Study Of Tunnei Data Based On Time Series Predicting

In recent years, more and more highway tunnels with monitoring system havebeen built with the country increasing the investment in road infrastructure. Theenvironment data of tunnel collected from the sensors are idle as not being paid enoughattention to. In order to predict the future trend of the pollutant within tunnels, thispaper attempts to study the tunnel environmental data from two aspects: single varianttime series prediction and multivariate time series prediction.Firstly, this paper introduces the basic concepts about time series forecastingmethods briefly and some Recurrent Neural Networks models for times seriesforecasting, particular about the Echo State Networks. According to the characteristicsof tunnel time series, a combination prediction algorithm based on Boosting Algorithmwas proposed. Then the algorithm was used for predicting CO concentration and VIvalue in the tunnel. The results showed that the proposed new method in predictingaccuracy and stability has achieved good results.Secondly, due to the ineffective use of the other tunnel data, a new method basedon Hidden Markov Model Regression used for multivariate time series forecasting wasproposed. Then the model was used for predicting CO concentration and VI value inthe tunnel with multivariate. The results showed that the proposed new method isstability and availability.Finally, the technology of sample bias correction from different distribution wasresearched. A new improved kernel mean matching was proposed after researching thebasic kernel mean matching. The two algorithms were compared in toy data and tunneldata. The results showed that the proposed new method has better results.