针对变负荷的多工况过程,提出了一种基于分段主元分析的监控方法。对于稳态工况,直接利用历史数据建立不同负荷下的主元监控模型。对于工况之间的过渡过程,根据先验知识可将其划分为跟踪时段和调节时段。在两大时段内分别将训练数据细分为多个子时段,进而在每一子时段内设定参考轨迹,利用训练数据与参考轨迹的残差建立主元监控模型,并采用改进的层次聚类算法合并特性相近的时段。在线监控时,根据负荷设定信息判断过程所处的工况,再选择相应的主元模型进行监控。在Alstom气化炉中的应用结果表明,该算法不仅能够避免传统多模型监控方法在工况过渡时出现的大量误警,也能在过渡过程中实现准确的故障检测。
Aimed at the multimode processes under varying load condition, a novel monitoring approach based on piecewise PCA is proposed in this paper. First, the PCA monitoring model at each steady mode is built using historical data. As for the transition mode, it can be divided into servo phase and regulatory phase through prior knowledge. Then, in both phases the training data are divided into several sub-phases, where the residual between the training data and the reference trajectory are used to establish monitoring model, and an improved hierarchical clustering algorithm is carried out to merge the transition sub-phases of high similarity. When in online monitoring, the mode will be determined by the set point value of load, and the process can be monitored by the pieeewise PCA model. Finally, the monitoring strategy is applied to Alstom gasifier, and the results demonstrate that the proposed method can achieve accurate detection of faults in transition mode.