<p>Traffic congestion becomes critical in enhancing the intelligent transport system because it supports determination of the propagation of traffic congestion, which affects the time expected to be completed, and environmental conservation. However, current solutions may lack necessary accuracy, they might take very long time to solve or they do not incorporate complex traffic patterns properly. In this paper, we introduce a new model called the Transformer-based Temporal Convolutional Gated Network (TransTCGNet), deep learning techniques along with a biologically inspired algorithm from physics known as Quantum Avian Navigation Optimization (QANO). This model uses temporal convolutional layers to extract features from time-series data assimilation; transformer mechanisms are used for long forward dependency on the data. One major enhancement in the current work is the use of QANO for tuning the model hyperparameters that help the model generalize well and reduce the computation time. Moreover, the employment of the positional encodings in the transformer structure strengthens the model in analyzing temporal dependencies inherent in the traffic datasets – the key factor for identifying congestion. As shown from our experimental results, this TransTCGNet is more efficient than the traditional models as it has training accuracy of 99% and as it is tested, it does NOT degrade in performance. Cross-sectional evaluations prove lower miss rates and shorter computational time comparing to previous model that further supports practical implementation of the model for traffic light control in real-life traffic systems. In addition to catering for the challenges posed in traffic congestion prediction, this research also presents propensity for progress in more intelligent modes of transport in urban centers leading to efficient decision making in smart transportation.</p>