Research On The Short-term Traffic Flow Prediction Method Based On Multi-scale Decomposition And Deep Learning

In order to further improve the accuracy of short-time traffic flow parameter prediction and provide a more reliable basis for dynamic traffic management decisions,this paper investigates the short-time traffic flow prediction method based on the measured traffic flow data of induction coils.The main work done is as follows.(1)Predictability analysis of short-term traffic flow dateTaking the induction coil measured traffic flow data as the research object,two dimensions of chaotic characteristics and approximate entropy are selected to analyze the predictability of short-time traffic flow data.On the one hand,the C-C method is used to reconstruct the traffic flow time series data in phase space,and the chaotic characteristics of the traffic flow are identified according to the relationship between the correlation dimension and embedding dimension of the new system after reconstruction,and the predictability of the short-time traffic flow time series data is analyzed based on the chaotic characteristics of the traffic flow.On the other hand,the approximate entropy value is used to measure the complexity of traffic flow at different time scales,and the predictability of short-time traffic flow time series data is analyzed based on the approximate entropy value.(2)Multi-scale decomposition method for short-time traffic flow dataIn order to obtain smooth and ordered traffic flow input data,the signal decomposition theory is introduced.Four decomposition algorithms,such as empirical modal decomposition,ensemble empirical modal decomposition,complete ensemble empirical mode decomposition with adaptive noise and variational modal decomposition,are used to decompose the original traffic flow time series data into multiple time series with different frequencies respectively,and the original traffic flow time series data are smoothed without changing the intrinsic characteristics of the data.Finally,the decomposed modal components of the four algorithms are reconstructed separately,and the decomposition performance of the four algorithms is analyzed by the reconstruction error.(3)Short-time traffic flow prediction model construction based on deep learningTwo deep network models,LSTM and GRU,are selected,and the parameters of the number of hidden layer units,batch processing size and learning rate of the two network models are optimized by using particle swarm optimization algorithm,and the short-time traffic flow prediction models based on LSTM deep network and GRU deep network are constructed for each modal component obtained by the four decomposition algorithms,respectively,and the prediction results of each modal The final prediction results are obtained by superimposing the prediction results of each modal component.(4)Example verification analysisIn order to verify the prediction effect of the proposed model,the actual measured traffic flow data of a mega-city highway is used for example verification.The mean absolute error,mean absolute percentage error and root mean square error are selected as evaluation indexes,and the prediction effect of this paper is compared and analyzed with the traditional prediction method,which proves the effectiveness and superiority of the method of this paper.