Research On Bus Arrival Time Prediction Method Based On GAN-LSTM

In recent years,with the rapid development of urban economy,more and more family cars aggravate the urban traffic congestion.Attracting more passengers to choose public transport can alleviate this situation.For urban passengers,one of the most concerned information is the bus arrival time.Accurate prediction of the bus arrival time can not only make passengers more reasonably plan their travels,reduce the anxiety while waiting for a bus,but also help better adjust the time schedule.Therefore,it is of great practical significance to study the accurate bus arrival time prediction model for the development of the smart city.This thesis proposes the combination of LSTM neural network and Conditional Generative Adversarial network to form a novel CGAN-LSTM prediction model.During experiments,this thesis optimized the network’ s architectural parameters,the optimization function,the loss function and activation functions.By introducing additional label information into the generator and discriminator,a zero-sum game can make the generator learn a distribution which is close to the real data.This thesis analyzes the bus GPS data acquisition mode and data format.Moreover,through the analysis of the bus driving process and influencing factors,This thesis designs an algorithm to obtain the bus arrival time.In this algorithm,the concept of the buffer is introduced which can help accurately decide the ground truth.Finally,the format transformation and the data normalization methods are applied for the obtained arrival time to provide sufficient training data for the CGAN-LSTM.Finally,the experimental results report that the prediction performance of the proposed CGAN-LSTM is promising.Meanwhile,this thesis also demonstrates the model under different conditions such as peak hours of working days or non working days,our model still have strong adaptability.Compared with LSTM and SVR models,the accuracy of this model is higher and more stable.Therefore,the CGAN-LSTM prediction model proposed in this thesis can not only well predict the bus arrival time,but also it is easier to modify according to the practical changes.