Research On Urban Macro Travel Speed Prediction Based On Classical Models And LSTM Model In Time Series Models

In this paper,the daily average speed of urban roads is predicted and deduced from a macro perspective.The object of this study is the aggregated macro travel speed of the city based on the big data of online car-hailing.This data is the average daily speed of the urban roads within the six districts in Beijing,which has high research value in the macro sense.At present,there are many research methods in the field of traffic flow prediction.But in general,the focus of this study is on the local traffic state,and there is a lack of research on the macro-state of traffic flow.Based on the study of urban macro travel speed,the author regards this data as a time series and decomposes it.It finds that there are fancy seasonal and trend characteristics in the data.Based on this deduction,the external factors affecting urban macro travel speed are analyzed.Then,different algorithms of time series model are introduced to build different macro prediction models.On the one hand,in the classical algorithm of time series model,this paper studies auto-regressive intergrated moving average(ARIMA)algorithm in detail.In order to improve the information capturing ability of the model,the STL-ARIMAX model is proposed,which combines time series decomposition process with dynamic regression model creatively.This model can capture more information of the data.At the same time,the prediction accuracy is improved.On the other hand,in the neural network algorithms of time series models,this paper uses the long short-term memory network(LSTM)in deep learning to predict macro speed.External factors can be input as additional feature dimensions,so LSTM model has higher flexibility and better prediction effect than classical prediction methods for time series.In this paper,based on the macro travel speed of the city,the characteristics of the data and its prediction model are studied in detail.At the same time,this paper also improves the flow of classical algorithms and the input of neural network algorithm.By doing so,we can better grasp the changing law of macro traffic flow speed and the fluctuation of speed caused by various external factors.Besides,the improved models can provide a scientific basis for the formulation of relevant traffic policies,as well as provide inspiration and reference for the related research on the macro characteristics of urban traffic in the future.