A Novel Wireless Capacity Prediction Algorithm Based on GCN-GRU Combining Surrogate Model-based Optimization

The rapid advancement of wireless networks makes it crucial for base stations to possess self-management and active adjustment capabilities. Accurate prediction of wireless network capacity is an important prerequisite to realizing base station intelligence. In this paper, an improved node capacity prediction algorithm for complex wireless network structures is proposed by fusing graph convolution network and the surrogate model-based optimization model. Firstly, we utilize the Graph Convolution Network-Gated Recurrent Unit model to construct a neural network based on user timing migration prediction under network topology by combining timing and non-timing wireless parameter indicators. Secondly, the surrogate model-based optimization is used to construct a wireless network capacity parameter accuracy calibration model that integrates users, resources, and network relationships. Extensive empirical performance comparisons demonstrate that the proposed algorithms can predict the wireless network capacity more efficiently and accurately.