Abstract: In order to quantify the reliability of airport bus line networks,a reliability prediction model of the air- port bus travel time is constructed based on the BP neural network.Next,an optimization model of the airport bus line network is constructed,the objective of which is to maximize the reliability,and such constraints as time,sites and services are taken into account.Then,the Hill-climbing algorithm is adopted to achieve the initial solutions for the lines,and a fitness function based on the reliability is established.Finally,a hybrid genetic algorithm with dif- ferent mutation and crossover rates is designed to solve the constructed optimization model.Case study results show that (1) during peak hours,the reliability of the bus line network at Nanjing Lukou International Airport is only 0. 62,the reliability of inner city roads is about 15% lower than that of outer city roads,and the overall reliability is at a low level; (2) the optimization process through the hybrid genetic algorithm is greatly affected by the cross- over and mutation rates,and lower crossover rate and higher mutation rate can increase the instability of the optimi- zation process; and (3) when the crossover rate is 0. 9 and the mutation rate is 0. 05，the objective function value of the constructed optimization model is 0. 79,and the reliability is 11. 5% higher than that before the optimization,which means that the optimization effect is significant.This method provides a scientific basis for the optimization of the bus line network of airports and the corresponding efficiency improvement of the external transport services.

Key words: airport bus, travel time reliability, BP neural network, genetic algorithm, line network optimization