Attention-based BiLSTM-CNN network for highway visibility prediction

Abstract

Under low visibility meteorological conditions, the traffic safety on highways is significantly threatened. During such situations, highway management authorities often need to implement relevant emergency measures to ensure traffic safety. Therefore, achieving accurate visibility forecasting for highways holds great significance. With the advent of the big data era, deep learning algorithms are widely used in transportation meteorology research and application areas due to their autonomous learning advantage of uncovering into data mapping relationships. In this paper, based on meteorological monitoring data collected from the Zhaotong District Operation Department of Yunnan Communications Investment & Construction Group CO., LTD., we propose an Attention-based BiLSTM-CNN network (ABCNet) visibility prediction model. ABCNet extracts bidirectional temporal features of meteorological element sequences and deep data space features. It utilizes an attention mechanism to adaptively adjust the weights of feature representations, aiming to obtain optimal feature information, thereby achieving performance metrics surpassing competing models. Experimental results demonstrate that ABCNet effectively achieves accurate visibility prediction for highways, carrying substantial practical significance.