Detecting Traffic Incident Based On Traffic Flow Analysis

Real-time traffic incident detection can better help to solve traffic congestion, and deal with traffic incidents quickly. When detecting traffic incident based on traffic flow forecasting, we should analyze the traffic flow, and the separation of white noise of traffic flow sequence is also very important, while the study of relatively rapid and effective prediction method and the method for judging traffic incident whether or not happened, are the keys of my dissertation.The main contents and results are:(1) In numerical experiments, under different decomposition levels, we get corresponding noise series via Wavelet Denoising by changing signal frequency fA, decomposition level N, and white noise intensity Q. The chi-square test valueχd2, meanμ, standard deviationσ? of these noise series were taken as the quantitative indicators of denoising. We found that the signal frequency is the main factor that influences the performance of wavelet threshold denoising. The signal frequency can not be more than 7/3980 (there are 7 sine wave periods in 3980 points), which is an unambiguous quantitative description of"low-frequency signal, high-frequency noise". The second important influencing factor is the decomposition level, its range of priority is 3~10 via Wavelet Denoising. The last is the white noise intensity. the available"frequency-noise intensity"range is between the lines The separated noises by these parameters are more similar to white noises via wavelet threshold denoising. Denoising the real highway traffic flow confirm the validity of our numerical experiment results.(2) Study the effects of differential method and the wavelet for traffic flow sequence. Find differential method denoising is superior to wavelet denoising for higher frequency noise on lower frequency signal sequence. Combination of both can separate white noise effectively.(3) Verify the power spectrum denoising is feasible. The power spectrum denoising is better than wavelet denoising, and also can be better without the interference of singularities.(4) Proposed wavelet weighted moving average method, and comparative study with the simple moving average method, exponential smoothing method found that wavelet weighted moving average method is a little bad than exponential smoothing method in the smooth traffic flow prediction, but better than simple moving average method. Wavelet weighted moving average is the best method in increased traffic flow prediction.(5) Proposed a new method for traffic incident identification based on mathematical statistics and robust statistics. Given step, and given the probability of traffic incidents by the example of traffic flow series.