Short-term intersection traffic prediction based on quadratic decomposition framework and two BiLSTM hybrid models

Accurate and reliable short-term traffic flow forecasts play a vital role in improving traffic efficiency, optimizing traffic layout, improving public travel experience, and improving emergency management capabilities. This study provides a hybrid prediction model consisting of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Improved Variational Mode Decomposition (WVMD) to decompose the original sequence into a series of modal functions, solving the problem of a single decomposition method is difficult to fully extract and separate different characteristic information of the sequence. Fuzzy entropy divides the intrinsic mode function (IMF) into high-frequency IMF and low-frequency IMF. In response to the problem of low prediction accuracy and difficulty in effectively predicting sequences with differences in volatility in the study of a single prediction model, we proposed a hybrid model that uses ABiLSTM and CNN-MABiLSTM for prediction. Compared with other eleven benchmark models and commonly used hybrid models on the data set at the intersection, it shows the best results, with high accuracy and applicability.