Base stations are large power consumers. Using AI models to accurately predict base station traffic not only helps information and communication infrastructure save energy and reduce carbon emissions, but also reduces operators investment costs and promotes the green and low-carbon development of mobile communications. In order to reduce network operating costs, through statistical analysis of a large number of base station data, machine learning algorithms based on Holt-winters linear model, LSTM regression model, TCN model, Transformer model, Mamba model and improved FECAM-TCN model are used to target different Scenarios are used to achieve comparative prediction of base station cell traffic, and the prediction results are significantly improved using the improved model. Based on the accurate traffic prediction results of the base station, operators can adjust network parameters and resource allocation to ensure stable connections and low latency during peak hours and high-traffic areas. This will improve user experience and reduce network congestion and weak signal issues.