基于粗集理论和支持向量机的动态预测新方法及应用

A Novel RS-SVM Dynamic Prediction Approach and Its Application

基于粗集的属性约简理论和SVM回归思想 ,提出了一种内嵌属性约简策略的SVM动态预测方法 (RS -SVM) ,并用于回转窑烧结带温度测量。该方法首先利用属性约简理论精选出与烧结带温度有重要关联的传感器信号 ,再利用SVM建立这些传感器信号与烧结带温度之间的非线性映射模型 ,并不断地跟踪预测误差动态修正SVM预测模型 ,从而提高了系统的抗干扰性能和容错能力。通过与直接SVM方法进行比较的实验 ,说明了此方法在回转窑烧结带温度预测的优越性。

Based on the idea of the attribute reduction of the rough sets theory (RS) and the support vector machine regression (SVM), a kind of RS-SVM dynamic prediction approach is presented and applied to predict the temperature of the rotary kiln sintering process. First, the sensor signal that is closely associated with the sintering temperature are refined by using the attribute reduction theory. Then, a nonlinear imaging model between those sensors signals and sintering temperature is established by utilizing SVM, and dynamically correct the SVM predictive model via continuous tracing the predictive error, thereby, the anti-interference and the fault-tolerant performances have been improved. Through the comparative experiments between the direct SVM approach and the RS-SVM approach proposed in this paper, the results show that the RS-SVM approach has superiority in the temperature predictive task of rotary kiln sintering process.