Method And Application Research On Traffic Flow Forecasting Based On Echo State Networks

With the rapid progress of urbanization in China,the traffic in the central city is seriously congested in some periods and regions.Accurate traffic flow prediction can provide a good forecast of road traffic condition for traffic control system,and alleviate the problem of urban congestion to a certain extent.However,there are two difficulties in the study of traditional traffic flow prediction,the first difficulty is the small forecast step length and poor accuracy of multi-step prediction of traffic flow,and the second difficulty is that the traffic flow prediction of multiple roads is time-consuming and inefficient.Based on the practical application,this paper considers the time dimension and spatial dimension in view of the problems existing in the current traffic flow prediction method.Based on the Echo State Network model,a single point multi-step traffic flow prediction model and a multi-point single-step traffic flow prediction model are proposed,and the two models are applied to solve the vehicle path optimization problem.The details are as follows:⑴ The time dimension.Aiming at the problem of the small forecast step length and low precision of multi-step traffic flow prediction,a single point and multi-step traffic flow prediction model based on Echo State Network is proposed,and its predicted time range is one day.Firstly,the time series are reconstructed by combining the periodicity of traffic flow in our model,and Principal Component Analysis is explored as a dimensionality reduction method.Then the Echo State Network model is used to predict the traffic flow time series.In this way,the original multi-step prediction problem can be transformed into multiple single-step prediction problems,and the prediction accuracy of the model will be improved.On the other hand,reducing the data dimension can improve the prediction efficiency,and removing the redundant data can further improve the prediction accuracy.In addition,an adaptive disturbance particle swarm optimization algorithm is used to optimize the parameters of the model.The availability of the proposed model was proved by predicting the time series of real traffic flow.The experiments demonstrate that the proposed model can effectively prevent the delay of prediction results and greatly improve the precision of multistep prediction.⑵ The spatial dimension.Aiming at the low efficiency of multiple road traffic flow prediction,a multi-point single-step traffic flow prediction model based on Echo State Network is proposed,and its predicted spatial range is multiple roads.First,the data is represented by a high-dimensional multivariable time series,in which the dimension of the variable is the total number of roads and the length of the variable is the length of the training sample.The high dimensional variability data is reduced to low dimension by Principal Component Analysis dimension reduction method.And then enter multiple Echo State Network models in parallel for prediction.Finally,the final prediction results are obtained after the Principal Component Analysis reconstruction.The main purpose of using Principal Component Analysis model in multi-point single-step model is to shorten the program running time and improve the prediction efficiency of the model.In addition,the simulation experiment of the model is carried out by using R language and the model is compared with the traditional Echo State Network single-step prediction model.Experimental results show that the model has a significant advantage in operational efficiency.Although the single-point multi-step prediction model and the multi-point one-step prediction model proposed in this paper are based on the combination of Principal Component Analysis and Echo State Network,the focus of time dimension is predictive step length and prediction accuracy,while the focus of spatial dimension is predictive efficiency and running time.According to the different emphases of the two models,two application models are proposed: The path planning model based on the single point multistep prediction model and the path optimization model based on the multi-point step prediction model.In addition,using R language to simulate the application of these two models,the experimental results show that the paths planned by these two path optimization models can take less time than the traditional methods.