Kim Nam-Seog, Chung Koohong, Ahn Seongchae, Yu Jeong Whon, Choi Keechoo
Department of Electrical Engineering and Computer Science, University of California at Berkeley, United States.
School of Civil, Environmental and Architectural Engineering, Korea University, Anam-Dong, Seongbuk-Gu, Seoul 136-713, Republic of Korea; California Department of Transportation Operations, Highway Operations, 111 Grand Ave, Oakland, CA 94623, United States.
Accid Anal Prev. 2014 Oct;71:29-37. doi: 10.1016/j.aap.2014.05.007. Epub 2014 May 28.
Filtering out the noise in traffic collision data is essential in reducing false positive rates (i.e., requiring safety investigation of sites where it is not needed) and can assist government agencies in better allocating limited resources. Previous studies have demonstrated that denoising traffic collision data is possible when there exists a true known high collision concentration location (HCCL) list to calibrate the parameters of a denoising method. However, such a list is often not readily available in practice. To this end, the present study introduces an innovative approach for denoising traffic collision data using the Ensemble Empirical Mode Decomposition (EEMD) method which is widely used for analyzing nonlinear and nonstationary data. The present study describes how to transform the traffic collision data before the data can be decomposed using the EEMD method to obtain set of Intrinsic Mode Functions (IMFs) and residue. The attributes of the IMFs were then carefully examined to denoise the data and to construct Continuous Risk Profiles (CRPs). The findings from comparing the resulting CRP profiles with CRPs in which the noise was filtered out with two different empirically calibrated weighted moving window lengths are also documented, and the results and recommendations for future research are discussed.
过滤交通碰撞数据中的噪声对于降低误报率（即对不需要进行安全调查的地点进行调查）至关重要，并且可以帮助政府机构更好地分配有限的资源。先前的研究表明，当存在真实已知的高碰撞集中位置（HCCL）列表以校准去噪方法的参数时，对交通碰撞数据进行去噪是可行的。然而，在实际中这样的列表通常不容易获得。为此，本研究引入了一种创新方法，使用广泛用于分析非线性和非平稳数据的集合经验模态分解（EEMD）方法对交通碰撞数据进行去噪。本研究描述了在使用EEMD方法分解数据以获得本征模态函数（IMF）集和残差之前，如何对交通碰撞数据进行变换。然后仔细检查IMF的属性以对数据进行去噪并构建连续风险概况（CRP）。还记录了将所得CRP概况与使用两种不同经验校准加权移动窗口长度过滤掉噪声的CRP进行比较的结果，并讨论了结果和对未来研究的建议。
