subsidy of the carbon credit reward and punishment mechanism rules in one cycle, represents the number of (n,m)-th type travelers who choose path r, ε is a constant, and the income neutrality constraint is expressed as: When implementing the carbon credit reward and punishment mechanism rules, the welfare of travelers needs to be improved, that is, the total travel cost of the cycle after the implementation of the carbon credit reward and punishment mechanism rules for (n,m) type travelers must be lower than the cost when the carbon credit reward and punishment mechanism rules are not implemented; The carbon credit reward and punishment mechanism rule that implements this constraint is called a Pareto improvement, which is expressed as formula (27), where It represents the sum of the time cost and monetary cost of the (n,m)th person who chooses the appearance mode on the i-th day without implementing the carbon credit reward and punishment mechanism rules: Design algorithm calculation, initialize the lower model to output travelers' periodic travel behavior choice decisions and online car-hailing market operations, and use them as the input of the upper model. Solve the upper model to obtain the output of the carbon credit reward and punishment mechanism rule setting, and then input the lower model cycle until the objective function value of the upper model twice does not change, then terminate the algorithm, and finally output the carbon credit reward and punishment mechanism rule setting, travelers' periodic travel behavior choice decisions, and online car-hailing market operations.

7. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 6 is characterized in that: In order to solve the multi-modal travel behavior selection decision model in step 3), the Lagrange multiplier is introduced to transform it into a complementary relaxation problem, where Z(f) is the objective function of the multi-modal travel behavior selection decision model established in step 3), f and λ are vector groups composed of flow and Lagrange multipliers respectively, and the vector λ 1,m , 2,m denote the Lagrange multiplier corresponding to the flow of the mth group of people with cars and the Lagrange multiplier corresponding to the flow of the mth group of people without cars, respectively, and denote the Lagrange multiplier corresponding to equations (16) and (17); from an economic point of view, λ 1 and λ 2 They represent the minimum travel cost of the car group and the car-free group in the equilibrium state, λ 1 and λ 2 are all positive numbers strictly greater than 0, L(f,λ) represents the Lagrangian function related to f and λ; When the above equation (28) satisfies the KKT condition, it is converted into the following equivalent problem: In the formula, represents the total cycle cost of the path currently selected by the mth group of people with cars, It represents the total cycle cost of the path currently selected by the mth group of people without cars.

8. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 7 is characterized in that: The steps of the algorithm to solve the multimodal travel behavior choice decision model are as follows: a. Initialize the parameters of each variable and construct the initial solution by finding the shortest path and performing all-or-nothing allocation. All-or-nothing allocation means loading the traffic of each initial group into the extended clock network according to the shortest path obtained above. The main iteration number j = 0; b. Based on the initial solution, generate a new path through the column generation method, iterate it, and evaluate the convergence, j = j + 1; c. Allocate traffic on the new path until the multi-modal travel behavior choice decision model reaches an equilibrium state; For the column generation method in step b, a new shortest path for each group of people needs to be obtained. The carbon credit cost as a cycle cost cannot be directly expressed as a point-to-point cost. It is processed by using the k shortest path method on the extended clock network. Since the carbon credit cost will also affect the cost between nodes, directly using the extended clock network that only contains the time and monetary costs between directly connected nodes will lead to inaccurate results. Therefore, the input extended clock network is modified as follows: For the driving node, in addition to the time cost and monetary cost, a virtual daily carbon credit cost κ is added a ×p c ; For bus nodes, subtract a virtual daily carbon credit cost κ c ×p r ; For online car-hailing nodes, the virtual carbon credit cost is adjusted according to the carbon credit reward and punishment mechanism rules. If κ b >0, subtract κ b ×p r , if κ b If <0, add κ b ×p c ; For each group of people, the shortest path is obtained according to the above-mentioned extended clock network and Dijkstra algorithm is applied. Each node on the generated shortest path is regarded as a deviation node. The path from the starting point to the deviation node is fixed, and the subsequent path is recalculated to generate a new candidate path until k paths are generated; For step c, define a function G(f,λ) related to (f,λ), where Ω(a,b) is the constructed auxiliary function, which is specifically expressed as: There is a mathematical property for the function G(f,λ): there is always a set of (f,λ) such that G(f,λ)=0, and when G(f,λ)=0 is the global function minimum, so the algorithm aims to make G(f,λ)=0, which is achieved by gradient descent and search step size. However, due to the non-convexity of the objective function, it will face local optimality during the descent process, making it difficult for the function to continue to descend. Therefore, when the function falls into the local optimal solution, the momentum method and random perturbation are introduced to make the function jump out of the local optimal solution and continue to descend. For the judgment of the local optimal solution, the judgment method used is: if the absolute value of the difference of G(f,λ) in five consecutive times is less than ε3, ε3 is a number close to 0, then it is considered that it is trapped in the local optimal solution. The mathematical expression is as follows, where f (j,k+n) , (j,k+n) They are the flow vector and Lagrange multiplier vector corresponding to the sub-iteration number k+n under the main iteration number j: |G(f (j,k+n) ,l (j,k+n) )-G(f (j,k) ,l (j,k) )|≤ε3 (34) The gradient d of the function G(f,λ) at the kth sub-iteration k The expression is as follows (35). Note that p and q are used to distinguish the expressions of r and m in calculating the gradient. In fact, they still represent the corresponding paths and populations. (j) , (j) ,j=0,1,2,... represents the flow vector of the path selected by each group at the jth main iteration and the Lagrange multiplier vector f and λ corresponding to the flow; f (j,k) , (j,k) ,j=0,1,2,...,k=0,1,2,... represents the flow vector of the path selected by each group in the kth sub-iteration under the jth main iteration and the Lagrange multiplier vector f and λ corresponding to the flow, They represent the number of people in the qth category who choose path p in the kth sub-iteration under the jth main iteration, and the number of people in the qth category who choose path p in the car-carrying and car-free groups, respectively. They represent the Lagrange multipliers of the qth type of people in the car-owning and car-free populations corresponding to the kth sub-iteration under the jth main iteration: The function G(f,λ) decreases by a step size ρ at the kth sub-iteration k is the maximum value that satisfies the following inequality, c is a constant approaching 0, Represents the transpose of the gradient matrix of the function G for the flow f and the Lagrange multiplier λ at the kth sub-iteration under the jth main iteration: When there is no local optimal descent, the descent rule of G(f,λ) for the kth sub-iteration is: f (j,k+1) =f (j,k) +r k d k ;l (j,k+1) =λ (j,k) +r k d k (37) In the formula, f (j,k+1) , (j,k+1) represents the flow f and Lagrange multiplier of the k+1th sub-iteration under the jth main iteration; When falling into a local optimum, the gradient obtained by the momentum method is used as the function descent search direction. The momentum method accumulates the historical gradient data since falling into the local optimum and the current gradient synthesis as the search direction. When the function jumps out of the local optimum, the ordinary gradient is used as the search direction again. The gradient expression of the momentum method in the kth sub-iteration is: Where, d k-1 is the gradient at the k-1 sub-iteration; Definition rules for random perturbations, including the perturbation term ε on the flow f and the perturbation term ε to the minimum cost λ , use a constant value of 1 to introduce random perturbations near the local optimal solution; not all types m and paths p need to be perturbed when they fall into the local optimal solution; some people are already in equilibrium, and the gradient value is checked to determine whether adjustment is needed: if the gradient of type m and path p is significantly non-zero, it indicates imbalance, and a threshold ε4 is used to determine when the absolute value of the gradient exceeds ε4 to apply random perturbations, and for ε f The perturbation in, when a steeper local optimal solution requires a larger perturbation to cross the local optimal point, the perturbation size is adjusted according to the frequency of encountering the local optimal point, while ensuring that the flow is always positive and does not exceed the given number of people; Therefore, when falling into the local optimum, the descent rule of G(f,λ) of the kth sub-iteration is: f (j,k+1) =f (j,k) +r k d k +e f ;l (j,k+1) =λ (j,k) +r k d k +e λ (39) When the traveler's periodic travel behavior selection decision is obtained for a given carbon credit reward and punishment mechanism rule, it is input into the upper model, the objective function of the carbon credit reward and punishment mechanism model decided by the leader is calculated, and a new set of carbon credit reward and punishment mechanism rules are generated according to the constraints. If the carbon credit reward and punishment mechanism rules generated twice are the same, the algorithm terminates, otherwise the newly generated carbon credit reward and punishment mechanism rules are returned as input to the lower model to continue the iterative calculation.

9. The multi-modal travel structure optimization method for people with heterogeneous time values ​​according to claim 8 is characterized in that: The steps of the algorithm are summarized as follows: a. Initialization: Use the k shortest path algorithm to obtain the initial path set R under zero flow conditions, the number of main iterations j = 0, and use all-or-nothing allocation to obtain f (0) , (0) , f (0) , (0) The flow vector and the Lagrange multiplier vector corresponding to the flow when the main iteration number is 0 are given the main program convergence value ε1 and the equilibrium convergence value ε2; b. Column generation: j=j+1; based on f (j) , (j) , f (j) , (j) is the flow vector and the Lagrange multiplier vector corresponding to the flow at the jth main iteration, and the new path candidate set R is obtained using the k shortest path method (j) ; c. Determine whether the main program converges: For each m∈M 1 ∪M 2 , the cost of the corresponding shortest path is obtained by solving the shortest path if represents the Lagrange multiplier of the (n,m)th group of people flow in the jth main iteration, and adds the path r to the path set R; if for any m∈M 1 ∪M 2 , all satisfied Then terminate the program, otherwise enter the equilibrium step; d. Balance: For each iteration k, if the function value G in the kth sub-iteration under the jth main iteration is (j,k) (f,λ) satisfies equation (37), then the momentum method is used to search the direction, and random perturbations are used to update f (j,k) , (j,k) Otherwise, use the gradient method to search for the direction and update f using the Armijo step size (j,k) , (j,k) ; e. Determine whether the equalization procedure has converged: For each iteration k, determine whether G (j,k) (f,λ)≤ε2, if the condition is met, return to the column generation step, otherwise re-enter the balancing step; f. Solving the model of carbon credit reward and punishment mechanism for leader decision-making: Input the flow vector f of travelers at this time, calculate the multi-objective function of reducing emissions and congestion, and solve the new carbon credit reward and punishment mechanism rules in combination with the constraints; g. Determine whether the newly generated carbon credit reward and punishment mechanism rules are the same as those in the previous round. If they are the same, the entire algorithm terminates and outputs the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision; otherwise, return to the column generation step.

10. A multi-modal travel structure optimization system for people with heterogeneous time values, characterized in that: The method for optimizing the multi-modal travel structure of a population with heterogeneous time value according to any one of claims 1 to 9 comprises: The extended clock network is used to describe the travel behavior selection decision of travelers in a cycle and convert it into a specific path. The structure of the extended clock network is 3 columns and Γ rows, where Γ is the number of days in the cycle, each column corresponds to the three travel modes of driving, online car-hailing and public transportation, and each row corresponds to each day in the cycle. The rows and columns determine the travel mode selected by the traveler on a certain day. The extended clock network uses each day as a decision point. In order to ensure the effectiveness of the extended clock network path, virtual links are added between various travel modes every day. The purpose of establishing the extended clock network is to accurately represent the travel behavior selection decision of travelers every day. The carbon credit reward and punishment mechanism rule module is used to establish the carbon credit reward and punishment mechanism rules, the purpose of which is to encourage travelers to turn to the public transportation mode by combining reward and punishment measures; the carbon credit reward and punishment mechanism rules are that travelers can choose one of the travel modes of driving, public transportation, and online car-hailing every day during the cycle, and their carbon credit accounts will gain points or deduct corresponding points. At the end of the cycle, travelers with positive carbon credit accounts will be rewarded, and travelers with negative carbon credit accounts will pay fines; according to the obtained path of the travelers in the extended clock network, the path is calculated according to the daily travel mode to calculate the carbon credit account surplus at the end of the cycle, and the carbon credit cost of the cycle is obtained; The multi-modal travel behavior selection decision model discretizes travelers into different types of travelers according to their time value. Time value measures the value that different travelers can create per hour, which is related to wages. After discretizing the population according to time value, the number of travelers of each type is obtained, and then the decision of each type of traveler is described, that is, the travel mode of each day in the cycle. The model goal is to minimize the total travel cost of all travelers in a cycle according to the decision of travelers, including time cost, monetary cost and carbon credit cost calculated based on the rules of carbon credit reward and punishment mechanism. Since the decisions of travelers are independent, the objective function of the multi-modal travel behavior selection decision model is obtained by linearly superimposing the objective function of each traveler, and the decision of each traveler on the travel mode of each day in a cycle is output; The online car-hailing market equilibrium model is used to describe the number of online car-hailing drivers and passenger waiting time when the online car-hailing market is in equilibrium under given online car-hailing market parameters, and then calculate the time cost and monetary cost of online car-hailing travel every day; The carbon credit reward and punishment mechanism model for leader decision-making is a two-level optimization model consisting of an upper and lower model, which is used to study how leaders design carbon credit reward and punishment mechanism rules under different carbon emission targets; the upper model takes different carbon emission targets as the objective function, and considers Pareto improvement and income neutrality to restrict the output of carbon credit reward and punishment mechanism rules. Pareto improvement means that after the implementation of the carbon credit reward and punishment mechanism rules, the total travel cost of a traveler in one cycle is less than that without the implementation of the carbon credit reward and punishment mechanism rules; the lower model is a multi-modal travel behavior selection decision model and an online car-hailing market equilibrium model, and an algorithm calculation is designed to initialize the lower model to output the traveler's periodic travel behavior selection decision and the online car-hailing market operation status, and use it as the input of the upper model to solve the upper model output as the carbon credit reward and punishment mechanism rules, and then input the lower model to cycle until the objective function values ​​of the upper models do not change twice before terminating the algorithm, and finally output the carbon credit reward and punishment mechanism rules and the traveler's periodic travel behavior selection decision and the online car-hailing market operation status.

Images