These kinase inhibitors limitations have suggested possible
directions of future research. First, the SOM was developed from data gathered in the afternoon peak period. The SOM’s prototype vectors may not fully cover the entire input space during the off-peak traffic conditions. Second, the SOM was trained with data from one freeway site. It would be interesting to test the transferability of the SOM to other sites. Third, fixed reaction times had been used in the processing of data. It is known that reaction times vary for the same driver and between drivers. That is, reaction time contributes to heterogeneities. However, without assuming fixed reaction times, it was very difficult, if not impossible to proceed with the analyses presented in this paper. Future research should explore a new methodology to estimate reaction time or incorporate reaction time into the SOM’s input or output. Fourth, during training, the number of neurons and the neighborhood radius of SOM are two crucial parameters affecting SOM’s clustering performance. The paper determined these parameters empirically based on the size of data set and the operation speed of computers. Analytical methods need to be further
developed to give a remark regarding how to determine these parameters in a more reasonable manner. Fifth, this research had manually inspected the weight distributions among the neurons (Figure 3) to ascertain the convergence of weights at the end of SOM training. An objective
method of assessing the weight convergence would be helpful in future SOM applications. Acknowledgment This project was supported partially by National Social Science Foundation of China (Major Program, Grant no. 11&ZD160). Conflict of Interests The authors declare no conflict of interests.
The location problem is one of the most studied issues in combinatorial optimization, which is widely applied in communication industry, transportation, and logistics industry. In China, railway freight transport center specifies the railway freight station, which is equipped with various kinds of facilities. Recently, many railway freight transport centers have Dacomitinib been constructed for the purpose of centralized and express transportation. The railway freight transport center location problem is very crucial for the construction of railway freight transport center, which is costly and influential. Many models have been set up to study this problem such as covering model, p-median model, and p-center model [1–4]. Recently, Racunica and Wynter  used two variable-reduction heuristics to solve the hub location problem in intermodal transport hub-and-spoke networks. Jesús and Paula  added a coverage constraint to the p-median model and applied three different algorithms to solve it. Most of the research in literature studied this problem in certain environment. However, many elements in the location problem are fluctuant, especially the transport demand.