Application of bayesian networks to estimate the probability of a transfer at a public transport stop

Optimizing transfers during public transport operations is one of the essential components of improving the quality of transport. Several factors influence the passenger's perception of a transfer: from the personal characteristics of the user of transport services to the parameters of the route network, trip characteristics and the design of transfer stops. The method of constructing Bayesian networks was used as one of the effective methods for solving problems of forecasting complex systems to find the relationship between different types of input data that affect the probability of making a transfer at a stop. The need for a transfer arises for a passenger when two reasons are combined: the need to make a trip between two transport areas and the lack of a direct public transport route between these transport areas. The number of needs for trip will depend on the number of residents in the departure zone, and the probability of not having a direct route will depend on the total number of routes departing from this zone. A simulation was carried out in the PTV Visum software environment (on the example of Lviv city) to determine the impact of these factors on the probability of changing at a stop. As a result, data were obtained on the total amount of passenger exchange at the stops of the public transportation system with distribution into the number of passengers disembarking at the stop, the number of passengers transferring at this stop, and the number of passengers going (up to 200 m) to another stop to transfer. The average waiting time for a transfer at a stop depends on both the number of routes passing through the stop and the regularity of traffic. Strict adherence to traffic schedules helps to reduce the average waiting time for a transfer. A comparison of the results of calculating the probability of a transfer at one of the stops using calculations based on field observation data and using modeling was carried out to check the adequacy of the modeling. The calculated probability is 0.16, the simulated probability is 0.12.