Research On Urban Traffic Flow Prediction Based On Mode Decomposition Algorithm And Deep Learning

With the continuous acceleration of urbanization and the expansion of transportation networks in China,traffic congestion and harmful gas emissions have become important problems faced by many cities.Short-term traffic flow prediction is an important issue in the field of transportation,the accurate and effective short-term traffic flow prediction results are crucial for alleviating road congestion and reducing harmful gas emissions.Currently,the common research methods of traffic flow prediction ignore the spatial-temporal correlation and data noise of traffic flow data.Therefore,this paper proposed a combination model based on decomposition algorithm and deep learning to conduct analysis and research on urban road network traffic flow by establishing a composite model.The main works of this paper are as follows:(1)Aiming at the problems that the urban traffic flow data has strong stochastic fluctuation and large amount of noise,which lead to the decline of prediction accuracy,a combined model of EEMD+Bi GRU was proposed.First,the Ensemble Empirical Mode Decomposition(EEMD)algorithm was used to decompose the original traffic flow data,and several Intrinsic Mode functions(IMF)and a Residual component(Res)were obtained;Secondly,according to the IMF components,the noise energy map was plotted to separate the noise from the original data.Finally,the remaining IMF components were input into the Bidirectional Gated Recurrent Unit(Bi GRU)for training,so as to learn the temporal features in the sequence and output the forecast results.Experimental results show that compared with the Long-Short Term Memory(LSTM)and EMD-based combination models,the EEMD+Bi GRU model has better prediction performance.(2)Aiming at the problem that EEMD method can not completely neutralize the added noise in the decomposition process,which leads to poor reconstruction of decomposition results,and the accuracy of traffic flow prediction also reduces,a combined model MEEMD-DBA(MEEMD-DBi LSTM-Attention)was proposed.First,the EEMD algorithm was compared with the Modified Ensemble Empirical Mode Decomposition(MEEMD)from several perspectives,and the MEEMD algorithm was used to decompose the original traffic flow data to obtain several IMF components and a Res component.Secondly,the noise was separated by calculating the energy value of IMF.Then,the remaining IMF components were input into Double-layer Bidirectional Long-Short Term Memory(DBi LSTM)for training.The Attention Mechanism was added in the training process to pay attention to important features in the data,and finally the traffic flow prediction results were output.The experimental results show that the proposed MEEMD-DBA combined model has a good prediction effect in short-term traffic flow prediction.(3)Aiming at the problems that the parameters of the mode decomposition algorithm are difficult to determine,the spatial-temporal matrix of the short-term traffic flow prediction model contains noise,and the model can not pay attention to the input characteristics of the traffic flow data adaptively,which leads to the decline of the prediction accuracy of the model,a spatial-temporal fusion combination model based on mode decomposition algorithm was proposed.Firstly,the correlation between multiple roads and the road to be studied was analyzed,and several roads with strong correlation were selected.Then the Mayfly Optimization Algorithm(MOA)was used to optimize the parameters of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)to find out the optimal parameter combination.Then the CEEMDAN was used to decompose the original traffic flow data;The decomposed IMF components were used to construct the spatial-temporal matrix,and the Gated Linear Unit(GLU)and Convolutional Neural Network(CNN)were used to extract the temporal-spatial characteristics of the traffic flow data.Combined with the two-stage Attention Mechanism,the feature learning ability of the model was enhanced,and finally the prediction results were output.The experimental results show that the proposed combination model has better prediction performance than other benchmark models.