An EWT-EnsemLSTM-LSSA Model for Metro Passengers Volume Prediction

Abstract

Metro passenger volume prediction is an incredibly significant issue when it comes to traffic flow prediction problems. The accurate prediction of passenger flow is not only beneficial but necessary to adjust public transportation systems. As such, metro passenger volume prediction has become a crucial issue in the realm of Intelligent Transportation Systems (ITS). In this paper, a novel model called EWT-EnsemLSTM-LSSA has been proposed to deal with the complex issue of passenger flow prediction. This model assembles empirical wavelet transform (EWT), long short-term memory (LSTM), support vector regression (SVR), and logistic mapping sparrow search algorithm (LSSA) to create a comprehensive and robust solution. To start with, EWT is implemented to decompose the original dataset into five wavelet time-sequence data series for further prediction. A cluster of LSTMs with varying hidden layers and neuron counts is then deployed to scrutinize and exploit the implicit information within the EWT-decomposed signals. Subsequently, the output of LSTMs is integrated into a non-linear regression method SVR. Finally, LSSA is engaged to optimize the SVR automatically. The EWT-EnsemLSTM-LSSA model is put to the test in three case studies, employing data collected from the metro of Minneapolis, America, and Hangzhou, China. The results of these experiments are truly remarkable, as they indicate that the proposed model outperforms its conventional counterparts by reducing the mean average error to 189.27 and the root mean square error to 260.36 in Minneapolis data, and the mean average error to 24.97 and the root mean square error to 41.75 in Hangzhou data.

Keywords

• Traffic volume prediction
• time series analysis
• empirical wavelet transform (EWT)
• long short term memory (LSTM)
• support vector regression (SVR)
• logistic mapping sparrow search algorithm (LSSA)