Bi-level Optimization Model For The Location Of Urban Rail Transit Stations Based On Simulated Annealing Algorithm

Statistics indicates,at the end of 2017,urban rail transit had been operated in 31 cities in China Mainland,which,to some degrees,had helped reduce the volum of traffic(mainly private cars)on the ground and the level of traffic congestion.At the same time,problems caused by the irrationality of station location have also attracted the attention of city builders and researchers.For example,ridership forecasting of some stations was overestimated,at the same time,there were many potential passengers,there were no stations.In addition,the phenomenon that distances between stations on express lines are too short also exist,which result in frequent stops and low efficiency,so that the advantage of express line is not used.Considering the high cost,large amount of work and the unrecoverable characteristics,it is of greatly practical significance,social value and economic value to improve the scientificalness and rationality of the location of urban rail transit stations.The study of the location of urban rail transit stations was based on the ridership forecasting at station-level and the optimazation of station spacing.To define the service area of urban rail transit stations,data collected of Harbin Metro Line 1 was used to research the location of passengers around stations.Furthermore,the service area of 800 m network distance was difined,which is less than the emprical value of 800 m Euclidean distance.Based on this value,a vector map of the construction of the Harbin Metro Line 1,station,service areas,and the walking network dataset of passengers were established in Arcgis 10.1.In the study of ridership forecasting,the phenomena of the different influence of explanatary variables in different locations of a service area to a station and of a line to different stations were taken into consideration.Based on the comparison of the four-step model,the multiple regression model based on the least squares method and the geographically weighted regression(GWR)model,the weighted floor area of the building of six categories of land use types including administrative management,education and culture,residence,business,recreation and medical and health care were selected as the initial explanatory variables,so did the land use diversity index,the number of feeder bus lines within 200 m,the station accessibility and the density of the road network,to establish the ridership forecasting moedel based on GWR.Taking into account passenger trip cost and urban rail transit operating cost,the comprehensive transportation cost of the urban rail transit is defined,based on which spacing of stations were optimized.The passenger trip cost includes the time cost of walking to the corresponding station,the waiting time cost at the station,the travel time cost,the time cost of leaving the station to the destination,and the ticket fee.For the operating costs,only the cost of purchasing trains and the cost of electric power consumption which were related to the station spacing were considered.Considering the public attributes of urban rail transit,it should provide services for as many travelers as possible.Therefore,the ridership forecasting model based on station level was taken as the upper level.At the same time,a rational station slection plan should be able to guarantee the cost of passengers and operators are both in a low level,making the plan both socially and economically valuable.Then,the station optimazation moedel was taken as the lower level so that the bi-level medel of urban rail transit stations location was eatablished.Considering the difficulty of solving the bi-level model and comparing the features of several heuristic algorithms,the simulated annealing algorithm was selected to search the optimal solution because of the feature of being able to accept worse results and adjustability of parameters.Finally,taking Harbin Metro Line 1 as an example,program was written according to the principle of simulated annealing algorithm in MATLAB R2015 a,the model was solved and the validity and practicability of the model were proved.