Self-Refined Generative Foundation Models for Wireless Traffic Prediction

TL;DR

This work tackles non-stationary wireless traffic forecasting in 6G by introducing TrafficLLM, a self-refined LLM that operates in-context without parameter fine-tuning. TrafficLLM iteratively improves predictions through a three-step loop of traffic prediction, feedback generation, and prediction refinement, aided by data-to-text transformation and a validation scheme for reliable math. Evaluations on a V2I channel dataset and an urban wireless-traffic dataset show substantial gains over ARIMA and GPT-based baselines, with MAE/MSE reductions up to ~39.27%/53.62% and ~37.33%/40.04%, respectively, and acceptable inference latency around 0.2s. The approach offers scalable, edge-friendly predictive capability leveraging cloud-based LLM inference, with potential for integration with pre-trained models and further enhancements in verification and refinement techniques.

Abstract

With a broad range of emerging applications in 6G networks, wireless traffic prediction has become a critical component of network management. However, the dynamically shifting distribution of wireless traffic in non-stationary 6G networks presents significant challenges to achieving accurate and stable predictions. Motivated by recent advancements in Generative AI (GenAI)-enabled 6G networks, this paper proposes a novel self-refined Large Language Model (LLM) for wireless traffic prediction, namely TrafficLLM, through in-context learning without parameter fine-tuning or model training. The proposed TrafficLLM harnesses the powerful few-shot learning abilities of LLMs to enhance the scalability of traffic prediction in dynamically changing wireless environments. Specifically, our proposed TrafficLLM embraces an LLM to iteratively refine its predictions through a three-step process: traffic prediction, feedback generation, and prediction refinement. Initially, the proposed TrafficLLM conducts traffic predictions using task-specific demonstration prompts. Recognizing that LLMs may generate incorrect predictions on the first attempt, this paper designs feedback demonstration prompts to provide multifaceted and valuable feedback related to these initial predictions. The validation scheme is further incorporated to systematically enhance the accuracy of mathematical calculations during the feedback generation process. Following this comprehensive feedback, our proposed TrafficLLM introduces refinement demonstration prompts, enabling the same LLM to further refine its predictions and thereby enhance prediction performance. Evaluations on two realistic datasets demonstrate that the proposed TrafficLLM outperforms LLM-based in-context learning methods, achieving performance improvements of 23.17% and 17.09%, respectively.

Figures (2)

• Figure 1: An illustrated example of the proposed TrafficLLM applied to uplink throughput (i.e., UL_throughput) prediction. The example prompt presents how to formulate UL_throughput prediction in a question-answer format. Given the input prompt that includes base station location and network measurements for input time, an LLM is expected to predict future UL_throughput for the target time by responding to question outlined in the question prompt. The proposed TrafficLLM automatically generates the feedback prompt by querying the LLM with multiple questions related to the prediction performance. The validation scheme is designed to identify potential inaccuracies during feedback generation and correct incorrect answers. The refinement prompt is further incorporated to allow the same LLM to revise its own predictions by providing actionable instructions.
• Figure 2: Prediction performance comparison between our proposed TrafficLLM and the baselines on the V2I dataset.