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

Abstract

Abstract: Sixth-generation (6G) communication is expected to support dense Internet of Things deployments, connected vehicles, immersive services, intelligent transport, remote health-care, and other data-intensive smart-city functions. The same ultra-dense radio access infrastructure, massive multiple-input multiple-output arrays, highfrequency radios, and edge-computing capacity that enable these services can substantially increase network energy demand. Static sleep schedules and reactive load balancing cannot follow rapid spatial–temporal variations in urban traffic without either wasting power or degrading quality of service (QoS). This paper reviews energy-aware 6G research and presents a simulation-oriented architecture that combines a hybrid convolutional neural network–long short-term memory (CNN–LSTM) traffic forecaster with a proximal policy optimization (PPO) deep reinforcement learning agent. The forecaster estimates localized short-term load, while the agent selects base-station sleep states, radio-resource allocations, beam and bandwidth policies, and edge-offloading actions. A software-defined networking control plane translates these decisions into zero-touch commands, and a closed feedback loop returns power and QoS measurements as rewards or penalties. The proposed evaluation uses NS-3, Python, Gymnasium, and Stable-Baselines3 to compare the learning policy with static and reactive baselines. Metrics include total energy, energy per bit, latency, throughput, block error rate, reliability, switching frequency, and policy stability. No uncollected performance resultis claimed. The work provides a reproducible research plan for reducing the operational and carbon cost of 6G infrastructure while protecting critical smart-city services.