Deep learning-enhanced task offloading strategy for internet of vehicles: dynamic optimization based on local-edge-cloud collaborative framework

Deep Learning-Enhanced Task Offloading for IoV Connected vehicles face inefficiencies, delays, and high energy use from resource-heavy applications. Existing machine learning-based offloading strategies rely on manual features, limiting adaptation to complex IoV dynamics. This paper proposes a deep learning scheme: CNN-LSTM fuses multi-source data for feature extraction; a deep network with GAN optimizes pricing and resource decisions; online learning adapts to changes. Simulations show local-edge-cloud collaboration with task success >75%, delay reduced by 15%-20%, and better utility balance.