In real-life freeway transportation system, a few number of incident observation （very rare event） is available while there are large numbers of normal condition dataset. Most of researches on freeway incident detec...In real-life freeway transportation system, a few number of incident observation （very rare event） is available while there are large numbers of normal condition dataset. Most of researches on freeway incident detection have considered the incident detection problem as classification one. However, because of insufficiency of incident events, most of previous researches have utilized simulated incident events to develop freeway incident detection models. In order to overcome this drawback, this paper proposes a wavelet-based Hotelling 7a control chart for freeway incident detection, which integrates a wavelet transform into an abnormal detection method. Firstly, wavelet transform extracts useful features from noisy original traffic observations, leading to reduce the dimensionality of input vectors. Then, a Hotelling T2 control chart describes a decision boundary with only normal traffic observations with the selected features in the wavelet domain. Unlike the existing incident detection algorithms, which require lots of incident observations to construct incident detection models, the proposed approach can decide a decision boundary given only normal training observations. The proposed method is evaluated in comparison with California algorithm, Minnesota algorithm and conventional neural networks. The experimental results present that the proposed algorithm in this paper is a promising alternative for freeway automatic incident detections.展开更多
文摘In real-life freeway transportation system, a few number of incident observation （very rare event） is available while there are large numbers of normal condition dataset. Most of researches on freeway incident detection have considered the incident detection problem as classification one. However, because of insufficiency of incident events, most of previous researches have utilized simulated incident events to develop freeway incident detection models. In order to overcome this drawback, this paper proposes a wavelet-based Hotelling 7a control chart for freeway incident detection, which integrates a wavelet transform into an abnormal detection method. Firstly, wavelet transform extracts useful features from noisy original traffic observations, leading to reduce the dimensionality of input vectors. Then, a Hotelling T2 control chart describes a decision boundary with only normal traffic observations with the selected features in the wavelet domain. Unlike the existing incident detection algorithms, which require lots of incident observations to construct incident detection models, the proposed approach can decide a decision boundary given only normal training observations. The proposed method is evaluated in comparison with California algorithm, Minnesota algorithm and conventional neural networks. The experimental results present that the proposed algorithm in this paper is a promising alternative for freeway automatic incident detections.
