Short-term Traffic State Forecasting Study Based On Data Mining

The result of real-time, accurate and efficient prediction is the key to serve city traffic management and public travel. The rapid development of intelligent transport system produces many large, multi-source and heterogeneous traffic data. However, the existing methods of short-term traffic flow forecasting are mainly about the analysis based on time series. Under the complex traffic conditions of urban road, it is necessary to dig out the inherent regularity from the mass of traffic data, so as to effectively serve the short-term traffic flow forecasting. In this environment, it is proposed not only the research background and significance but also the present situation of domestic and foreign. Then the technical route of this paper is completed mainly including three issues about data quality, spatial data mining and time series data mining.The traffic data is necessary to pre-process because of many factors, such as malfunction from traffic detector equipment and uncertainty from complex road network. The error data is recognized through outlier detection algorithm which identify the pseudo-error and combination idea of traffic flow mechanism and threshold for further processing. It can be repaired on the basis of historical data estimation. The missing data can be identified and fulfilled by EM algorithm. At last it can get high-quality data after the adaptive exponential smoothing.According to the temporal characteristics of the traffic flow data, short-term traffic forecasting is presented two research directions differently based on spatial and time series data mining. For spatial data mining, the urban traffic network is divided by using system clustering method. Then it selects the appropriate kernel function and parameter optimization method to achieve the best effect of ?-SVR model. For time series data mining, the research is mainly focused on time series segmentation. By dividing the time series based on SAGA-FCM segmentation method, the traffic data feature extraction is proposed. Through the model of?-SVR to predict the divided periods of time, the experimental examples show the better prediction effect after clustering segmentation.