Predicting Stay Time Using An Rnn-based Encoder With Neural Arithmetic Logic Unit

The relationship between transportation and development has always been the focus of urban planners and policy makers.With the continuous advancement of information technology,the intelligent transportation system improves the utilization rate of transportation facilities in the city,provides real-time route planning for urban residents,guides intelligent driving methods,and effectively feedbacks and predicts traffic conditions.The specific traffic demand gives a feasible reference strategy.The travel of urban private vehicles is often related to the driving purpose.The stay time of urban private vehicles can reflect the driving purpose of private vehicles to a certain extent.Knowing the stay time of urban private vehicles in advance effectively improves the services of various location-based applications.Quality and operational efficiency,and can provide additional reference information for the future traffic flow in the parking area,which is conducive to traffic planning and flow distribution.Therefore,predicting the stay time of urban private vehicles can effectively improve the operating efficiency of urban intelligent transportation systems.Due to the randomness,instability and uncertainty of the staying behavior of urban private vehicles,it is a difficult task to effectively predict the stay time.Therefore,this paper has carried out data cleaning based on the data information of urban private vehicles,carried out a comprehensive data analysis based on the effective stay data of urban private vehicles,and carried out corresponding algorithm design according to the characteristics of the staying behavior of urban private vehicles to improve the accuracy rate of predicting the stay time of private vehicles in the city is improved.The main work of this paper is as follows:As the data of urban private vehicles is difficult to obtain,this paper designs an OBD-based vehicle information system to obtain the trajectory information of real urban private vehicles.For the original data obtained from urban private vehicles,this paper adopts various data preprocessing methods to clean it,which improves the value density,effectiveness and usability of urban private vehicle data.This paper proposes an algorithm for detecting staying behavior of private vehicles in the vehicle information data,and uses the DBSCAN clustering algorithm to extract the temporal and spatial characteristics of the staying behavior of urban private vehicles,and analyzes Characteristics of staying behavior of urban private vehicles.In this article,we propose a neural network-based urban private vehicle stay time prediction model to solve the problem of predicting stay time.Specifically,the prediction model we proposed consists of three parts: encoder,anomaly module and decoder.First,we encode the staying behavior of urban private vehicles as hidden vectors to avoid the impact of time sparsity.The encoder module uses multi-layer perceptron to learn spatiotemporal features from historical trajectory data,such as the difference between the stopping point and the corresponding stopping time the inherent relationship between.Then,we construct an abnormal module with neural arithmetic logic unit in the prediction model,thereby enhancing the function of the neural network to process linear relationships,so that the proposed prediction model has better prediction capabilities.In addition,we use pruning techniques to prevent overfitting of the prediction model.This paper conducts extensive experiments based on largescale real-world private car trajectory data.The experimental results show that the performance indicators of this model in RMSE,MAE,MAPE,SMAPE,KL divergence are generally better than comparison methods.