Abstract:This paper aims at the problems of short prediction time and untimely early warning of current prediction algorithms.By using Extreme Learning Machine (ELM) algorithm,it builds a time-series forecasting model and constructs a forecasted value slope trend by least squares fit,using Gaussian mixture clustering to obtain dynamic security trend thresholds,combining predicted slope trend and dynamic safety trend threshold to realize slope trend early warning.The results show that the ELM algorithm has better early warning ability than modeling methods such as GRU neural network,and the slope trend early warning can detect abnormal changes in runtime earlier,and achieve accurate and timely early warning.