Based On SARIMA And LSTM Model Smart City Traffic Forecast

In recent years,the construction of smart cities has become the engine of the rapid economic development in our country.Especially with the wide application and development of smart transportation,the demand for data application and analysis in transportation problems has become increasingly prominent.In order to alleviate traffic congestion and ensure good traffic operation,it is necessary to analyze and predict the potential conditions in traffic flow during smart traffic regulation and management,as well as to plan the traffic route in advance.The aim is to improve the traffic efficiency and traffic safety.However,due to the complex and changeable traffic flow and many influencing factors,it is difficult to predict the trend using such data.This thesis analyzes the network and actual smart traffic data,discusses how to construct a reasonable prediction model to improve the prediction accuracy of traffic flow,and provides a certain application reference for traffic regulation.Based on the basic characteristics of traffic flow data,we extracts the main features of short-term traffic flow data and conducts test analysis in combination with common prediction models.Considering the feasibility of the model and the prediction accuracy of short-term traffic flow,the analysis is based on the time series prediction model and the recurrent neural network model,respectively.After optimizing and testing the model parameters to obtain the corresponding prediction results,the two basic models are further combined,and then the combined model is used to predict the traffic flow data.In the construction and testing of predictive models,the traffic flow data is firstly sorted and analyzed,mainly dealing with a small number of missing values and individual outliers in the data set.Then,the data is sorted into the data type of the experimental input.A short-term traffic flow prediction model of the SARIMA model is established,the determination of the three main parameters p,d,and q in the model is analyzed.The parameters of the minimum error model are determined by comparison.Then establish a recurrent neural network model to predict short-term traffic flow,and compare the prediction effects of LSTM and GRU models under different time steps and mapping dimensions.It is found that when the training data set is increased,the prediction effect of the model is also improved.On this basis,the combined model is used for prediction.The traffic flow data is firstly predicted by the SARIMA model,then the residual error is predicted by the LSTM model,and the prediction result is input to the BP neural network.The nonlinear mapping transformation is used to obtain the prediction result.The prediction accuracy of traffic flow can be estimated by error analysis.Our results show that the combined model can be improved in a certain extent.