Abstract: To improve the performance and resource utilization rate of cloud platforms, this paper proposed a load prediction combination model based on ARMA-CNN-SVR, improved the accuracy of predicting cloud platform resource usage by integrating the advantages of multiple prediction models. Based on this load forecasting combination model, the paper further optimized the elastic scaling strategy, effectively solved the lag problem of resource adjustment, enhanced the initiative and intelligence of the cloud platform, and significantly improved resource utilization rate and service quality.