Research On Time Series Data Mining And The Application To Traffic Flow Forecasting

The traffic flow forecasting which is critical problem of traffic guiding system and traffic control system in intelligent transportation systems(ITS) leads to great effect on traffic guiding and control. It is the time series analysis problem in nature. There have been many research methods for time series now, and the common one is modeling method. Because of the complexity and unstability of time series, the traditional modeling method is unable to obtain the accurate model of time series system, so forecasting becomes more difficult. Data mining which is developing in recent years is multi-subject data analysis technique and it is applied to various study fields since it emerged. Data mining can be applied to alomost all data, so the good thoughtway is provided to researchers, and at the same time data mining for time series is becoming recent research focuses. There are time variety and non-linear characteristic in traffic flow representing time series data, the satisfied results can't be achieved by traditional method. It is a good idea that utilizes data mining technique to seek the evolution law within the research object. This paper put forward the frame of time series feature pattern mining(TSFPM) on the basis of domestic and international research on time series and put forward the suitable mining algorithm based on characteristic of traffic flow and realized it. The basic thinking of the frame is that translates numerical value time series into symbol series easy to mine and then utilize suitable mining algorithm to mine the symbol series. The patterns which are mined will be cut down according to practicality and forecasting rules will extract finally. This paper provided relevant concepts and definitions of time series feature pattern mining(TSFPM) and analyzed and researched the mining process. In the experimentation with traffic flow data the validity of TSFPM was validated and the impact on performance of the algorithm of the key parameter in algorithms was analysed.