Traffic Flow Labelling for Congestion Prediction with Improved Heuristic Algorithm and Atrous Convolution-based Hybrid Attention Networks

Abstract

The quality of life and the development of urban areas are impacted by traffic-related issues. The delayed response of priority and emergency vehicles, such as police cars and ambulances, jeopardizes public safety and well-being. Further, repeated episodes of congestion affect driver’s temperament by wasting time and causing frustration. Prevailing forecasting techniques are inadequate to address the complexities of urban infrastructure that include autonomous vehicles, connected infrastructure, and integrated public transport. In this article, a new model has been proposed using heuristic methods for real-time traffic management and control applications. The adaptive weighted features are utilized in the atrous convolution-based hybrid attention network for efficient traffic congestion prediction. The features are optimally selected by Mean Square Error of Grass Fibrous Root Optimization (MSE-GFRO) and combined with the optimal weights and thus, are offered the adaptive weighted features. The prediction model combines deep Temporal Convolutional Network (DTCN) and gated recurrent unit (GRU) based on an attention mechanism to predict traffic congestion on the basis of adaptive weighted features. Experimental analysis is performed over distinct optimization models and classifiers to demonstrate the efficiency of the implemented model.