A method for predicting bus passenger flow

Abstract

The present invention discloses a method for predicting public transportation passenger flow, which comprises constructing an individual travel decision model based on individual travel data and public transportation vehicle operation data as input, then establishing motion rules for individual travel agents, establishing motion rules for public transportation vehicle agents and interaction rules between the two agents; by inputting actual characteristic information of the two agents and the environment, and simulating the actual state according to pre-set motion rules and interaction rules based on this information, thereby outputting the required passenger flow data and completing passenger flow prediction. The method can well simulate the actual travel behavior of public transportation individuals, can effectively reflect the operating status of urban public transportation systems, and plays an important role in accurately describing passenger travel behavior, exploring passenger flow change patterns, and analyzing passenger flow distribution characteristics.

Description

Technical Field

[0001] The present invention relates to the technical field of public transportation, and in particular to a method for predicting public transportation passenger flow. Background Art

[0002] To mitigate the negative impact of the rapid increase in private vehicles on urban roads, meet the transportation needs of urban residents as much as possible and ensure fairness in their travel, multimodal bus systems have rapidly become a key component of the comprehensive transportation system of large cities, playing an irreplaceable role in providing large-scale urban transportation services and alleviating urban traffic pressure.

[0003] The service capacity of public transportation is a key factor in judging the level of public transportation operations in a city. Therefore, continuously improving the public transportation service capacity of cities of different levels to meet the diverse travel needs of urban residents and improve the quality of residents' travel is an important task in the current construction of major cities. The reasonable planning of the public transportation network is one of the prerequisites for improving public transportation service capacity, and accurate prediction of public transportation passenger flow can provide an important basis for scientific planning of bus lines and networks, which is helpful for operational planning decisions such as timetable compilation.

[0004] Public transit passenger flow forecasting has long been a hot topic in transportation research, both domestically and internationally. Certain achievements have been made in bus trip OD estimation, bus route selection, and bus passenger flow allocation. Furthermore, with the development of emerging technologies such as computer technology and artificial intelligence, various methods are being continuously optimized and improved. However, there is still room for further research in urban public transit passenger flow forecasting methods.

[0005] In previous studies, most of the research objects have been concentrated on the comprehensive transportation system of large cities, and less consideration has been given to conventional ground public transportation in small and medium-sized cities. The travel of individual travelers in the public transportation network is affected by multiple factors such as their individual attributes, the operating status of public transportation vehicles, and the road conditions. Existing models for public transportation passenger flow prediction rarely consider individual travel characteristics and the operating status of public transportation vehicles. At the same time, at this stage, the research on public transportation passenger flow distribution based on multi-agent simulation technology is mainly focused on urban rail transit. There is little research on multi-agent simulation models for ground public transportation passenger flow, and previous studies have not fully considered the impact of real-time bus information on passengers' travel route selection. Therefore, a public transportation passenger flow prediction method is urgently needed to solve the above problems. Summary of the Invention

[0006] The present invention provides a public transport passenger flow prediction method, which can simulate the actual travel behavior of public transport passengers and effectively reflect the operating status of the urban public transport system.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for predicting public transportation passenger flow, comprising the following steps:

[0008] S1, obtained individual travel data and bus operation data as input;

[0009] S2, building an individual travel decision model;

[0010] S3. Based on the constructed individual travel decision model, establish the movement rules of the individual travel agent, the movement rules of the bus vehicle agent, and the interaction rules between the two;

[0011] S4. By inputting the actual characteristic information of the two intelligent agents and the environment, and simulating the actual state based on this information according to the pre-set movement rules and interaction rules, the required passenger flow data is output to complete the passenger flow prediction.

[0012] Preferably, in step S2, based on individual travel data, the individual's boarding and alighting station information is inferred based on the time matching method and travel chain theory, and according to the different functional levels of multimodal buses, the individual bus travel transfer behavior is analyzed, the individual multimodal bus travel chains in small and medium-sized cities are identified and divided, and the virtual OD pairs of the travel individuals are established to obtain complete multimodal bus individual travel data; then, based on the Markov decision process theory and the individual's historical travel data, a travel decision model for the multimodal bus individual is constructed.

[0013] Preferably, the individual bus travel data includes IC card swiping data, bus GPS data and bus route station location information, wherein the GPS data is supplemented by interpolation and matched with the station latitude and longitude to obtain the vehicle arrival time, and then matched with the passenger's IC card swiping time to identify the boarding station; different bus trip chains are divided according to the trip chain theory, and the inference process is simplified by combining passenger travel assumptions and bus station characteristics to obtain the alighting stations of different types of bus trip chains.

[0014] Preferably, the public transportation network that carries and limits the movement trajectory and range of the two types of intelligent agents is: PT-Network=<r′,s r ′,b r ′,w r ' -b ,t r ' -ser >, where r′ represents all routes included in the multimodal bus network, covering routes at different levels in the network; s r ′ represents the set of all line sites consisting of the sites included in each line; b r ′ represents the set of all line vehicles consisting of the operating vehicles contained in each line; w r ' -b Represents the set of all route trips consisting of the operating trips of each route; t r '-ser Represents the set of all line operating hours consisting of the operating hours of each line.

[0015] Preferably, the individual agent P-Agent is represented as: P-Agent=<(O,D...),S P ,A P ,R P ,F P >, where (O, D…) refers to the distribution set of the starting and ending points of the individual agent, including all the historical starting and ending points of the travel individual within a certain period of time; S P is the state set of the individual travelers; A P is the action set of the individual traveler; R P is the direct reward value set of individuals; F P is the set of individual future return values.

[0016] Preferably, the movement rules of individual agents are established:

[0017] Step 1: Determine the travel chains between each pair of virtual OD points of individual agents within a certain historical period, form a set of candidate paths, determine the OD points and departure times of the trips, import them into the public transportation network, optimize the travel paths, and input the selected paths into the NetLogo simulation platform.

[0018] Step 2: Determine whether the departure time has arrived, and then move to the boarding station at the walking speed. If the departure time has not arrived, continue to wait at the terminal;

[0019] Step 3: Wait in line at the boarding station for the pre-boarding vehicle. If the pre-boarding vehicle arrives first and is not full, board the vehicle. Otherwise, remain in the current state and continue waiting. If a vehicle on another candidate route arrives first and is not full before the individual boards, calculate the direct rewards and corresponding state transition probabilities for the current route and the pre-boarding route, and based on the calculated results, board the vehicle or remain in the current state and continue waiting.

[0020] Step 4: When in the vehicle and on the way, the individual agent determines the get-off station based on the determined travel route. If the individual agent reaches the get-off station at the time of getting off and does not need to transfer, the individual agent proceeds to the travel destination along the preset route. If the individual agent reaches the get-off station at the time of getting off but needs to transfer, the agent moves to the boarding station at walking speed. If the individual agent does not reach the get-off station at the time of getting off, the agent maintains the current state.

[0021] Preferably, the bus vehicle intelligent agent B-Agent is expressed as: B-Agent=<(r,b,di b ,s r ,t ser ),S B ,AB >, where (r,b,di b ,s r ,t ser ) describes the inherent properties of the bus agent, which represents the bus route, vehicle number, departure interval, set of passing stations and operating time period to which the B-Agent belongs; S B is the state set of the bus; A B A set of actions for buses.

[0022] Preferably, the motion rules of the bus agent are established:

[0023] Step 1: The bus will depart at the departure time according to the operating time period;

[0024] Step 2: After arriving at the stop, determine whether it is an intermediate stop. If so, calculate the stop duration based on the number of people getting on and off the bus simulated in the bus network; if not, calculate the stop duration directly based on the number of people getting off the bus;

[0025] Step 3: After the stop time is reached, the bus will depart for the next stop or the operation ends, and the vehicle will enter the bus station to queue up for the next bus to depart.

[0026] Preferably, the Ceder model is used to calculate the docking time:

[0027]

[0028]

[0029] in, Represents the stop time of a single-door boarding and alighting bus; Represents the stop time of a bus with separate boarding and alighting services; t b Represents basic docking time; Represents the average boarding time per person; represents the average time it takes to get off the bus; δ b Represents the number of people on board; δ a Represents the number of people getting off the bus.

[0030] Preferably, the formula for calculating the number of passengers that can be boarded is: N ap =N bap -N ip -N ap +N bp ; Among them, N ap Represents the number of people that can board; N bap Represents the vehicle's interior capacity; N ip Represents the number of people in the car; N ap Represents the number of people getting off the bus; N bpRepresents the number of people on board.

[0031] Compared with the existing technology, the beneficial effects of the present invention are as follows: by calculating the individual bus boarding and alighting stops to divide the travel chain and thus establish an individual travel decision model, and combining the state transition of bus trips with multi-agent simulation technology, a multi-modal bus recent passenger flow forecast driven by individual travel data is realized, which can well simulate the actual travel behavior of individual bus travelers and effectively reflect the operating status of the urban public transportation system. It plays an important role in accurately describing passenger travel behavior, exploring passenger flow change patterns, and analyzing passenger flow distribution characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0033] In the attached figure:

[0034] Figure 1 It is a flow chart of the public transportation passenger flow prediction method of the present invention;

[0035] Figure 2 is a graph of individual travel state transitions in the multimodal public transportation network of the present invention;

[0036] Figure 3 This is a flow chart of the movement rules of individual intelligent agents traveling in the present invention;

[0037] Figure 4 It is the state transition diagram of the bus intelligent body of the present invention;

[0038] Figure 5 This is a flow chart of the motion rules of the bus intelligent body of the present invention;

[0039] Figure 6 It is the overall flow chart of the simulation model of the present invention;

[0040] Figure 7 This is a schematic diagram of the bus route layout in a specific embodiment of the present invention;

[0041] Figure 8 It is the basic animation interface of the simulation platform of the present invention;

[0042] Figure 9 This is the interface for displaying the real-time number of people getting on and off at stations 1 to 4 of the present invention. DETAILED DESCRIPTION

[0043] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0044] Example: Figure 1 As shown, a method for predicting bus passenger flow includes the following steps:

[0045] S1, obtained individual travel data and bus operation data as input;

[0046] S2, building an individual travel decision model;

[0047] S3. Based on the constructed individual travel decision model, establish the movement rules of the individual travel agent, the movement rules of the bus vehicle agent, and the interaction rules between the two;

[0048] S4. By inputting the actual characteristic information of the two intelligent agents and the environment, and simulating the actual state based on this information according to the pre-set movement rules and interaction rules, the required passenger flow data is output to complete the passenger flow prediction.

[0049] In step S2, constructing an individual travel decision model includes:

[0050] Identification and estimation of individual boarding and alighting points for bus trips: Based on bus IC card swipe data and GPS data obtained from bus systems in small and medium-sized cities, sparse GPS data is interpolated and matched with the longitude and latitude of the bus stops to obtain the vehicle arrival time. The time matching method is then used to obtain the passenger's IC card swipe time and identify the boarding point. Different bus trip chains are divided according to trip chain theory. The estimation process is simplified by combining passenger travel assumptions and bus stop characteristics to obtain the alighting points for different types of bus trips. This completes the preliminary processing of individual trip data, obtains basic information about individual bus trips, and provides the initial basic data input for building a passenger flow prediction simulation model.

[0051] The boarding station identification is as follows:

[0052] Step 1: Select research data.

[0053] Select the bus route r to be studied and obtain any passing station s i The longitude and latitude (x i ,y i ), select a certain operating vehicle b on the route and a certain operating vehicle number w b GPS data and IC card data of any passenger p riding in the vehicle.

[0054] Step 2: Clear redundant data: Calculate the time interval between each card swipe record and the two previous and next card swipe records. p,k , if 0<IT p,k <L S,S+1 / v b (L S,S+1 is the distance between the current boarding station and the downstream adjacent station, calculated using the haversine formula; vb is the bus operating speed), it means that the card swiping time interval is short. It can be considered that passenger p has an accompanying person and shares an IC card with the accompanying person. These two card swiping times are regarded as one trip, and the time recorded earlier is taken as the current card swiping time;

[0055] Step 3: GPS data interpolation processing: If the bus GPS data is a sparse sample with low frequency return and has no correlation with the vehicle's arrival and departure, then the GPS data will be supplemented by interpolation. In the selected GPS data, the latitude and longitude of any two adjacent GPS records are (x b,w,j ,y b,w,j )、(x b,w,j+1 ,y b,w,j+1 ), the recording times are RT b,w,j , RT b,w,j+1 , the interval time is Λ, and Λ data are inserted between two GPS records with an interpolation interval of 1s. The latitude and longitude of the inserted λth GPS data are calculated:

[0056]

[0057]

[0058] Among them, the corresponding recording time is RT b,w,j +λ;

[0059] Step 4: Determine the vehicle's arrival location, arrival time, and departure time: For any station s on route r i , use the haversine formula to calculate the train number w completed in step 2 b The distance between all GPS data and the site, select the latitude and longitude of the GPS data closest to the site As the vehicle arrival location, the corresponding recording time As the arrival time AT b,w,j ; Identify the time AT in GPS data b,w,j The moment when the speed changes from 0 to non-zero state is the corresponding vehicle departure time DT b,w,j ;

[0060] Step 5: Identify the passenger boarding station: b The associated IC card records are sorted by the card swiping time. It is known that the kth card swiping time of passenger p is TT p,k (1≤k≤K, K is the number of times the passenger swipes the card on the day), compare the swiping time of any card swiping record with the arrival time of the train at any station, when AT b,w,i ≤TT p,k <AT b,w,i+1 When AT b,w,i Corresponding sitesi BS is the boarding station where the passenger swipes the card p,k .

[0061] Get-off point estimation:

[0062] According to the travel chain theory, individual bus trips are divided into closed bus trip chains, non-closed bus trip chains, and non-bus trip chains. The alighting station for all trips except the last one in a non-closed bus trip chain is calculated in the same way as that for a closed bus trip chain. The alighting station for the last trip of the day needs to be calculated with the help of historical travel data and the characteristics of downstream stations, which is the same as the processing method for non-bus trip chains.

[0063] The calculation of the alighting stations of a non-closed bus travel chain specifically includes:

[0064] Step 1: According to the boarding station BS p,k , determine whether the current card swiping record is the last one of the day. If k < K, go to step 2, otherwise go to step 4;

[0065] Step 2: Determine the boarding station BS for the current train p,k and the next bus boarding station BS p,k+1 Whether they belong to the same line, if so, the next bus stop will be used as the current bus stop, i.e. AS p,k =BS p,k , otherwise go to step 3;

[0066] Step 3: Determine the BS p,k+1 The set of stations CAS from upstream to downstream that the current ride passes through in the circular area with the maximum walking distance D as the radius is the center of the circle p,k ={cas m |0≤m≥M}; if the set is an empty set, then the trip chain belongs to a non-public transportation trip chain; if the number of stations in the set M=1, then the only station in the set is the alighting station AS of the current bus p,k If M>1, calculate the number of p,k Any site cas m BS p,k+1 The travel time t to the destination m , select the station with the shortest travel time cas * =argmin(t m ) is the current get-off station AS p,k , the calculation formula is as follows:

[0067]

[0068] Among them, v w is walking speed; d m For site cas m With BS p,k+1 The distance between m-1,m For site cas m-1 With cas m The length of bus routes between b is the bus operating speed;

[0069] Step 4: Determine the boarding station BS of the last ride p,k and the boarding point BS for the first ride p,1 Whether they belong to the same line, if so, the boarding station of the first ride will be used as the alighting station of the last ride, i.e. AS p,k =BS p,1 , otherwise go to step 5;

[0070] Step 5: Determine the boarding station BS for the first ride p,1 The set of stations from upstream to downstream that the passenger passed by on the last trip within the circular area with the center as the circle and the maximum walking distance D as the radius. If the set is empty, the passenger travel chain belongs to a non-closed bus travel chain. Otherwise, determine the alighting station of the last trip according to step 3.

[0071] The estimated drop-off points for non-public transportation chains are:

[0072] Step 1: Determine the boarding station BS for the current train p,k Downstream Site Assembly DAS p,k ={das n |1≤n≤N};

[0073] Step 2: Calculate the number of passengers at any station n The probability of getting off at PA p,n , the calculation formula is:

[0074] PA p,n =f n,1 ·f n,2 / ∑(f η,1 ·f η,2 );

[0075] Among them, f n,1 、f n,2 They represent the development intensity around the station and the characteristic parameters of public transportation accessibility; f n,1 =BF n / ∑BF η , f n,2 =R n / ∑R η , BF nFor site das n The average number of passengers boarded during a certain period indirectly reflects the development intensity around the station; R n For passing sites das n The number of bus routes.

[0076] Step 3: According to the passenger's arrival at any station n The probability of getting off at PA p,n , use the roulette method to calculate the current train's get-off station AS p,k .

[0077] The transfer behavior of individual urban bus travelers can be divided according to the transfer mode and the transfer space distance. The individual travel chain can be divided according to the transfer time threshold, specifically as follows:

[0078] Step 1: Determine the two previous and next boarding stations BS based on the passenger boarding station identification results p,k and BS p,k+1 Whether they belong to the same route. If so, the passenger's trip on that day is a unimodal bus trip. If not, determine whether the two bus routes belong to the same functional level. If so, the passenger's trip on that day is still a unimodal bus trip. If not, the passenger's trip is a multimodal bus trip.

[0079] Step 2: According to the alighting station AS p,k , determine the site AS p,k The corresponding bus operating frequency w b , obtain the arrival schedule of the vehicle of this shift, and convert AS p,k The station name is matched with the station in the arrival timetable, and the vehicle is obtained at the station AS p,k Arrival time AT b,w,j ; Take 0.25 of the current bus travel time as the passenger's get-off time Δt p , then the passenger's getting off time AT b,w,j,l , the calculation formula is as follows:

[0080] AT b,w,j,l =AT b,w,j +Δt p ;

[0081] Step 3: Calculate a certain get-off time AT b,w,j,l BS with the next boarding station p,k+1 Corresponding boarding time TT p,k+1 The time interval μ p,l , μ p,l =TT p,k+1 -AT b,w,p,l ;

[0082] Step 4: When the passenger transfers at the same station, that is, the passenger's two boarding stations belong to the same line, determine the maximum waiting time WT of the passenger at the station based on the departure intervals of the lines at different functional levels of the multimodal bus. p,k , the maximum waiting time is used as the maximum transfer time threshold η for transfers at the same station p,k , that is, η p,k =WT p,k When the passenger transfers at different stations, that is, the two boarding stations are not on the same line, the maximum walking distance D that the passenger can accept and the passenger's walking speed v w Calculate the maximum transfer time threshold;

[0083] Step 5: Set the travel time interval μ p,l and the maximum transfer time threshold η p,k For comparison, if μ p,k ≤η p,k , then the bus individual completes a single multi-modal or single-modal bus trip on that day, and μ p,l is the transfer time; otherwise, the bus individual completes multiple multi-modal or single-modal bus trips on the same day, μ p,l For the activity time.

[0084] After completing the identification and division of individual travel chains, the identification of multimodal bus individual travel chains and the establishment of virtual OD pairs are carried out: the key information of individual travel chains is identified to meet the data input requirements of the passenger flow prediction simulation model and further improve the individual travel data;

[0085] It is necessary to define virtual O and D points for each individual in the multimodal public transportation network according to their historical travel habits. Considering the spatial location attributes of the individual's departure station (arrival station) in the road network, the Manhattan distance between different departure stations (arrival stations) is calculated and the departure stations (arrival stations) are clustered. A virtual O point (D point) is defined for each departure station cluster (arrival station cluster); DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is the most classic density clustering algorithm. It can determine the clustering structure of the research object according to the density of the sample data distribution. By setting a set of "neighborhood" parameters, one or more core objects in a set of samples can be obtained. In addition, some outliers can be found during clustering and it is not sensitive to outliers in the data set. Finally, the virtual OD pairs of the core objects (travel individuals) are determined.

[0086] The establishment of a virtual OD pair is as follows:

[0087] Step 1: Identify all complete travel chains of individual buses and number them, extract the starting boarding station of each travel chain, and form a set D p,BS ,Each object in the collection contains the number of each travel chain and the longitude and latitude coordinates of the corresponding boarding point;

[0088] Step 2: Set the neighborhood parameter ε according to the range of bus stops at different functional levels, and adjust MinPts accordingly according to the specific research object; p,BS Perform core object search and obtain the core object set Ω p,BS ;

[0089] Step 3: From the set Ω p,BS A core object is randomly selected as a seed to execute the clustering algorithm, and all sites reachable by its density are searched, thus forming the first cluster. Then The core objects contained in Ω p,BS Eliminate them, and then randomly select a seed from the updated set to generate the next cluster, and repeat until the set is empty;

[0090] Step 4: According to each cluster The trip chain number corresponding to each object in the filter is used to filter out the terminal alighting stations of the trip chain corresponding to these numbers, and form corresponding sets Each object in the collection also contains the number of each travel chain and the latitude and longitude coordinates of the corresponding alighting station;

[0091] Step 5: According to the neighborhood parameters (ε, MinPts) Search for core objects, remove outliers, and obtain a core object set

[0092] Step 6: Calculate the set and The center points of each corresponding set together constitute multiple virtual OD point pairs of passenger p; the calculation formula is:

[0093]

[0094]

[0095] in, Represents the X coordinate of the center point of each set; Represents the Y coordinate of the center point of each set; x b,w,n Represents the X coordinate of the objects contained in each collection; y b,w,nrepresents the Y coordinate of the objects contained in each set; n represents an object in each set; and N represents all objects contained in each set.

[0096] After identifying and segmenting individual travel chains and establishing virtual OD pairs, the individual forms a complete travel chain. Considering the continuity of travel habits, individuals will have multiple trips in the same travel chain. These travel records form the fundamental data source for the multimodal public transit near-term passenger flow forecasting model. To ensure accurate data input for the individual travel decision-making model, key information about the individual travel chain must be identified. This information includes the individual's starting and ending points, travel time to the boarding stop, boarding stop, waiting time at the stop, vehicle stop time, boarding time, alighting stop, alighting time, transfer time, and travel time to the final destination.

[0097] like Figure 2 As shown, in a multimodal bus network, the individual traveler, as a decision maker, first makes a departure time selection decision in the "trip destination (i.e., the starting point of the next trip)" state, and after determining the departure time, it will transfer to the "departure time selected" state. Then, it will select the boarding station according to its own travel principles and transfer to the "boarding station selected" state. At the boarding station, the individual will select the bus route between the starting and ending points of the trip, and after the selection is completed, it will reach the "in-car en route" state; in the "bus route selection" state, the individual has determined whether to take the vehicle directly to the destination or get off at certain stations and transfer to other routes to reach the destination. If no transfer is required, the individual will transfer from the "in-car en route" state to the "get-off station selected" state, and go to the destination, returning to the "travel destination" state; if a transfer is required, the boarding station will be selected in the "get-off station selected" state, starting a new round of action selection decisions and state transfers; the state transfer of the travel process of the individual traveler in the bus network;

[0098] Among them, the state space S is:

[0099] S={Destination / Origin, Departure-time, Boarding-stion, En-route, Alighting-station};

[0100] Among them, Destination / Origin is the latitude and longitude coordinates of a pair of virtual OD points in the individual's history; Departure-time is the specific departure time information selected by the individual; Boarding-stion is the latitude and longitude of the boarding station selected by the individual and the collinear route information contained in the station; En-route is the vehicle, operating train number and other information corresponding to the route of the individual's selected boarding station; Alighting-station is the route, vehicle, operating train number, latitude and longitude and other information of the alighting station selected by the individual.

[0101] Action space A:

[0102] A={Select-departure-time,Select-boarding-station,Select-bus-route,Select-alighting-station,To-the-destination} Each action space is a candidate data set, among which the Select-departure-time candidate set contains all departure time information of the individual's historical travel; the Select-boarding-station candidate set contains all boarding station information of the individual's historical travel; the Select-bus-route candidate set contains the collinear bus route information corresponding to all boarding stations of the individual's historical travel; the Select-alighting-station candidate set contains all historical alighting station information corresponding to the individual's "bus route selection"; the To-the-destination candidate set contains the latitude and longitude coordinates of all established virtual OD points.

[0103] The individual travel decision model includes direct reward R, future reward F, and state transition probability P. The action selection in each state is based on the future reward F to determine the state transition probability P, so the action with the largest P is selected as the current one and the transition is made to the next state. When the transition is made to the next state, the direct reward R of the action selection becomes known and the historical experience value set is updated. First, the corresponding R, F, and P in the state and action space are defined. The direct reward value is represented by the travel time cost, but the time cost of the individual during the travel process is actually a penalty value, which contradicts the definition of the reward value. Therefore, the negative of the time cost is used to represent the direct reward corresponding to each action.

[0104] (1)R SDT : The direct reward of departure time selection is determined by two factors: the individual's expected arrival time and the total historical travel time. When choosing a departure time, the individual will first determine the expected arrival time, and then determine whether the trip is early (negative value), on time (0) or delayed (positive value) based on the difference Δt between the actual arrival time and the expected arrival time. The individual's actual arrival time is determined by the sum of the departure time and the actual total travel time. The calculation formula of Δt is: Δt = t depart +t trip -t desire , where t depart Indicates the departure time, t trip represents the actual travel time, t desire Indicates the expected arrival time.

[0105] Since individuals take punctuality as their travel goal in the actual travel process, in order to conform to the reality of life, when defining direct rewards, both early arrival and delay are positive numbers, so that the reward of punctuality is minimized; R SDT It can be expressed as: Where Δt early Indicates the time difference of early arrival, Δt late The time difference indicating the delay.

[0106] (2)R SBS :The individual will determine the boarding station for this trip at the destination, then R SBS It can be defined as the travel time t of an individual from the end of the trip to the boarding point D / O,BS Negative number of: R SBS =-t D / O,BS ;

[0107] Individuals can arrive at the boarding station by walking, bicycle, car, etc., usually walking; when an individual needs to transfer to another station to wait for the bus at a certain alighting station, R SBS It is defined as the travel time t of an individual from the get-off station to another boarding station BS,BS' Negative number of: R SBS =-t BS,BS' ;

[0108] (3)R SBR :Individuals choose bus routes at the boarding station. The direct reward corresponding to each bus route is reflected in the total time required for the route, including the waiting time t of the individual at the current station. wait , the bus stop time at the station t stop and the travel time t between the boarding station and the final arrival station travel , when the individual needs to transfer midway, R SBR It also includes the individual transfer travel time t BS,BS' . t wait Specifically, it refers to the difference between the time when an individual arrives at a station and the time when the vehicle arrives at the current station; t stop Specifically, it refers to the difference between the time when a vehicle arrives at a station and the time when it leaves the station; t travel Specifically, it refers to the total time that an individual is in the vehicle; when an individual reaches the destination directly, R SBR Expressed as: R SBR =-(t wait +t stop +t travel );

[0109] When an individual needs to transfer to reach the terminal, R SBR Expressed as: R SBR =-(twait +t stop +t travel +t BS,BS' );

[0110] (4)R TD : The individual's get-off station and travel destination have been determined in the aforementioned action selection, so the direct reward to the destination is the travel time t between the individual's get-off station and the travel destination TD,D / O Negative number of: R TD =-t TD,D / O ;

[0111] 2. Future Return F

[0112] The future reward F is the basis for calculating the state transition probability. The direct reward value of an individual when making an action choice in a dynamic public transportation information environment is unknown, and its direct reward value can only be accurately known after the individual actually completes the action. Therefore, based on the continuity of the individual's travel habits, the future reward can be estimated through historical experience. Each action choice of the traveling individual has a corresponding action selection set, and each action in the set will have multiple records in the historical travel. These recorded values ​​are all the direct reward values ​​R corresponding to the individual after completing the action at that time. The future reward F is calculated based on the average of the direct reward values ​​R of each action history record.

[0113] (1)F SDT : The future return of departure time selection is the average of the difference between the actual arrival time and the expected arrival time of a departure time selected in the individual's historical travel OK; F SDT Expressed as:

[0114]

[0115] in, It represents the average of the early arrival time difference in historical trips, The average time difference of delays in historical trips.

[0116] (2)F SBS :F SBS The average historical travel time of an individual from the destination to the boarding station A negative number; if the individual transfers from a get-off station to another get-on station, then F SBS The average historical travel time for an individual to transfer from a get-off station to another get-on station Negative number of F SBS Respectively expressed as:

[0117]

[0118]

[0119] (3)F SBR :Individuals need to select bus routes when the boarding station has been selected; considering that individuals are always in a dynamic travel environment, they will adjust and reselect the travel route that can minimize their travel time cost in a timely manner according to the arrival of vehicles at the station. Therefore, when individuals are waiting for the bus at the station, they first select the average waiting time of each route based on the historical travel data. The negative number of the bus stop time at the station The negative of and the average travel time between the boarding station and the final arrival station The sum of the negative numbers of determines the optimal route for this trip; when an individual needs to transfer midway, the average transfer time of the individual must also be considered. If F is a negative number, SBR It can be expressed as:

[0120]

[0121]

[0122] At the same time, according to the principle of minimizing travel time cost, individuals also need to consider that the first vehicle that arrives during the waiting process may not be the vehicle for the pre-boarded route, and in this case, the future reward F needs to be recalculated. SBR ', F SBR 'The difference is that the individual does not take into account the current site F SBR ' is expressed as:

[0123]

[0124]

[0125] Then, after waiting for a while at the boarding station, when the first bus that arrives is the one for the individual's pre-boarding route, the individual gets on the bus and completes the bus route action selection; if the bus is not the one for the individual's pre-boarding route, the individual needs to select the bus route based on the F between the path of the current arriving vehicle and the other paths. SBR ', compare the state transition probabilities and then decide whether to get on the car to complete the current action selection.

[0126] 3. State transition probability P

[0127] Each time an individual makes an action choice, they need to calculate the state transition probability to determine the final action choice. In this paper, the state transition probability P is calculated using a logistic regression expression. The action with the largest P is the optimal choice for the current action decision, and the individual gives priority to this action. The state transition probabilities P are expressed as follows:

[0128]

[0129]

[0130]

[0131] There are five states of travel individuals in a multi-modal bus network. Then, the state transition probability matrix M (5) Defined as:

[0132]

[0133] As an external environment, the public transportation network mainly plays the role of carrying and restricting the movement trajectory and range of the two types of intelligent agents. The public transportation network is: PT-Network=<r′,s r ′,b r ′,w r ' -b ,t r ' -ser >, where r′ represents all routes included in the multimodal bus network, covering routes at different levels in the network; s r ′ represents the set of all line sites consisting of the sites included in each line; b r ′ represents the set of all line vehicles consisting of the operating vehicles contained in each line; w r ' -b Represents the set of all route trips consisting of the operating trips of each route; t r ' -ser Represents the set of all line operating hours consisting of the operating hours of each line.

[0134] The individual agent in the public transportation network system has its own attributes, behaviors, and goals. It can perceive and judge the external network environment to select a travel path, thereby completing a set of actions and reaching the destination. The individual agent P-Agent is expressed as: P-Agent=<(O,D…),S P ,A P ,R P ,F P >, where (O, D…) refers to the distribution set of the starting and ending points of the individual agent, including all the historical starting and ending points of the travel individual within a certain period of time; S P is the state set of the individual travelers; A P is the action set of the individual traveler; R P is the direct reward value set of individuals; F P is the set of individual future return values.

[0135] Among them, reference Figure 3 As shown, establish the movement rules of individual agents:

[0136] Step 1: Determine the travel chains between each pair of virtual OD points of individual agents in a certain historical period to form a candidate path set; generate R SDT 、R SBS 、R SBR 、R TD The direct return value set, the initial set is empty, and the corresponding future return value F is calculated at the same time SDT 、F SBS 、F SBR 、F SBR ', state transition probability P D / O,DT 、P DT,BS 、P BS,ER ;

[0137] Step 2: Determine the OD point and departure time for this trip. Set the location of the virtual OD point in the established bus network environment as the starting and ending points of the individual's trip. Number each individual according to their travel card number and place all traveling individuals at the corresponding travel O point according to the actual travel situation on that day. Determine the departure time, boarding station, and preferred travel path based on the state transition probability in step 1. Input the selection results into the simulation platform and set the principle of individual intelligent agents waiting at the station to first come, first served.

[0138] Step 3: Simulation starts. The vehicle is at the destination of the trip and waits at the destination according to the departure time entered in step 2. If the departure time is reached, the vehicle moves to the boarding station at walking speed and executes step 4. If the departure time is not reached, the vehicle continues to wait at the destination.

[0139] Step 4: When the boarding station is selected, record the R SBS Individuals queue up at the boarding station according to the order of arrival and the preferred travel route to wait for the pre-boarding vehicle. If the pre-boarding vehicle arrives first and the vehicle is not full before the individual boards, the individual agent performs the boarding action and proceeds to step 5; if the pre-boarding vehicle arrives first but the vehicle is full before the individual boards, the individual agent maintains the current state and continues to wait for the second vehicle to arrive; if a vehicle from another candidate route arrives first and the vehicle is not full before the individual boards, the F of the current route and the pre-boarding route must be calculated. SBR 'The corresponding P BS,ER ', determine whether to get on the bus. If so, proceed to step 5. If not, continue waiting. If other candidate route vehicles arrive first and the vehicle is full before the individual gets on, maintain the current state and continue waiting.

[0140] Step 5: In the vehicle on-the-go phase. Determine the get-off station based on the determined travel route; if the get-off time at the get-off station is reached and the individual does not need to transfer, the individual agent gets off the vehicle and proceeds to step 6; if the get-off time at the get-off station is reached but the individual needs to transfer, the agent moves to the boarding station at walking speed and returns to step 4; if the get-off time at the get-off station is not reached, the agent continues to maintain the current state;

[0141] Step 6: When the get-off station is selected, record the R SBR , the individual agent follows the preset path to the destination of the trip;

[0142] Step 7: At the end of the trip, the trip ends and the trip enters the end of the trip state, waiting for the next trip; record the R TD , and record the R according to the departure time and arrival time SDT ;

[0143] Step 8: Update each action selection set according to the direct reward values ​​generated in this trip.

[0144] The bus agent has fixed operating hours, routes, departure times, and departure intervals in a multimodal bus network. The arrival time at each stop will fluctuate due to the actual road conditions and the number of people getting on and off the bus. The bus agent B-Agent is expressed as: B-Agent = < (r, b, di b ,s r ,t ser ),S B ,A B >, where (r,b,di b ,s r ,t ser ) describes the inherent properties of the bus agent, which represents the bus route, vehicle number, departure interval, set of passing stations and operating time period to which the B-Agent belongs; S B is the state set of the bus; A B is the action set of the bus, S B Transfer reference Figure 4 shown.

[0145] Among them, reference Figure 5 As shown, the motion rules of the bus vehicle agent are established:

[0146] Step 1: Input the bus routes, set of passing stations, distance between stations, vehicle number, vehicle capacity, departure interval and operating time period of all bus vehicle agents within the bus network to be studied, and set the vehicle speed on the road; departure interval, vehicle capacity and speed are set according to the functional level of the line. At the same time, enter the bus readiness stage;

[0147] Step 2: Simulation starts. If the first bus on each route arrives at its departure time according to the operating time period, the process proceeds to step 3. If it does not arrive at its departure time, the process remains in a waiting state.

[0148] Step 3: The bus in the en route state travels at a predetermined speed, and the travel time of the vehicle is determined according to the distance between two adjacent stops. When the vehicle's travel time on the road section ends, the vehicle arrives at the stop and allows travelers to get on and off. If not, the bus remains in the en route state.

[0149] Step 4: The vehicle determines whether it stops at an intermediate station based on the set of stops entered in Step 1. If so, the dwelling duration is calculated based on the number of passengers getting on and off as simulated in the bus network. If it does not stop at an intermediate station, it means the vehicle has reached the terminal station, and the dwelling duration can be directly calculated based on the number of passengers getting off. The dwelling duration is calculated using the Ceder model as follows:

[0150]

[0151]

[0152] in, Represents the stop time of a single-door boarding and alighting bus; Represents the stop time of a bus with separate boarding and alighting services; t b Represents basic docking time; Represents the average boarding time per person; represents the average time it takes to get off the bus; δ b Represents the number of people on board; δ a represents the number of people getting off the bus; the unit is seconds. According to the field survey of domestic scholars, the unit boarding and alighting time of individual travelers is 2.26 seconds per person and 1.21 seconds per person respectively. This paper uses these values ​​for simulation operation;

[0153] At the same time, the number of passengers that can board the vehicle needs to be calculated. When the number of passengers that can board is 0, the parking time ends directly: N ap =N bap -N ip -N ap +N bp Among them, N ap Represents the number of people that can board; N bapRepresents the vehicle's interior capacity; N ip Represents the number of people in the car; N ap Represents the number of people getting off the bus; N bp Represents the number of people on board.

[0154] Step 5: When the vehicle stops at an intermediate station, the stop time ends and the process returns to step 3; if the vehicle is at the terminal station, the stop time ends and the process returns to step 6; if the stop time does not end, the bus remains in a stopped state.

[0155] Step 6: After this operation is completed, the vehicle enters the bus station and queues up for the next bus to depart.

[0156] Based on the above two types of intelligent agents, as well as the movement and interaction rules of the intelligent agents, the overall process of the simulation model can be constructed as follows: Figure 6 shown.

[0157] In a specific embodiment, in order to verify the effectiveness of the simulation model, two bus routes with different functional levels but on the same line and 100 individual travel agents were set as examples. The bus operation and the interaction of passengers getting on and off the bus were simulated in a circular operation mode, and the number of people getting on and off at each station was displayed in the monitoring window as the passenger flow prediction result. The bus route layout of the example is as follows Figure 7 The bus network, bus vehicle agent, and individual travel agent settings are shown below:

[0158] 1. Public transportation network settings

[0159] Lines L1 and L2 are bus routes in the trunk and backbone transportation networks, respectively. L1 includes four stations: S1, S2, S3, and S4, while L2 includes four stations: S1, S3, and S4. The two lines share the same line: S1, S3, and S4. The distance between adjacent stations on Line L1 is 500 meters. On Line L3, stations S1 and S3 are 1000 meters apart, and S3 and S4 are 500 meters apart.

[0160] 2. Bus vehicle intelligent body settings

[0161] The parameter settings of buses on lines L1 and L2 are shown in the following table:

[0162]

[0163] 3. Travel Individual Agent Settings

[0164] The numbering, OD points, candidate paths, and departure time settings of the individual agents are shown in the following table. To simplify the calculation process, the departure time and departure station of the individuals are directly given here. When all individuals arrive at their destination, the current simulation round ends and a new round of simulation begins.

[0165]

[0166]

[0167] 5.3.2 Simulation algorithm flow

[0168] Based on the basic conditions of the set example, write the simulation code in NetLogo. The relevant code algorithm flow is as follows:

[0169] Step 1: Initialization phase: construct the route with a time granularity of 0.01 seconds per simulation step and a spatial granularity of 1.63 meters per pixel, and generate environmental elements such as the station body;

[0170] Step 2: Initialize the individual agent code, start time, and path set by importing them. Individuals are generated in the order of the code and assigned a first appearance time node. The path set is represented by a three-layer nested array. The innermost layer contains a two-dimensional vector representing the path node, where the first dimension represents the bus route on which the path is located and the second dimension represents the station. The second layer is the path composed of the node array. The outermost layer represents the ordered set of all paths of the individual agent.

[0171] Step 3: Generate a bus agent and set the time it appears at the first stop according to the departure schedule. Set its Boolean variable operating to False, indicating that the bus is not currently operating. The my_passengers variable is used to store all the individual agents on the bus. Initially, this set is empty. Initially, set its wait variable to True to prepare for its first stop at the departure stop.

[0172] Step 4: Enter the main loop phase, check the current time node, and if the individual agent that meets the time condition appears at the first node of the selected route, put the individual agent into the set named passengers_here at the corresponding station and record the passenger's arrival time arrival_time. The arrival_time of the first passenger is defined as its initialization time, and its on_board variable is set to False to indicate that it is not on the bus at the beginning;

[0173] Step 5: Check the departure time of the bus, change the operating variable of the bus that meets the departure conditions, set it to True, and appear at the starting station;

[0174] Step 6: Individual agents get on and off the bus. The process of getting on and off the bus is divided into the following six subroutines:

[0175] (1) Update the location information of individuals in the vehicle. For vehicles arriving at a station, determine whether my_passengers is an empty set. If not, update the location variable pos of all passengers to the station and delete the corresponding element of the station in the route to prepare for the subsequent boarding and alighting condition judgment.

[0176] (2) When an individual gets off the bus, search for the last individual to get on the bus and get off at this station (based on the individual's route node) at 1.21 second intervals to ensure a "last in, first out" order. Change the individual's on_board variable to False and set its arrival_time to the current time.

[0177] (3) Update the passengers_here collection of this station and include individuals in the my_passengers collection of the bus whose on_board is False and whose station is not its terminal into the collection.

[0178] (4) Update the my_passengers collection of the bus and remove individuals whose on_board is False.

[0179] (5) Individual boarding: Count the number of elements in the bus's my_passengers set. If the number is less than 30, enter the individual boarding subroutine. Search the passengers_here set at this station every 2.26 seconds and select the individual who needs to board the bus and has the smallest arrival_time, so as to satisfy the "first come, first served" order. Set the on_board variable of the found passenger to True.

[0180] (6) Execute subroutines (3) and (4) again to ensure that changes in each set are updated in a timely manner.

[0181] Step 7: Bus vehicle driving, including the following subroutines;

[0182] (1) Determine whether the bus has completed all designated stops. If so, set the operating status to False. Otherwise, enter the following subroutine.

[0183] (2) Combine the bus route information and location information to determine whether the bus has arrived at the station where it needs to stop. If it has not arrived, it continues to run; otherwise, it stops at the station and sets the wait variable to true and the stop start time stop_timing to the current time.

[0184] (3) If the current time is greater than 2 seconds from stop_timing (basic stop time), and there is no individual who needs to get off the bus and the bus is fully loaded, or there is no individual who needs to get on the bus at the station, then set wait to False.

[0185] (4) Vehicles whose variable wait is False will continue to travel at the set speed.

[0186] Step 8: If all individual agents have reached their destinations in this round and all the vehicles' operating variables are False, the time is refreshed and the next simulation round begins from 0. All bus variables are reinitialized, but passengers retain their previous knowledge of all route waiting times to make route selection decisions in the new simulation round.

[0187] 5.3.3 Simulation result output

[0188] 1. Interface display

[0189] The basic animation interface of the NetLogo simulation platform is as follows Figure 8 As shown in the figure; there are 4 stops in total, the yellow vehicles are running on line L1, and the orange vehicles are running on line L2. The number of waiting people on the two lines and the cumulative number of getting on and off at each stop are displayed in real time above the stop number, and the number of remaining passengers in this round of simulation is displayed in the lower right corner of the interface; there are three types of sliders: "setup", "go" and "total_rounds", which represent reset, start / pause and total rounds respectively, for controlling the overall simulation progress.

[0190] Result Output

[0191] During the initialization run, the individual agent goes through each path in the order of the candidate paths. Only after all paths have been completed can it start making autonomous decisions. Therefore, there are 4 rounds of initialization run. Starting from the 5th round, the individual agent starts making autonomous decisions and maintains a relatively stable state with the minimum time cost from this time on.

[0192] The monitoring window of the NetLogo simulation platform displays the running time of each simulation round, the real-time number of people on the bus between stations, the total real-time number of people on the bus, and the real-time number of people getting on and off the bus at stations 1 to 4. Figure 9 At the end of each round of simulation, the actual number of people getting on and off at each station and the simulation running time are shown in the following table:

[0193]

[0194] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for predicting public transport passenger flow, characterized by: The steps include: S1, obtained individual travel data and bus operation data as input; S2, building an individual travel decision model; S3. Based on the constructed individual travel decision model, establish the movement rules of the individual travel agent, the movement rules of the bus vehicle agent, and the interaction rules between the two; S4. By inputting the actual characteristic information of the two intelligent agents and the environment, and simulating the actual state based on this information according to the pre-set movement rules and interaction rules, the required passenger flow data is output to complete the passenger flow prediction; In step S2, based on individual travel data, the individual boarding and alighting station information is inferred based on the time matching method and trip chain theory. Based on the different functional levels of multimodal public transportation, the individual bus transfer behavior is analyzed, and the multimodal public transportation individual travel chains in small and medium-sized cities are identified and divided. Virtual OD pairs of travel individuals are established to obtain complete multimodal public transportation individual travel data. Then, an individual travel decision model is constructed based on the Markov decision process theory and the individual's historical travel data. The individual agent Expressed as: ,in, It refers to the distribution set of the starting and ending points of individual agents, including all the historical starting and ending points of the travel individuals within a certain period of time; is the state set of individual travelers; is the set of actions of the individual travelers; is the set of direct reward values ​​of individuals; is the set of individual future return values; Establish the movement rules for individual agents: Step 1: Determine the travel chains between each pair of virtual OD points of individual agents within a certain historical period, form a set of candidate paths, determine the OD points and departure times of the trips, import them into the public transportation network, optimize the travel paths, and input the selected paths into the NetLogo simulation platform. Step 2: Determine whether the departure time has arrived. If so, move to the boarding station at walking speed. If not, continue waiting at the destination. Step 3: Wait in line at the boarding station for the pre-boarding vehicle. If the pre-boarding vehicle arrives first and is not full, board the vehicle. Otherwise, remain in the current state and continue waiting. If a vehicle on another candidate route arrives first and is not full before the individual boards, calculate the direct rewards and corresponding state transition probabilities for the current route and the pre-boarding route, and based on the calculated results, board the vehicle or remain in the current state and continue waiting. Step 4: While in the vehicle, the agent determines the destination based on the determined travel route. If the agent reaches the destination at the time of getting off and does not need to transfer, the agent proceeds to the destination along the preset route. If the agent reaches the destination at the time of getting off but does need to transfer, the agent moves to the destination at walking speed. If the agent does not reach the destination at the time of getting off, the agent maintains the current state. The bus intelligent body Expressed as: ,in, Describes the inherent properties of the bus agent, which are represented in turn The bus route, vehicle number, departure interval, set of stops along the way and operating time period; is the state set of the bus; is the action set of the bus; Establish the motion rules of the bus agent: Step 1: The bus will depart at the departure time according to the operating time period; Step 2: After arriving at the stop, determine whether it is an intermediate stop. If so, calculate the stop duration based on the number of people getting on and off the bus simulated in the bus network; if not, calculate the stop duration directly based on the number of people getting off the bus; Step 3: After the stop time is reached, the bus will depart for the next stop or the operation ends, and the vehicle will enter the bus station to queue up for the next bus to depart.

2. A method for predicting public transportation passenger flow according to claim 1, characterized in that: Individual bus travel data includes IC card swipe data, bus GPS data, and bus route station location information. GPS data is interpolated and matched with the station's longitude and latitude to obtain vehicle arrival times, which are then matched with passengers' IC card swipe times to identify boarding stations. Different bus trip chains are divided according to trip chain theory, and the inference process is simplified by combining passenger travel assumptions and bus station characteristics to obtain alighting stations for different types of bus trip chains.

3. A method for predicting public transportation passenger flow according to claim 1, characterized in that: The bus network that carries and restricts the movement trajectories and ranges of the two types of agents is: ,in, Represents all routes included in the multimodal bus network, covering different levels of routes in the network; Represents the set of all line sites consisting of the sites included in each line; Represents the set of all route vehicles consisting of the operating vehicles included in each route; Represents the set of all route trips consisting of the operating trips of each route; Represents the set of all line operating hours consisting of the operating hours of each line.

4. A method for predicting public transportation passenger flow according to claim 3, characterized in that: The Ceder model is used to calculate the docking time: ； in, Represents the stop time of a single-door boarding and alighting bus; Represents the stop time of buses with separate boarding and alighting services; Represents basic docking time; Represents the average boarding time per person; represents the average time it takes to get off the bus; Represents the number of people on board; Represents the number of people getting off the bus.

5. A method for predicting public transportation passenger flow according to claim 4, characterized in that: The formula for calculating the number of passengers that can be boarded is: ;in, Represents the number of people that can board; Represents the vehicle's interior capacity; Represents the number of people in the car; represents the number of people getting off the bus; Represents the number of people on board.

Images