A general approach to traffic and demand forecasting for shared mobility
By building traffic and demand prediction methods for traffic networks in shared travel environments, the problem that existing technology cannot predict traffic and demand for multiple types of shared traffic networks is solved, and accurate prediction of multiple types of shared traffic networks is achieved, and traffic planning is helped to formulate and management policies.
Patent Information
- Application Number
- CN202210588077.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-26
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-05-26
AI Technical Summary
The existing online car-hailing traffic and demand prediction methods cannot predict the traffic and demand of multiple types of shared transportation networks at the same time, and the accuracy of the results depends on a large amount of historical data and cannot be applied to multiple types of shared transportation networks.
A method for traffic network traffic and demand prediction in a shared travel environment was established. Through topological structure, constraint relationship and generalized random ride-sharing user equilibrium problems, a variational inequality model with elastic demand was constructed, and Monte Carlo simulation and weighted cost average were used. Calculate with the projection method to obtain traffic network traffic and residents' travel needs.
It realizes traffic and demand forecasts for multiple types of shared transportation networks, and provides a general method that can help enterprises and institutions accurately predict urban traffic status and formulate appropriate traffic planning plans and management policies.
Smart Images

Figure CN115062825B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for predicting flow and demand, and in particular to a general method for predicting flow and demand of shared travel. Background Art
[0002] A method for predicting online car-hailing traffic based on deep learning (application number CN202011455308.7) establishes a bidirectional long short-term memory network model with historical traffic volume in the past 8-24 hours as input to predict the online car-hailing traffic data for the next hour, effectively improving the accuracy of traffic prediction in some special circumstances (such as extreme weather, holidays, large-scale events, etc.). A method and device for predicting the supply and demand difference of online car-hailing based on multi-tenancy (application number CN202111388947.0) predicts orders and available capacity based on order data, driver location data, and driver status data in the target area, and then determines the supply and demand difference prediction results of the target area. These two inventions are flow prediction and demand prediction methods for shared travel modes, respectively, but the accuracy of the results depends on the input of a large amount of historical flow data, order data, location data, and status data of online car-hailing, and it is not possible to achieve simultaneous prediction of flow and demand of multiple types of shared transportation networks. Summary of the invention
[0003] The present invention establishes a general method for predicting the flow and demand of shared travel, which is used to predict the traffic network flow and residents' travel demand in a shared travel environment. Based on the urban traffic network in a shared travel environment, the present invention extracts the topological structure, establishes constraint relationships, constructs travel costs, and establishes an elastic demand (ED) function, abstracting the flow and demand prediction of the actual traffic network into a generalized stochastic ridesharing user equilibrium with elastic demand (general stochastic ridesharing user equilibrium with elastic demand, general SRUE-ED) problem, which can be used for multiple types of shared transportation users, multiple types of behavioral selection models, multiple types of demand models, and multiple types of shared transportation services. In response to this problem, the present invention constructs a parameterized fixed point model and a variational inequality (VI) model with a unique solution; and by establishing a Monte Carlo (MC) method, a weighted cost average (WCA) method, and a projection method, a globally convergent calculation method is proposed to obtain the traffic network flow and residents' travel demand.
[0004] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0005] A general method for predicting the flow and demand of shared travel includes the following steps:
[0006] S1. Obtain the topological structure and travel mode of the urban transportation network;
[0007] S2. Establish traffic flow conservation constraints and ride-sharing matching constraints;
[0008] S3, establish a generalized travel cost function;
[0009] S4. Construct the generalized stochastic ridesharing user equilibrium problem;
[0010] S5. Establish an elastic demand function and construct a generalized random carpooling user equilibrium problem with elastic demand;
[0011] S6. Construct a fixed point model with parameters;
[0012] S7. Construct a variational inequality model equivalent to the generalized stochastic ridesharing user equilibrium problem;
[0013] S8. Introduce the projection method, Monte Carlo simulation method and weighted cost average method to design the calculation method of variational inequality model.
[0014] Furthermore, the topological structure and travel mode in S1 are:
[0015] Construct a shared transportation network G = (N, A), where N represents a node set, n represents a node, n∈N; A represents a road segment set, a represents a road segment, a∈A;
[0016] Travel demand occurs at the starting point and ends at the destination. A pair of starting and destination points with travel demand constitutes an OD pair in the shared transportation network, represented by w. The travel demand between OD pair w is q w ; Let W represent the set of OD pairs, w∈W;
[0017] The path is formed by connecting the first segments; let p∈P w represents a path connecting OD to w, P w represents the path set connecting OD pair w; P represents the path set, P = ∪ w P w ;
[0018] The travel mode set in the shared transportation network is I, which can be divided into three subsets: self-driving drivers, carpooling drivers (and carpooling passengers, represented by the sets SD, RD and R respectively, that is, I = SD∪RD∪R; i represents a certain travel mode, i∈I; the travel mode subsets SD, RD and R may include multiple travel modes, represented by the value of i; let i=1∈SD represent self-driving drivers; i=2,3∈RD represent carpooling drivers carrying 1 or 2 carpooling passengers respectively; i=4,5∈R represent carpooling passengers carried by carpooling drivers with i=2,3 respectively;
[0019] Let x a,i represents the flow of mode i on road section a, represents the traffic of mode i on path p between OD pair w; there is a relationship between the two:
[0020]
[0021] In the formula, represents the path-section conversion coefficient: if path p passes through section a, then otherwise make represents the path flow vector;
[0022] Let x a The traffic flow on section a is equal to the sum of the traffic flows of all drivers on the section, that is:
[0023]
[0024] Let x = (x a ,a∈A) T represents the road flow vector; t a =t a (x a ) = t a (x) represents the travel time on road section a, which is about x a Or a function of x; let represents the travel time on path p between OD pair w, which is calculated as:
[0025]
[0026] The supply and demand relationship in the ridesharing market is based on the travel demand between each OD pair; is the ride-sharing supply, is the demand for ridesharing, which is:
[0027]
[0028] Furthermore, the constraint relationship in S2 is:
[0029] Traffic Flow Conservation Constraints
[0030]
[0031] Rideshare matching constraints
[0032]
[0033] Where N i represents the number of carpooling passengers carried by carpooling driver i∈RD, It means mapping the carpooling driver i∈RD to the carpooling passenger i′∈R that he carries, that is, The shared transportation network includes N2 = 1 and N3=2, then the specific carpool matching constraints are:
[0034]
[0035] RD| represents the number of elements in the set RD, |RD| = 2; |P w | represents the set P w The number of elements in is the number of paths between OD pairs w; let the matrix A = [0, N, -1], the dimension is (|RD|·∑ w |P w |)×((2|RD|+1)·∑ w |P w |); 0 means RD|·∑ w |P w |)×1 zero vector; N and 1 are both (|RD|·∑ w |P w |)×(|RD|·∑ w |P w |) is a diagonal matrix, the diagonal elements are N i and 1; the vector expression of the co-occurrence matching constraint is:
[0036] Af=0 (8).
[0037] Furthermore, the generalized travel cost function in S3 is:
[0038] The path travel cost function of each travel mode in the shared transportation network is as follows:
[0039]
[0040] In the formula, represents the travel cost of mode i on path p between OD pairs w; The section flow x is calculated from the path flow f, and then the are calculated from the path flow f; so is a function of the path flow f; ρ i represents the time value of mode i, I represents the time cost of mode i on path p between OD pairs w; i (·) represents the inconvenience cost function of mode i, represents the inconvenience cost of OD for mode i on path p between w, which comes from the discomfort of sharing the space in the car with strangers and the path travel time is proportional to; and M i (·) represent the base price and floating price of OD to w ride-sharing drivers, represents the compensation received by the ride-sharing driver, represents the fare paid by the carpooling passenger; represents the miscellaneous costs of mode i on path p between OD pairs w, such as tolls, fuel costs, and waiting time for buses (passengers);
[0041]
[0042] According to the market clearing principle of economics, the additional cost of maintaining supply and demand balance also needs to be considered. In the formula, Lagrange multiplier representing the co-occurrence matching constraint; and The mapping relationship is opposite, that is and make represents the additional cost vector, represents the Lagrange multiplier vector;
[0043] Generalized travel cost function for shared transportation networks for:
[0044]
[0045]
[0046] make is the generalized travel cost vector between OD pairs w; represents the generalized travel cost vector.
[0047] Furthermore, the specific process of constructing the generalized stochastic ridesharing user equilibrium problem in S4 is as follows:
[0048] Since travelers’ perception of travel time is affected by many factors, such as their psychological state, travel purpose, and income level, but the price and reward of ride-sharing are accurately calculated by the App, the error between perceived travel cost and actual travel cost mainly comes from the randomness of perceived travel time;
[0049]
[0050] In the formula, represents the random perception error of mode i travel time on road segment a; Has the property of: expectation variance limited; It follows an independent distribution and its value is independent of the traffic; its probability density function is a finite positive number with a finite number of discontinuities; represents the perceived travel time of mode i on road segment a; let t = (t a ,a∈A) T represents the road segment travel time vector, represents the perceived travel time vector, represents the random perceptual error vector;
[0051] make represents the perceived travel cost of mode i on path p between w by OD, is the random perception error of mode i on path p between OD pairs w, then:
[0052]
[0053]
[0054] In the formula, are all random variables; for any OD pair w, represents the minimum perceived travel cost, which is also a random variable;
[0055] In summary, the equilibrium condition for generalized random carpooling is:
[0056]
[0057]
[0058]
[0059] In the formula, is the probability that a traveler between OD pair w chooses to travel in mode i on path p, and its value is related to the generalized travel cost between OD pairs There is a connection.
[0060] Furthermore, the specific process of constructing the generalized stochastic ridesharing user equilibrium problem with elastic demand in S5 is as follows:
[0061] make represents the generalized travel cost of OD for travelers between w The satisfaction function is as follows:
[0062]
[0063] OD travel demand q between w w There is an elastic demand function:
[0064]
[0065] Where D w (·) is about satisfaction S w A bounded, non-negative, non-increasing, continuously differentiable function of ; Represents travel demand q w The positive upper bound of w ,w∈W) T represents the satisfaction vector;
[0066] In summary, the equilibrium condition for generalized random carpooling users with elastic demand is:
[0067]
[0068]
[0069]
[0070]
[0071] because Then the satisfaction function can be written as The generalized stochastic ridesharing user equilibrium condition can be integrated as follows:
[0072]
[0073]
[0074]
[0075] Furthermore, since λ in the generalized random carpooling user equilibrium condition is a Lagrange multiplier corresponding to the carpooling matching constraint, according to the complementary relaxation theorem, if the value of λ is given in advance, the carpooling matching constraint is automatically satisfied; in addition, since And elastic demand D w (S w(f,λ)) and the probability of selection are greater than zero, so It is also automatically satisfied; if the Lagrange multiplier λ is given in advance, the generalized random carpooling user equilibrium problem can be expressed as the following fixed point model with parameters:
[0076]
[0077] Where λ is the parameter of the model, f(λ) represents the solution of the parametric fixed point model, and f(λ) is an implicit vector function about λ. Since λ is only a fixed parameter and will not change the solution of the parametric fixed point model, it is easy to prove that f(λ) is the only path flow solution of the parametric fixed point model.
[0078] Furthermore, the specific process of constructing variational inequalities in S7 is:
[0079] The variational inequality model is constructed based on the implicit vector function f(λ) in S6:
[0080] In the feasible set Ω λ Find a λ * ∈Ω λ , so that:
[0081]
[0082] In the formula, Ω λ is the feasible set of λ, M is a sufficiently large integer such that The absolute value of The upper and lower bounds can never be reached; it can be proved that the VI model has a unique solution λ * ,f(λ * ) satisfies the generalized random carpooling user equilibrium condition;
[0083] Solve the variational inequality model and obtain the unique Lagrange multiplier λ * ; Then apply the obtained λ * Solve the fixed point model with parameters and obtain the unique path flow solution f(λ * ).
[0084] Furthermore, the specific process of designing the calculation method in S8 is:
[0085] Calculation method f (k) Both represent the path flow vector of the kth iteration, which can be equivalent to f; λ (l) ,t (k) , T (n) , C (n) , C (m) , φ(m) , g (m) , γ (k) , β (k) , α (k) , τ (l) , S (0) , φ (0) , f (0) , γ (0) , β (0) , (l-1) , Symbols of similar forms are the same, indicating only the value of a variable at a certain iteration, and have the same meaning as variables without parenthetical superscripts (l), (k), (n), (m), (0), (l-1), (k-1), etc.;
[0086] Monte Carlo simulation method:
[0087] Part I:
[0088] S10, initialization; set the number of simulations n = 1, initial satisfaction The maximum acceptable error of satisfaction n ; Incoming path flow f (k) ; Call the projection method, by f (k) Calculate λ (l) ;
[0089] S11, Monte Carlo simulation; according to the path flow f (k) , calculate the travel time t of the road section (k) ; and generate random perception error ξ to obtain the perceived travel time vector T (n) =t (k) +ξ;
[0090] S12. Search for the shortest path; calculate generalized travel costs And find the minimum generalized travel cost between any OD pair w Path;
[0091] S13. Estimate satisfaction; the satisfaction is obtained by averaging the results of n simulations:
[0092]
[0093] S14, stop condition test; calculate the satisfaction function S of any OD for w w Variance
[0094]
[0095] Then calculate the maximum number of simulations n0:
[0096]
[0097] If n>n0, execute S15; otherwise, n:=n+1, execute S11.
[0098] S15. Calculate elastic demand; any OD pair w travel demand q w for:
[0099]
[0100] After the calculation is completed, execute S20;
[0101] Part II:
[0102] Monte Carlo simulation method:
[0103] S20, initialization; set the simulation number m = 1, initial auxiliary flow mode The maximum acceptable error of auxiliary path flow ε m ;
[0104] S21, Monte Carlo simulation; generate random perception error ξ, and obtain the perceived travel time vector T (m) =t (k) +ξ;
[0105] S22. All-or-nothing allocation; calculation of generalized travel costs Find any OD pair w with the minimum generalized travel cost The "path-mode combination" And the demand q between OD and w w All loaded in On the temporary auxiliary path flow i∈I; thus obtaining the temporary auxiliary path flow vector
[0106] S23, random network loading, estimation of auxiliary path flow; auxiliary path flow The average result of m simulations is obtained;
[0107]
[0108] S24, stop condition check; calculate the maximum number of simulation times m0:
[0109]
[0110] If m>m0, output auxiliary path flow vector Otherwise, m:=m+1, execute S21;
[0111] Weighted Cost Averaging Method:
[0112] S31, initialization; acceptable error ε>0, step size numerator γ (0) =0, step size denominator β (0) =0, number of iterations k = 1; set the initial path flow f (0) ; Calculate λ by projection method (l) ; Calculate generalized travel costs
[0113] S32, iterative process; according to Perform an all-or-nothing allocation to obtain f (k) ; Then replace the previous λ (l) With f (k) Substitute it and obtain the auxiliary path flow φ through the Monte Carlo simulation method (k) ; Update γ (k) , β (k) and step size α (k) :
[0114] γ (k) :=(k+1)ln(k+1)
[0115] β (k) :=β (k-1) +γ (k)
[0116]
[0117] Update generalized travel costs
[0118]
[0119] S33, check the stop condition; if Execute the Monte Carlo simulation method once and output
[0120] Otherwise, let k:=k+1 and execute S32;
[0121] Projection method:
[0122] S41, initialization; acceptable error κ>0, projection step length τ (0) >0, number of iterations l = 1; set the initial Lagrange multiplier λ (0) ; Incoming path flow f (k) ;
[0123] S42, iteration; given λ (l-1) , update τ (l) and λ (l) :
[0124] τ(l) :=1 / l
[0125] λ (l) : =λ (l-1) -τ (l) (Af (k) )
[0126] S43, check the stop condition; if Stop iteration and output λ (l) ; Otherwise, let l:=l+1 and execute S42.
[0127] Furthermore, the step size of the weighted cost averaging method in S8 adopts the following weighted values:
[0128] γ (k) :=(k+1)ln(k)+1)
[0129] β (k) :=β (k-1) +γ (k)
[0130]
[0131] Compared with the existing shared transportation network flow and demand forecasting methods under specific conditions, the present invention provides a general method for flow and demand forecasting that can be widely applied to various types of shared transportation networks, which can help relevant enterprises and institutions to more accurately and conveniently predict urban traffic conditions, thereby formulating appropriate transportation planning schemes and management policies. BRIEF DESCRIPTION OF THE DRAWINGS
[0132] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0133] Figure 1 (a) is the flow chart of the Monte Carlo simulation method;
[0134] (b) is a flow chart of the weighted average method;
[0135] (c) is a flowchart of the projection method;
[0136] (d) is the overall framework diagram of the calculation method;
[0137] Figure 2 is a flow chart of the method of the present invention;
[0138] Figure 3 Share the transportation network topology diagram for the case. DETAILED DESCRIPTION
[0139] The present invention will be further described below in conjunction with examples and accompanying drawings.
[0140] The present invention provides a general method for predicting the flow and demand of shared travel, such as Figure 1 As shown, the following process is included:
[0141] 1. Obtain the topological structure and travel patterns of the urban transportation network.
[0142] The topology of the example sharing transportation network is as follows Figure 3 As shown. The network has 24 nodes, 76 road segments and 528 OD pairs. Construct a shared transportation network G = (N, A), where N represents a node set, the number of nodes |N| = 24, n represents a node, n∈N; A represents a road segment set, the number of road segments |A| = 76, a represents a road segment, a∈A. Let W represent the set of OD pairs, the number of OD pairs |W| = 528, and any OD pair w∈W. For each OD pair of the case shared transportation network, select 10 paths, that is, for P w |=10.
[0143] The travel mode set in the shared transportation network is I, which can be divided into three subsets: solo drivers, carpool drivers (ride-sharing drivers) and carpool passengers (ride-sharing passengers), represented by the sets SD, RD and R respectively, that is, I = SD∪RD∪R. i represents a travel mode, i∈I. The travel mode subsets SD, RD and R may include multiple travel modes, represented by the value of i. Let i=1∈SD represent a solo driver; i=2,3∈RD represent carpool drivers carrying 1 or 2 carpool passengers respectively; i=4,5∈R represent carpool passengers carried by carpool drivers with i=2,3 respectively.
[0144] Let x a,i represents the flow of mode i on road section a, It represents the traffic of mode i on path p between OD pair w. There is a relationship between the two:
[0145]
[0146] In the formula, represents the path-section conversion coefficient: if path p passes through section a, then otherwise make represents the path flow vector.
[0147] Let x a The traffic flow on section a is equal to the sum of the traffic flows of all drivers on the section, that is:
[0148]
[0149] Let x = (x a ,a∈A) T represents the road flow vector; t a =t a (x a ) = t a (x) represents the travel time on road section a. In the case traffic network, the BPR function is used to calculate the travel time of the road section:
[0150]
[0151] In the formula, t a,0 is the free travel time of section a; c a is the traffic capacity of section a. a,0 and c a The values are shown in Table 1.
[0152] Table 1. Parameter values of road sections in the target city traffic network
[0153]
[0154]
[0155] make represents the travel time on path p between OD pair w, which is calculated as:
[0156]
[0157] The supply and demand relationship in the ride-sharing market is based on the travel demand between OD pairs. is the ride-sharing supply, is the demand for ridesharing, which is:
[0158]
[0159] 2. Establish traffic flow conservation constraints and ridesharing matching constraints.
[0160] (1) Traffic flow conservation constraints
[0161]
[0162] (2) Rideshare matching constraints
[0163]
[0164] Where N i represents the number of carpooling passengers carried by carpooling driver i∈RD, It means mapping the carpooling driver i∈RD to the carpooling passenger i′∈R that he carries, that is, The shared transportation network includes N2 = 1 and N3=2, then the specific carpool matching constraints are:
[0165]
[0166] The example shared traffic network has 528 OD pairs, and 10 paths are selected for each OD pair, that is, ∑ w |P w |=528×10=5280. Then the dimension of matrix A is (2×5280)×(5×5280)=10560×26400, 0 represents a 10560×5280 zero matrix, and N and 1 are both 10560×10560 diagonal matrices.
[0167] 3. Establish a generalized travel cost function.
[0168] In the case sharing transportation network, set γ i is the inconvenience coefficient. and Both or Both can be written m i is a floating pricing parameter. Then the generalized travel cost of the case sharing transportation network is:
[0169]
[0170] The travel cost parameter settings in the case sharing transportation network are shown in Table 2:
[0171] Table 2 Travel cost parameters of the case sharing transportation network
[0172]
[0173] 4. Construct the generalized random carpooling user equilibrium (generalSRUE) problem.
[0174] Random perception error of road sections in case sharing traffic network Follows Gumbel distribution It can be shown that the path random perception error Follows Gumbel distribution θ is the perception discrete parameter, θ = 0.05. It can be shown that the path and mode selection probabilities in the case sharing transportation network are based on the Logit behavior selection model. The generalSRUE condition of the network is:
[0175]
[0176]
[0177]
[0178] 5. Establish an elastic demand (ED) function and construct the generalized random ridesharing user equilibrium (generalSRUE-ED) problem with elastic demand.
[0179] make represents the generalized travel cost of OD for travelers between w The satisfaction function is as follows:
[0180]
[0181] The elastic demand function in the case sharing transportation network is in exponential form, specifically:
[0182]
[0183] In this network, μ is the demand elasticity parameter, μ = 0.05. For 528 OD pairs, the travel demand q w The positive upper bound of As shown in Table 3:
[0184] Table 3. The positive upper bound of travel demand in the case sharing transportation network
[0185]
[0186]
[0187]
[0188]
[0189]
[0190]
[0191]
[0192] because S w =S w (f,λ), the generalSRUE-ED condition of the case-sharing transportation network is:
[0193]
[0194]
[0195]
[0196] 6. Establish a fixed point model with parameters.
[0197] Since λ in the generalSRUE-ED condition is the Lagrange multiplier corresponding to the co-multiplication matching constraint, according to the complementary relaxation theorem, if the value of λ is given in advance, the co-multiplication matching constraint is automatically satisfied; in addition, since And elastic demand D w (S w (f,λ)) and the probability of selection are greater than zero, so is also automatically satisfied. Therefore, if the Lagrange multiplier λ is given in advance, the general SRUE-ED problem of the case-sharing transportation network can be expressed as the following parameterized fixed point model:
[0198]
[0199] Where λ is the parameter of the model, f(λ) represents the solution of the parametric fixed point model, and f(λ) is an implicit vector function about λ. Since λ is just a fixed parameter and does not change the solution of the parametric fixed point model, it is easy to prove that f(λ) is the only path flow solution of the parametric fixed point model.
[0200] 7. Construct a variational inequality model equivalent to the generalSRUE-ED problem.
[0201] The variational inequality (VI) model is constructed based on the implicit vector function f(λ):
[0202] In the feasible set Ω λ Find a λ * ∈Ω λ , so that:
[0203]
[0204] In the formula, Ω λ is the feasible set of λ, M is a sufficiently large integer such that The absolute value of The upper and lower bounds can never be reached. It can be proved that the VI model has a unique solution λ * ,f(λ * ) satisfies the generalSRUE-ED conditions.
[0205] Solve the VI model and obtain the unique Lagrange multiplier λ * ; Then apply the obtained λ * Solve the fixed point model with parameters and obtain the unique path flow solution f(λ * ).
[0206] 8. Introduce the projection method, Monte Carlo simulation method and weighted cost average method, and design the calculation method of the variational inequality model.
[0207] When calculating the path flow of the case-sharing transportation network, the maximum acceptable error ε of the satisfaction in the Monte Carlo simulation method n = 0.5, the maximum acceptable error of the auxiliary path flow vector The acceptable error of the weighted cost average method is ε = 10 -5 . Applying the proposed calculation method, after 99 iterations, the unique path flow solution f(λ * ). Multiplying the path flow by the “path-section correlation coefficient” can give the section flow result x=(x a ,a∈A) T , as shown in Table 4.
[0208] Table 4. Flow results of road sections in the case sharing transportation network
[0209]
[0210] Any OD demand for w w As shown in Table 5.
[0211] Table 5 OD travel demand of the case sharing transportation network
[0212]
[0213]
[0214]
[0215]
[0216] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention is described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A general method for traffic and demand forecasting in shared mobility, Features: The steps include: S1. Obtain the topological structure and travel mode of the urban transportation network; The topological structure and travel mode in S1 are: Construct a shared transportation network G = (N, A), where N represents a node set, n represents a node, n∈N; A represents a road segment set, a represents a road segment, a∈A; Travel demand occurs at the starting point and ends at the destination. A pair of starting and destination points with travel demand constitutes an OD pair in the shared transportation network, represented by w. The travel demand between OD pair w is q w ; Let W denote the set of OD pairs, w∈W; The path is formed by connecting the first segments; let p∈P w represents a path connecting OD to w, P w represents the path set connecting OD pair w; P represents the path set, P = ∪ w P w ; The travel mode set in the shared transportation network is I, which can be divided into three subsets: self-driving drivers, carpooling drivers and carpooling passengers, represented by the sets SD, RD and R respectively, that is, I = SD∪RD∪R; i represents a certain travel mode, i∈I; the travel mode subsets SD, RD and R may include multiple travel modes, represented by the value of i; Let i = 1∈SD represent a self-driving driver; i = 2, 3∈RD represent a carpooling driver carrying 1 or 2 carpooling passengers respectively; i = 4, 5∈R represent the carpooling passengers carried by the carpooling drivers i = 2, 3 respectively; Let x a,i represents the flow of mode i on road section a, represents the traffic of mode i on path p between OD pair w; there is a relationship between the two: In the formula, represents the path-section conversion coefficient: if path p passes through section a, then otherwise make represents the path flow vector; Let x a The traffic flow on section a is equal to the sum of the traffic flows of all drivers on the section, that is: Let x = (x a ,a∈A) T represents the road flow vector; t a =t a (x a ) = t a (x) represents the travel time on road section a, which is about x a Or a function of x; let represents the travel time on path p between OD pair w, which is calculated as: The supply and demand relationship in the ridesharing market is based on the travel demand between each OD pair; is the ride-sharing supply, is the demand for ridesharing, which is: S2. Establish traffic flow conservation constraints and ride-sharing matching constraints; The constraint relationship in S2 is: Traffic Flow Conservation Constraints Rideshare matching constraints Where N i represents the number of carpooling passengers carried by carpooling driver i∈RD, It means mapping the carpooling driver i∈RD to the carpooling passenger i′∈R that he carries, that is, The shared transportation network includes and The specific carpool matching constraints are: |RD| represents the number of elements in the set RD, |RD|=2; |P w | represents the set P w The number of elements in is the number of paths between OD pairs w; let the matrix A = [0, N, -1], the dimension is (|RD|·∑ w |P w |)×((2|RD|+1)·∑ w |P w |); 0 means (|RD|·∑ w |P w |)×1 zero vector; N and 1 are both (|RD|·∑ w |P w |)×(|RD|·∑ w |P w |) is a diagonal matrix, the diagonal elements are N i and 1; the vector expression of the co-occurrence matching constraint is: Af=0(8); S3, establish a generalized travel cost function; S4. Construct the generalized stochastic ridesharing user equilibrium problem; S5. Establish an elastic demand function and construct a generalized random carpooling user equilibrium problem with elastic demand; S6. Construct a fixed point model with parameters; S7. Construct a variational inequality model equivalent to the generalized stochastic ridesharing user equilibrium problem; S8. Introduce the projection method, Monte Carlo simulation method and weighted cost average method to design the calculation method of variational inequality model.
2. The general method for predicting the flow and demand of shared travel according to claim 1, Features: The generalized travel cost function in S3 is: The path travel cost function of each travel mode in the shared transportation network is as follows: In the formula, represents the travel cost of mode i on path p between OD pairs w; The section flow x is calculated from the path flow f, and then the are calculated from the path flow f; so is a function of the path flow f; ρ i represents the time value of mode i, I represents the time cost of mode i on path p between OD pairs w; i (·) represents the inconvenience cost function of mode i, represents the inconvenience cost of OD for mode i on path p between w, which comes from the discomfort of sharing the space in the car with strangers and the path travel time is proportional to; and M i (·) represent the base price and floating price of OD to w ride-sharing drivers, represents the compensation received by the ride-sharing driver, represents the fare paid by the carpooling passenger; represents the miscellaneous costs of mode i on path p between OD pairs w, including tolls, fuel costs, and waiting time; According to the market clearing principle of economics, the additional cost of maintaining supply and demand balance also needs to be considered. In the formula, Lagrange multiplier representing the co-occurrence matching constraint; and The mapping relationship is opposite, that is and make represents the additional cost vector, represents the Lagrange multiplier vector; Generalized travel cost function for shared transportation networks for: make is the generalized travel cost vector between OD pairs w; represents the generalized travel cost vector.
3. The general method for predicting traffic and demand of shared travel according to claim 1, Features: The specific process of constructing the generalized stochastic ridesharing user equilibrium problem in S4 is: Since travelers’ perception of travel time is affected by multiple factors, such as their psychological state, travel purpose, and income level, but the price and reward of ride-sharing are accurately calculated by the App, the error between perceived travel cost and actual travel cost mainly comes from the randomness of perceived travel time; In the formula, represents the random perception error of mode i travel time on road segment a; Have nature expectations variance limited; It follows an independent distribution and its value is independent of the traffic; Its probability density function is a finite positive number with a finite number of discontinuities; represents the perceived travel time of mode i on road segment a; let w = (t a ,a∈A) T represents the road segment travel time vector, represents the perceived travel time vector, represents the random perceptual error vector; make represents the perceived travel cost of mode i on path p between w by OD, is the random perception error of mode i on path p between OD pairs w, then: In the formula, are all random variables; for any OD pair w, represents the minimum perceived travel cost, which is also a random variable; In summary, the equilibrium condition for generalized random carpooling is: In the formula, is the probability that a traveler between OD pair w chooses to travel in mode i on path p.
4. The general method for predicting traffic and demand of shared travel according to claim 1, Features: The specific process of constructing the generalized stochastic ridesharing user equilibrium problem with elastic demand in S5 is as follows: make represents the generalized travel cost of OD for travelers between w The satisfaction function is as follows: OD's travel demand for w is q w There is an elastic demand function: Where D w (·) is about satisfaction S w A bounded, non-negative, non-increasing, continuously differentiable function of ; Represents travel demand q w The positive upper bound of w ,w∈W) T represents the satisfaction vector; In summary, the equilibrium condition for generalized random carpooling users with elastic demand is: because Then the satisfaction function can be written as The generalized stochastic ridesharing user equilibrium condition is integrated as follows:
5. The general method for predicting traffic and demand of shared travel according to claim 1, Features: The specific process of constructing the fixed point model with parameters in S6 is as follows: Since λ in the generalized random carpooling user equilibrium condition is the Lagrange multiplier corresponding to the carpooling matching constraint, according to the complementary relaxation theorem, if the value of λ is given in advance, the carpooling matching constraint is automatically satisfied; in addition, since And elastic demand D w (S w (f,λ)) and the probability of selection are greater than zero, so It is also automatically satisfied; if the Lagrange multiplier λ is given in advance, the generalized random carpooling user equilibrium problem can be expressed as the following fixed point model with parameters: Where λ is the parameter of the model, f(λ) represents the solution of the parametric fixed point model, and f(λ) is an implicit vector function about λ. Since λ is only a fixed parameter and will not change the solution of the parametric fixed point model, it is easy to prove that f(λ) is the only path flow solution of the parametric fixed point model.
6. The general method for predicting traffic and demand of shared travel according to claim 1, Features: The specific process of constructing variational inequalities in S7 is: The variational inequality model is constructed based on the implicit vector function f(λ) in S6: In the feasible set Ω λ Find a λ * ∈Ω λ , so that: In the formula, Ω λ is the feasible set of λ, M is a sufficiently large integer such that The absolute value of The upper and lower bounds can never be reached; it can be proved that the VI model has a unique solution λ * ,f(λ * ) satisfies the generalized random carpooling user equilibrium condition; Solve the variational inequality model and obtain the unique Lagrange multiplier λ * ; Then apply the obtained λ * Solve the fixed point model with parameters and obtain the unique path flow solution f(λ * ).
7. The general method for predicting traffic and demand of shared travel according to claim 1, Features: The specific process of the design calculation method in S8 is: Calculation method f (k) Both represent the path flow vector of the kth iteration, which can be equivalent to f; λ (l) ,t (k) 、T (n) , C (n) , C (m) , φ (m) , g (m) , γ (k) , β (k) , α (k) , τ (l) , S (0) , φ (0) , f (0) , γ (0) , β (0) , (l-1) , Symbols of similar form are used in the same way, they only represent the value of the variable at a certain iteration, and have the same meaning as the variable without superscript; Monte Carlo simulation method: Part I: S10, initialization; set the number of simulations n = 1, initial satisfaction The maximum acceptable error of satisfaction n ; Incoming path flow f (k) ; Call the projection method, by f (k) Calculate λ (l) ; S11, Monte Carlo simulation; according to the path flow f (k) , calculate the travel time t of the road section (k) ; and generate random perception error ξ to obtain the perceived travel time vector T (n) =t (k) +ξ; S12. Search for the shortest path; calculate generalized travel costs And find the minimum generalized travel cost between any OD pair w Path; S13. Estimate satisfaction; the satisfaction is obtained by averaging the results of n simulations: S14, stop condition inspection; Calculate the random variable for any OD pair w Variance Then calculate the maximum number of simulations n 0 : If n>n 0 , then execute S15; otherwise, n:=n+1, execute S11; S15. Calculate elastic demand; any OD pair w travel demand q w for: After the calculation is completed, execute S20; Part II: S20, initialization; set the simulation number m = 1, initial auxiliary flow mode The maximum acceptable error of auxiliary path flow ε m ; S21, Monte Carlo simulation; generate random perception error ξ, and obtain the perceived travel time vector T (m) =t (k) +ξ; S22. All-or-nothing allocation; calculation of generalized travel costs Find any OD pair w with the minimum generalized travel cost The "path-mode combination" ; And the demand q of OD for w w All loaded in On the temporary auxiliary path flow ; Thus, the temporary auxiliary path flow vector is obtained ; S23, random network loading, estimation of auxiliary path flow; auxiliary path flow The average result of m simulations is obtained; S24, stop condition check; Calculate the maximum number of simulations m 0 : If m>m 0 , output auxiliary path flow vector ; Otherwise, m:=m+1, execute S21; Weighted Cost Averaging Method: S31, initialization; acceptable error ε>0, step size numerator γ (0) =0, step size denominator β (0) =0, number of iterations k = 1; Set the initial path flow f (0) ; Calculate λ by projection method (l) ; Calculate generalized travel costs S32, iterative process; according to Perform an all-or-nothing allocation to obtain f (k) ; Then replace the previous λ (l) With f (k) Substitute it and obtain the auxiliary path flow φ through the Monte Carlo simulation method (k) ; Update γ (k) , β (k) and step size α (k) : c (k) :=(k+1)ln(k+1) = (k) :=b (k-1) +g (k) Update generalized travel costs S33, check the stop condition; if Execute the Monte Carlo simulation method once and output Otherwise, let k:=k+1 and execute S32; Projection method: S41, initialization; acceptable error κ>0, projection step length τ (0) >0, number of iterations l = 1; set the initial Lagrange multiplier λ (0) ; Incoming path flow f (k) ; S42, iteration; given λ (l-1) , update τ (l) and λ (l) : t (l) :=1 / l λ (l) :=λ (l-1) -τ (l) (Of (k) ) S43, check the stop condition; if Stop iteration and output λ (l) ; Otherwise, let l:=l+1 and execute S42.
8. The general method for predicting traffic and demand of shared travel according to claim 1, Features: The step size of the weighted cost averaging method in S8 uses the following weighting values: c (k) :=(k+1)ln(k+1) b (k) :=b (k-1) +g (k)
