Research On Clustering Mining For Passenger Flow Time Series

With the rapid development of the Intelligent Transportation System, A lot of traffic flow data is accumulated. It seems to be a problem in the field of traffic control that how to use this massive data to provide meaningful information for traffic planning and control optimization. The data mining technology serves as a powerful tool provides a theoretical basis and technical implementation means to solve this problem. Clustering analysis as an important data mining techniques can found the hidden data distribution and pattern in the huge amounts of data. Using cluster analysis method to analyze massive amounts of traffic time series data of passenger flow has great application value. On the one hand, it can be found the typical passenger flow change rules. On the other hand, it also can be used to find abnormal traffic condition and analyzes the reasons of abnormal appeared. At the same time, it can be compressed data to prepare data for predict and other mining.The traffic time series data of passenger flow with the characteristics of high dimension, high noise and frequent fluctuations, it bring a lot of inconvenience to the series clustering mining. Therefore, how to effectively compress the time series is a key problems need to be resolved in the process of clustering mining. This thesis analyzes the deficiencies and the applicability of the KPS algorithm when use it to deal the volatile traffic time series data of passenger. And then put forward an enhanced time series dimension reduction algorithm EKPS. Finally, through the theoretical analysis and experiments proved that the proposed improved algorithm can keep high compression ratio under the condition of good to keep the original time series morphological change trend, achieved the purpose of effectively reduce the dimension of time series.Afterward key point extraction dimension reduction pretreating of the passenger flow time series using EKPS algorithm in this thesis, choosing the classic FCM clustering algorithm to analysis the series data set based on the fuzzy characteristics of traffic data of passenger flow. Aiming at FCM’s defect of the sensitivity to the initial clustering center, demand of determining the number of clustering in advance and absence of consideration of the distance between classes, the thesis combined with the Hierarchical Clustering algorithm designed a hybrid second-order FCM algorithm. Finally, using the real traffic data of passenger as test subjects to verify the algorithm, the experimental results show that the algorithm can effectively work on the traffic time series data of passenger.