AI-Driven Energy-Efficient 6G Smart City Networks: Traffic Prediction, Deep Reinforcement Learning, and Zero-Touch Control

Abstract

The rapid evolution of sixth-generation (6G) communication networks is expected to enable dense Internet of Things deployments, connected vehicles, intelligent transportation, immersive services, and other data-intensive smart-city applications. However, the ultra-dense infrastructure required to support these services can significantly increase network energy consumption. Conventional static sleep scheduling and reactive resource management may be unable to adapt effectively to rapid spatial and temporal variations in urban traffic while maintaining quality-of-service (QoS) requirements. This paper proposes an AI-driven framework for energy-efficient 6G smart-city networks that integrates spatial-temporal traffic prediction, deep reinforcement learning, and zero-touch network control. A hybrid convolutional neural network–long short-term memory (CNN–LSTM) model is employed to forecast localized short-term traffic demand, while a proximal policy optimization (PPO) agent uses the predicted demand together with network, radio, computing, and QoS states to determine energy-aware resource-management actions. These actions include base-station activation and sleep control, radio-resource allocation, beam and bandwidth management, and edge-computing offloading. A safety and policy layer constrains the learned actions before their execution through a software-defined networking (SDN) control plane. A closed-loop monitoring mechanism continuously incorporates energy and QoS measurements into the control process. The proposed evaluation methodology uses NS-3, Python, Gymnasium, and Stable-Baselines3 to compare the framework against static, reactive, prediction-based, and reinforcement-learning baselines. Evaluation metrics include total energy consumption, energy per bit, latency, throughput, reliability, block error rate, switching frequency, inference overhead, and policy stability. The framework is designed to provide a reproducible approach for investigating sustainable 6G network operation while maintaining the service requirements of smart-city applications.