پیش بینی زمان توقف اتوبوس برمبنای الگوریتم K-نزدیک ترین همسایه KNN

ABSTRACT Bus Dwell Time Prediction Based on KNN

In order to promote competitiveness with other modes of travel such as the passenger car, bus transit agencies need to improve the quality of service to attract more people. Providing the information about bus arrival time makes public transit more attractive [1]. Bus travel time between two points consists of two parts: bus travel time on segments between every two bus stations, and bus dwell time at bus stations. This article mainly focuses on prediction method of bus dwell time.

Bus dwell time is composed by three parts: bus door opening, passengers up and down, bus door closing [2].Bus dwell time is influenced by many factors. In the TCQSM (Transit Capacity and Quality of Service Manual), factors that affect the dwell time consist of passenger boarding and alighting volumes, fare payment method, in-vehicle circulation, and stop spacing [3]. Prediction models can be set up between dwell time and these influence factors.

Martin Milkovits [4] established bus dwell time model using data from the automatic passenger counting, automatic fare counting, and automatic vehicle location systems installed on Chicago Transit Authority buses. Peng Qingyan [5] added coefficients related to weather, time (flat or peak period) and station location, to adjust the primary models. Because of the size of BRT station, Influence of passengers walking time should be considered [6].

Beside of these mathematical regression models, Machine learning algorithms also have been used to establish the relationship between bus dwell time and influence factors [7, 8].

However, all of these models are poor in Generalization ability and practicality. Because, different bus stop with different situation need different prediction model, and, how many passengers will get on the bus and how many get off are both needed to be predicted. With the advent of big data era, relationship between the bus dwell time and influence factors can be ignored. What we need to study is its changing rule. Time series based methods has been used in bus dwell time prediction using history data of bus dwell time [9]. Compared with previous models, this kind of model can be used in practice without need of adjustment according to bus style, station form and also without need of prediction the number of passengers will on and down.

This article mainly focus on analysis of bus dwell time changing rule and propose a dynamic model based on KNN to predict bus dwell time at bus station.