Extension of traffic flow pattern dynamic classification by a macroscopic model using multivariate clustering

In this paper, we evaluate the performance of a dynamic approach to classifying flow patterns reconstructed by a switching-mode macroscopic flow model considering a multivariate clustering method. To remove noise and tolerate a wide scatter of traffic data, filters are applied before the overall modeling process. Filtered data are dynamically and simultaneously input to the density estimation and traffic flow modeling processes. A modified cell transmission model simulates traffic flow to explicitly account for flow condition transitions considering wave propagations throughout a freeway test stretch. We use flow dynamics specific to each of the cells to determine the mode of prevailing traffic conditions. Flow dynamics are then reconstructed by neural methods. By using two methods in part, i.e., dynamic classification and nonhierarchical clustering, classification of flow patterns over the fundamental diagram is obtained by considering traffic density as a pattern indicator. The fundamental diagram of speed-flow is dynamically updated to specify the current corresponding flow pattern. The dynamic classification approach returned promising results in capturing sudden changes on test stretch flow patterns as well as performance compared to multivariate clustering. The dynamic methods applied here are open to use in practice within intelligent management strategies, including incident detection and control and variable speed management.