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。关键词粗集理论SVM属性约简动态预测回转窑Abstract Basedo nth eid eaof th eat tributere ductionof th eor ughse tsth eory( RS)an dth esu pportve ctorm achineer gerssion( SVM)，aki ndof RS -SVM dynamic prediction approach is presented and applied to predict the temperature of the rotary kiln sintering process. First, the sensor signal that isclosely associated with the sintering temperature are refined场using the attribute reduction theory. Then, a nonlinear imaging model between those sen-。signals and sintering temperature is established场utilizing SVM，and dynarnically correct the SVM predictive model via continuous tracing the predictiveerror, thereby, the anti-interference and the fault-tolerant performances have been improved. Through the comparative experiments between thedirect SVM approach and the RS-SVM approach proposed in this paper, the results show that the RS-SVM approach has superiority in the temperaturepredictive task of rotary kiln sintering process.Keywords Roughs ets SVM Attributesre duction Dynamicp rediction Rotary ki