Wavelet embedded Attentive Bi-LSTM for Short-term Passenger Flow Forecasting

Subway Passenger flow forecasting is an essential task of intelligent traffic. In view of the periodic, nonlinear and explosive characteristics of subway passenger flow, a multidimensional passenger flow forecast method is designed to incorporate wavelet transform and deep learning. This method consists of two parts. First of all, the wavelet threshold denoising algorithm is leveraged to smooth the raw data, alleviating imbalanced distribution. Secondly, an attention module is embedded in the Bi-directional Long Short-term memory neural network (Bi-LSTM) to extract the correlation of multi-dimensional time series. Instead of equivalent treatment of all dimensions, the proposed fusion Bi-LSTM method based on Wavelet and Attention recalibrates the dimensional weights in accordance with data properties. The experimental results indicate that the proposed method has preferable feasibility and superiority in numerical evaluation.