Short-term Accumulations Forecasting Using License Recognition Data

Lei Jin, Binglei Xie

The emerging accumulation-based emerging perimeter control strategy in the context of Macroscopic Fundamental Diagrams (MFD) need to predict accumulations dynamically. The objective of this study is to present short-term accumulations forecasting problem. Forecasting models and its inputs vectors of traffic accumulations should be studied firstly. To avoid the drawback of fuzzy descriptions for link traffic states derived by traffic arrivals, this study newly proposes a method to monitor the dynamics of link accumulations in the context of MFDs. This method yields a full description of the MFD by relating the number of circulating vehicles (accumulations) to network flow (arrivals). The precise traffic data are extracted from discrete records of automatic license plate recognition database. In addition, this paper studies possible applications and accuracy levels of four machine learning models for short-term accumulation forecasting: back-propagation neural network (BPNN), wavelet neural network (WNN), radial basis function neural networks (RBFNN) and support vector regression (SVR). WNN and BPNN models are found to be adaptive and have accuracy levels only a sixth that of RBFNN and SVR models.