基于时序大数据机器学习的状态趋势预警研究

针对目前预测算法的预测时间短和预警不及时等问题，采用极限学习机（ELM）算法，构建了时序预测模型，并通过最小二乘法拟合构建预测值斜率趋势，采用高斯混合聚类得到了动态安全趋势阈值，再结合预测斜率趋势和动态安全趋势阈值实现了斜率趋势预警。结果表明，相比于门控循环单元结构（GRU）神经网络等建模方法，ELM算法具有更好的预警能力，并且斜率趋势预警能够较早发现运行时异常变化，实现准确且及时的预警。

State Trend Early Warning Based on Time Series Big Data Machine Learning

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.