Short-time Traffic Flow Combination Prediction Method Based On Time Series Decomposition And Multi-feature Fusion

The current road infrastructure has been difficult to meet the growing traffic demand.The huge road pressure makes traffic congestion and traffic accidents occur frequently.The construction of an advanced Intelligent Transportation System(ITS)is one of the effective measures to alleviate traffic problems.Traffic flow prediction is the key link of ITS to provide active service and the basis for traffic managers to implement guidance and control.With the rapid development of emerging technologies such as traffic big data analysis and artificial intelligence,massive data sources and advanced prediction methods are provided for accurate traffic flow prediction.The research on short-term traffic flow prediction at home and abroad is usually based on the optimization model,and the factors affecting traffic flow changes are less analyzed,which makes it difficult for the model to effectively extract the key features of traffic flow.Due to the interference of many factors and the short acquisition time interval,the traffic flow shows strong volatility.The traditional prediction method cannot model and analyze its deeper nonlinear characteristics,and the single model has limited ability to extract the potential characteristics in the traffic flow,resulting in the prediction accuracy cannot meet the actual traffic scene needs.Therefore,because of the above problems,combined with feature mining and intelligent prediction theory,taking short-term traffic flow as the research object,the following research is carried out :Firstly,to ensure the quality of traffic flow data and weather data used in the study,the missing values in the data are filled.The characteristics of traffic flow are analyzed from the time and space dimensions,and the influence of weather factors and holidays on traffic flow is fully considered.The correlation between the characteristics is quantified by Pearson correlation.Then,the Pearson correlation coefficient is used to select the feature variables that are more relevant to the target traffic flow from the feature set.After the features are refined through the time window,the Maximum Relevance and Minimum Redundancy(MRMR)are used to analyze their impact on the traffic flow at the predicted time,and the importance of the features is sorted.The optimal feature subset of the traffic flow is selected by the experimental method,which reduces the feature dimension while ensuring the prediction accuracy of the model.It is input into the Extreme Learning Machines(ELM)model for prediction,and the improved Sparrow Search Algorithm(ISSA)is used to optimize the weights and biases in ELM.The results show that the short-term traffic flow prediction model based on MRMR feature selection and ISSA-ELM can effectively mine multiple features in traffic flow to improve the prediction accuracy,and its prediction effect is better than other comparative prediction models.Finally,to further improve the accuracy of traffic flow prediction,a short-term traffic flow prediction method based on time series decomposition and a Q-learning combination model is proposed.The traffic flow is decomposed into trend,period,and residual sequences with different characteristics by using the time series decomposition method.The residual sequence with strong volatility is decomposed by variational mode decomposition to improve the predictability of the traffic flow sequence.The optimal feature set of the reconstructed sequence is selected as the input of the model.Several basic models are constructed to predict the sequence set respectively.The Q-learning method is used to optimize the weight of the prediction results of each model,and the final prediction value is obtained by weighted summation.The traffic flow in the real scene is used for example verification,and the portability of the model and its role in traffic guidance are analyzed.The results show that compared with other models,the established model has a better prediction effect on road traffic flow and higher prediction accuracy.It is also applicable to domestic urban road scenes.