Detecting Traffic Congestion Incidents Based On Time Series

Traffic safety, traffic congestion and environmental pollution are the three majorproblems plagued the field of international transportation today. For China, with theyearly surging of automobile amount, traffic congestion becomes the most importantissue to be solved. The intelligent transportation system, which is based on trafficflow parameters forecasting and traffic incident detection, is recognized as the bestway to solve these three problems. Therefore, the study of traffic flow parametersforecasting and traffic incident detection have important theoretical significance andpractical value.The main contents and results of this paper are as follows:(1) The mathematical proof that the low-order non-linear transformation canimprove the signal-to-noise ratio is given. Numerical experiments show that thelow-order non-linear transformation not only can be used to estimate the "trend" ofcomplex time series, but also have the following three advantages:â‘ Improve thesignal-to-noise ratio of the input signal;â‘¡Reduce the influence of the "bad" datum inthe complex time series;â‘¢Improve the decomposition between noise and signal.Meanwhile, comparison of the results of autoregressive model, moving average model,exponential smoothing model and combined forecasting model is given by theselected time series forecasting indicators. The research belongs to stationary methodresearch of "mechanism+identification" prediction strategy.(2) The impact of the low-order non-linear transformation (power functiontransformation) on wavelet threshold denoising is studied. Experiments show thatpower function can restrain the sensibility of wavelet threshold denoising to outliers.(3) A modified traffic incident detection method based on time series, theprinciples and steps of realization are given. The probability of traffic congestionincident is also given through example.