AI-Based Traffic Prediction Method for Reducing 5G Terminal Power Consumption

Frequent power wastage occurs when a device maintains its connection during brief idle periods.Instead of focusing solely on the network side, this research explores what the device itself could do by identifying short non-transmission intervals directly from streaming traffic.A lightweight prediction model examines recent traffic patterns and estimates whether a short idle period is likely to occur.When such an interval is detected, the device transmits a release preference via the user equipment assistance information (UAI) framework defined in 3GPP Release 16, thereby requesting a state transition from RRC CONNECTED to RRC_IDLE.Experiments were conducted using a Samsung Galaxy Flip7 in KT's commercial 5G SA network environment, based on a one-hour YouTube streaming session at 480p resolution.The deployed model achieved an accuracy of 87.7% and a recall of 90.2%, reducing average power consumption by approximately 4% compared with the baseline configuration.