Dual-Scale Cellular Traffic Prediction Based on FastDTW for Graph Neural Network

Cellular traffic prediction is crucial to the network efficiency as it provides prior knowledge to guide the wireless resource allocation. The artificial intelligence (AI) based traffic prediction can provide better prediction performance and has drawn increasing interests. Nevertheless, most existing works utilized Euclidean structure data, which can provide traffic prediction to only certain grid area rather than specific spot, such as each base station (BS). To give more precise BS level prediction, in this work we propose a dual-scale spatiotemporal prediction framework based on fast dynamic time warping (FastDTW) algorithm. The proposed framework can capture both the overall neighborhood relationship of historical traffic and the local neighborhood relationship graph to improve the prediction accuracy. More importantly, due to the utilization of FastDTW, it solves the disadvantage of the high complexity of the dynamic time warping (DTW) algorithm and completes the prediction at the BS level more efficiently. Simulation results confirms the generality and performance of the proposed method, which outperforms the benchmark algorithms.