Research And Application Of Neural Network Ensemble In Bus Time Running Time Prediction

Now, the prediction of bus run time is becoming an important research hotspot in APTS (Advanced Public Transportation System) for its special significance.And it becomes a research difficulty because that it is related to with numerous factors and is a complicated nonlinear relation.Neural network ensemble (NNE) has been success widely used in many fields such as regression and classification for its unique properties.And it acquires a lot of increase in precision both prediction and classification.So NNE is used in the prediction of bus run time for increasing the generalization ability of the system and the prediction precision of running time.In order to achieve this goal,this paper makes the research and discussion in studying time prediction model and how to improve the generalization ability of the NNE.Firstly, the current popular prediction model of bus run time is analysed,input model with easy operation and simple principle input model is setted up based on analyzing the weather,holiday and so on which are related to the running time.Then three kinds of individual network which contain holiday type, normal type and weather type are trained.The two prediction models which are putted forward in this paper are based on the three kinds of individual network.Secondly, in order to ensure the accuracy of individual network and improve the diversity of individual networks,a method of second training is putted forward.Then genetic algorithm is used to optimize weights of individual networks and improve the generalization ability based on the minimum error. This model is named as PBRTST(The prediction of bus running time based on the second training).Lastly, the data in training set are classified and used different training strategies after analyzing the time series of bus running.Because of the performance of individual network has no consistency in data sample space and traditional integration method cannot dynamic adjust weights,generalized regression neutral network is used to dynamic adjust the weights of individual network.So it can improve the prediction accuracy and the generalization ability, this model is called PBRTSL(The prediction of bus running time based on the sort learning algorithm).