Bus Arrival Time Prediction Model Based On Clustring And Recurrent Neural Network

With the development of economy and advancement of technology,the concept of smart city emerged.Through the use of advanced information technology,the city's intelligent management and operation can be realized,creating a better life for everyone in the city.Smart public transport is an important part of a smart city and meets the needs of people for green travel and quick travel.Bus arrival time forecasting is the basis for realizing smart public transport.Increasing the accuracy of bus arrival time forecasting is of great significance to the realization of smart public transport and smart cities.First of all,this paper analyzes the different factors affecting the arrival time of buses,preprocesses the GPS data of buses,and generates the basic data set of buses.Secondly,the PBCR model of bus arrival time prediction combined with clustering and recurrent neural networks is proposed.Then,the PBCR model is parallelized on the Spark platform.Finally,the effectiveness of the model is verified by experimental methods.The innovations proposed in this paper include the following two aspects:1)A PBCR model of bus arrival time prediction combined with clustering and recurrent neural networks is proposed.The core idea of ??the PBCR model is to use the fuzzy K-means clustering algorithm to divide the time period,and use the recurrent neural network to predict the bus-to-station time in each period.Experimental results show that the proposed PBCR model is feasible and effective.2)Parallelization of the PBCR model on the Spark platform.Bus GPS data has the characteristics of large data volume.It is necessary to parallelize the PBCR model in order to better mine data information and process massive data.The key and difficulty in implementing PBCR model parallelization lies in the realization of the parallelization of the recurrent neural network.In this paper,the parallelism of the recurrent neural network is implemented in the Spark platform using data parallelism.The experimental results show that the parallelized PBCR model has good scalability,and reduces the running time of the algorithm by extending the number of nodes in the cluster.