Real-time Analysis And Short-term Forecast Of Bus Passenger Flow

Giving priority to the development of the public transportation is one of the effective measures to solve traffic problems in large-and middle-size cities. Facing with the severe urban traffic problems, the Ministry of Transportation has launched the demonstration project of public transport oriented cities, and chosen15cities including Beijing, Jinan and others to be the first batch of cities of this project. In order to realize the efficient operation of the public transport system, more efforts need to be done, and improving the management level of bus line operation is one of the key measures. The bus passenger flow analysis and prediction, especially the dynamic bus passenger flow analysis and prediction is the base to realize the optimization of the bus dispatch. Only a comprehensive analysis and accurate prediction of the short-term bus passenger flow can help managers grasp the real-time running status of bus lines, and make quick response to the fluctuations of short-term passenger flow and adjust the operation scheduling scheme effectively. Then it can actually realize the real-time scheduling and maximize the economic benefit of the enterprise as well as the social benefit.Since the traditional data collecting method for bus passenger flow is unable to meet the real-time data demand of the system, this study designed a real-time data collecting system for bus passenger flow. This system combined the passenger flow data which collected by IC system and coin detection system with the GPS data. Using the real-time upload feature of the GPS data, it uploaded the data of passenger flow and bus locations to the scheduling center. Combined with the utilization rate of IC card obtained from the survey, it can calculate the aboard flow data of the bus stop, bus line and network in a real-time manner.On the basis of real-time passenger flows data, the paper further conducted data mining of the IC data and GPS data to analyze passenger flow characteristics. It studied the methods of recognizing the stops for passenger boarding/alighting under the environment of complete data and partial data. The paper also studied the passenger transfer, and the time distribution feature, dynamic feature and the short-term feature of passenger flow. It analyzed the passenger flow fluctuation on weekdays, weekends and within a month, and quantitatively analyzed the time distribution and directional feature of the passenger flow within a day as well as the weekly fluctuation feature.The study identified three prediction models, namely the traditional BP neural network model, the improved BP neural network model and the RBF neural network model to predict the short-term passenger flow and evaluated the modeling result. Based on the passenger flow analysis, three categories of influence factors were determined to be the main inputs of the neural network, passenger flow in adjacent time intervals before the prediction time, passenger flow of the same time interval in adjacent weekdays before the prediction day, and passenger flow of the same time interval in adjacent weeks before the prediction week, and confirmed the neuron number of the input layer and hidden layer about the BP neural network model. The collated IC card data ws used to train the network and make prediction, The paper analyzed and evaluated the predicted results of each model, and the results showed that the prediction accuracy of the improved BP model and RBF model were better than the traditional BP model.