Study On Short-term Traffic Flow Prediction Methods For Urban Roads

With the increasing development of technology,traffic congestion has become one of the core issues that the government has focused on.In order to effectively mitigate or even solve this problem,intelligent transportation systems,especially short-term traffic flow prediction,have become the key.The real-time and accuracy of the short-term traffic flow prediction ensure the smoothness of the road and the stable operation of the system,laying the foundation and technical support for solving the traffic problems.Based on the analysis of a large number of short-term traffic flow prediction methods,this paper presents two major prediction models based on the shortcomings of the current short-term traffic flow model and two major upsurges of artificial intelligence: neural networks and deep learning.The specific work of this article is as follows:(1)Construct a short-term traffic flow prediction model based on adaptive GA-Elman neural network.Firstly,the differential data smoothing method is used to deal with traffic flow to eliminate the influence of data trend on the results.Secondly,for the defect that Elman neural network is easily trapped into local extremes,the genetic algorithm is used to optimize;the different implied results of Elman neural network are compared and analyzed.For the prediction error of the number of layers,the optimal number of layers is selected;the number of nodes in the hidden layer is adaptively selected by the number of input samples,so that the best prediction model is obtained.Through case analysis and comparison with Elman neural network,the advantage of this model was confirmed.(2)Construct short-term traffic flow prediction model based on spectral analysis and DBN-Elman neural network.Firstly,the spectral analysis method of traffic flow data is decomposed into the trend part and the stochastic volatility part.The stochastic volatility part is used as the input instead of the original traffic flow data,and the disturbance caused by the trend part is eliminated.Secondly,using deep belief network in deep learning combined with Elman neural network,the best prediction model was obtained after feature learning pre-training and fine-tuning operations;Through comparative analysis of examples,it shows that the model is highly efficient and advantageous.