Short-term traffic flow prediction via weight optimization of composite models

Abstract

ABSTRACT Accurate prediction of peak-hour traffic flow is crucial for congestion management. Neural network models have varying performance in different domains, limiting their effectiveness in traffic data. To address this, we propose the SSA-BiLSTM-BP-Elman ensemble model. It integrates bi-directional long short-term memory (BiLSTM), back propagation (BP) neural network, and Elman network. The model uses the sparrow search algorithm (SSA) to optimize the weight distribution to improve the prediction accuracy. Initially, the BiLSTM, BP, and Elman models are parameter-optimized by the Bayesian approach. SSA then assigns unique weights to each model based on the characteristics of the extracted data features. Results from an application to UK high-speed traffic flow data show that SSA significantly improves the accuracy of model predictions. This integrated model effectively utilizes the strengths of each model by assigning appropriate weighting coefficients by SSA, thus improving the overall prediction performance.