Short-term Traffic Flow Prediction Based On K-Means Clustering And GRU Network

In order to alleviate congestion and improve the efficiency of road network traffic,many cities in China have begun to build Intelligent Transport System(ITS).Short-term traffic flow prediction is an indispensable part of ITS.It has laid a data foundation for traffic control and guidance,and plays an important role in traffic planning and management.Short-term traffic flow data is a typical time series data.Due to its uncertainty and dynamics,how to better excavate the regulation of short-term traffic flow changes from the historical short-term traffic flow sequence,for the prediction of further improving the model accuracy and stability are of great significance.This paper proposes a short-term traffic flow prediction method combining clustering and Gated Recurrent Unit(GRU)neural network.First,this paper analyzes the advantages and disadvantages of the current short-term traffic flow prediction model,studies clustering and neural network related algorithms,analyzes the short-term traffic flow parameters and preprocessing methods;Second,the stationarity test was established on test data,and a deep GRU prediction model was established.Regularization and dropout methods were used to prevent the model from overfitting.The short-term traffic flow data of Pe MS(Performance Measurement System)in California was used to predict the performance of the model.After verification,on the basis of this,the optimal hyperparameters of the model were found through the grid search method,and the deep GRU prediction model was optimized.Finally,for the problem that the training data selection has a significant impact on the prediction results of the GRU neural network,K-Means clustering algorithm was used to clusters historical short-term traffic flow data,establishes a short-term traffic flow pattern library,and uses the KNN(K-Nearest Neighbor)classification algorithm to determine the historical short-term that was most similar to the short-term traffic flow change trend of the date to be predicted,using all historical short-term traffic flow data in this category as training data to predicted,and using Pe MS data to verify the performance of KMeans-GRU prediction model.The KMeans-GRU prediction model proposed in this paper is used to predict the short-term traffic flow of multiple detection stations.The results show that,compared with the prediction results of the traditional GRU network,the Stacked Auto Encoders(SAEs)and the Autoregressive Integrated Moving Average(ARIMA),the error of KMeans-GRU model was the lowest,and it’s average prediction accuracy has been increased by 3.43%,6.39%,and 8.34%,respectively,which proves that the KMeans-GRU prediction model has higher prediction accuracy and stronger stability and generalization ability.