Deep Neural Network Architecture Optimization for Edge Computing Based on Evolutionary Algorithms

Abstract

Vehicular Edge Computing (VEC) is a crucial component of Intelligent Transportation Systems (ITS), enabling low-latency and energy-efficient services by offloading computation to the network edge. However, optimizing system performance in such environments requires careful edge server placement, especially in dynamic vehicular contexts characterized by high mobility and unpredictability. Achieving optimal performance under the constraints of latency, energy consumption, and mobility remains a significant challenge. This research proposes a comprehensive framework for optimizing deep learning architectures in VEC, utilizing advanced evolutionary algorithms. Building on real-world vehicular mobility traces, the framework employs the Synergistic Fibroblast Optimized Efficient Deep Neural Network (SFO-Eff-DNN) to identify optimal configurations and edge server placements. The dataset includes details about task offloading under different mobility levels, the data was preprocessed using Min-Max normalization to ensure smooth learning. Among the algorithms evaluated, Synergistic Fibroblast Optimization (SFO) consistently produces well-distributed Pareto-optimal solutions and effectively handles trade-offs between competing objectives. The DNN is utilized to learn complex patterns in vehicular mobility and network conditions, which helps predict the best configurations for edge server placements. The proposed system efficiently minimizes latency and energy consumption while ensuring scalability and adaptability to real-world scenarios. Results demonstrate that SFO-Eff-DNN achieves superior convergence speed and energy efficiency, making it well-suited for time-sensitive deployments. Comparative simulations validate that this approach outperforms traditional methods, providing valuable insights for deploying efficient and robust edge intelligence architectures in next-generation intelligent transportation systems.