| Summary |
Appropriate traffic management and route planning are effective ways to avoid traffic congestion. The primary goal of effective traffic management is to be able to predict traffic conditions accurately.
To conduct research, this study gathers the vehicle detector data provided by the Kaohsiung City Transportation Bureau and proposes a multi-module deep learning prediction model. The prediction model proposed in this study includes multiple modules, which not only uses historic data to do prediction but take spatiotemporal conditions into consideration. Before training the prediction model cluster analysis method is used to divide traffic states into different groups. Traffic states in different groups regarded are as under different spatiotemporal conditions. Then the corresponding prediction modules are trained for different group traffic states, so that the model can predict with appropriate parameters according to the differences in spatiotemporal conditions. To perform a highly accurate prediction result it is vital that models could capture important temporal and spatial features and to estimate the correspondence between them, so that model can covert features to outputs. |
