Leveraging transfer learning with LSTM Gans for adaptive traffic signal control

Traffic congestion has become a persistent challenge in urban areas, leading to significant delays and economic losses. Several Intelligent Transportation Systems (ITS) have been developed to address this issue, but traditional methods for traffic signal decision-making often fall short due to inefficiencies such as excessive delays and energy wastage. To overcome these limitations, this study presents a novel transfer learning-based Long Short-Term Memory-Generative Adversarial Network (TL-LSTM-GAN) model. The system optimizes traffic signal control for priority vehicles in both daytime and nighttime conditions. The proposed system improves traffic conditions, reduces congestion, and enhances energy efficiency by addressing the limitations of current methods. It leverages transfer learning through a ResNet-50 discriminator pre-trained on ImageNet to enhance feature recognition and decision accuracy. An experimental study was conducted using evaluation metrics to compare the performance of the TL-LSTM-GAN model with state-of-the-art methods, and the results demonstrate its superior effectiveness. This application underscores the model's potential to significantly reduce traffic congestion and energy usage, making it a valuable contribution to advanced metropolitan transportation systems.