Authors:Fabian Ohler1;Karl-Heinz Krempels1 andSandra Möbus2
Affiliations:1RWTH Aachen University and Fraunhofer FIT, Germany;2RWTH Aachen University, Germany
Keyword(s):Passenger Demand Forecast, Public Transportation Forecast, Transit Passenger Volume Prediction.
Abstract:Using a forecast of the public transportation capacity utilisation, the buses can be adapted to the demand to avoid overfull buses leading to delays. An efficient utilisation of the buses at disposal can improve customer satisfaction as well as economic efficiency. The basis for our forecasts provide fragmentary measurements of passengers boarding and alighting buses at stops over the year 2015. In an attempt to improve the accuracy of the forecast, several external factors (e. g. weather, holidays, cultural events) were incorporated. We tackle the problem of forecasting public transportation capacity utilisation by forecasting the number of boarding and alighting passengers. Then we use these to adjust previous passenger count and the result as input for next forecast. Using multiple linear regression, support vector regression, and neural networks we evaluate different ways to model the external factors. Best results were achieved by neural networks with a median absolute error of≈4.16 in the forecast passenger count. They were able to keep more than 80% of the forecasts within a tolerance of 10 passengers. Since the error in the forecasts does not accumulate along the trips, chaining the forecasts in the described way is a viable approach.(More)
Using a forecast of the public transportation capacity utilisation, the buses can be adapted to the demand to avoid overfull buses leading to delays. An efficient utilisation of the buses at disposal can improve customer satisfaction as well as economic efficiency. The basis for our forecasts provide fragmentary measurements of passengers boarding and alighting buses at stops over the year 2015. In an attempt to improve the accuracy of the forecast, several external factors (e. g. weather, holidays, cultural events) were incorporated. We tackle the problem of forecasting public transportation capacity utilisation by forecasting the number of boarding and alighting passengers. Then we use these to adjust previous passenger count and the result as input for next forecast. Using multiple linear regression, support vector regression, and neural networks we evaluate different ways to model the external factors. Best results were achieved by neural networks with a median absolute error of ≈4.16 in the forecast passenger count. They were able to keep more than 80% of the forecasts within a tolerance of 10 passengers. Since the error in the forecasts does not accumulate along the trips, chaining the forecasts in the described way is a viable approach.