结合LLE的HMM回转窑喂煤量趋势预测研究

Research of LLE-combined HMM on Kiln Coal Feeding Trend Prediction

参照流形学习的非线性变换思想,提出一种结合局部线性嵌入(LLE)的隐马尔可夫模型(HMM)进行回转窑喂煤量变化趋势的预测.LLE-HMM通过LLE对特征数据进行非线性特征变换,接着将变换后的特征数据量化成具体的观测符号值,然后利用HMM建立回转窑喂煤量趋势的预测模型.通过对回转窑生产过程数据的仿真,结果表明与PCA-HMM、ICA-HMM相比,LLE-HMM的测量精度高,跟踪性能好,能满足喂煤量变化趋势预测的要求.

According to the thought of nonlinear transform by manifold learning, presents a hidden Markov model （ HMM ） combine with locally linear embedding （ LLE ） to predict the rotary kiln coal feeding trend. LLE-HMM conduct the nonlinear feature transform on characteristic data through LLE,then quantify the transformed data into the observation symbol value, and use HMM to establish prediction model of rotary kiln coal feeding trend. Through the simulation of rotary kiln production process data, the results show that LLE-HMM have a higher measurement accuracy,better tracking performance compared with PCA-HMM.ICA-HMM, can satisfy the prediction of coal feeding change requirements.