Research On Highway Short-term Traffic Flow Hybrid Prediction Methods Based On Artificial Neural Network

With the continuous increase of highway mileage and car quantity in China,a series of problems such as road traffic congestion,frequent traffic accidents and environmental pollution are becoming more and more serious.The vigorous development of intelligent transportation provides new ideas for solving these transportation problems.Traffic guidance and control are the core of intelligent transportation.Accurate and real-time highway short-term traffic flow prediction are an important basis for realizing traffic guidance and control,the prediction of traffic flow is affected by various time-varying and complex factors.Therefore,from the perspective of improving the prediction accuracy of the prediction model,studying the prediction method of short-term traffic flow on the highway has important application value.Based on the analysis of the spatial-temporal characteristics of highway short-time traffic flow,this paper proposes a data repair algorithm based on Manifold Distance with K Nearest Neighbor(MD-KNN).First,establish a historical database based on the collected multi-site traffic flow data;second,use the autocorrelation coefficient to measure the autocorrelation of the data,select a suitable time scale to construct the state vector and use the manifold distance to measure the similarity between the missing vector and the state vector,select the appropriate K state vectors to repair the missing value;finally,compare with the historical average method,linear interpolation method and KNN algorithm.The comparison results show that the MDKNN algorithm is superior to the traditional data repair algorithms.According to the spatial-temporal characteristics of short-term traffic flow,this paper proposes two short-term traffic flow prediction models.(1)the short-term traffic flow prediction model based on improved Mean Square Error Long Short-term Memory network(IMSE-LSTM).This model considers only temporal correlation and takes the time series of the traffic flow as the input sample,then uses LSTM to predict traffic flow,and optimize the LSTM layer through the dropout layer and improved loss function to improve the prediction accuracy of the model;(2)the short-term traffic flow hybrid prediction model based on spatial-temporal characteristics.This model takes into account the time and space correlation of traffic flow,and uses IMSE-LSTM and Radial Basis Function(RBF)neural network respectively to build the model,and then use the entropy method to perform weighted fusion on the two single prediction models to construct the IMSE-LSTM-RBF hybrid prediction model and obtain the final prediction result.The two prediction models proposed in this paper are compared with the ARIMA,SVR,LSTM,and GRU models.The experimental results show that the prediction models proposed in this paper have higher prediction accuracy and generalization performance,and the prediction effect of the hybrid prediction model is better and more stable,indicating that adding spatial characteristics to the prediction model can make the prediction accuracy higher.