针对即时学习(JITL)策略建模速度慢和模型利用效率不高等问题,提出一种改进的JITL策略。改进的JITL策略从两个角度出发:离线阶段使用K均值聚类算法对历史数据进行预分类,相似样本选择的范围由所有历史数据变为对应模态的历史数据;在线阶段结合模型更新策略,通过减少模型更新频率的方式提高模型利用效率,从而间接提高建模速度。把改进的JITL策略用于非线性多模态过程的故障检测,使用即时特征分析(JITFA)算法作为改进JITL策略中的模型计算统计量。将所提方法应用于一个数值例子和一个Benchmark案例,并与JIT-FA等五种不同方法进行对比,验证了所提策略和故障检测方法的优越性。

To address the traditional strategy’s drawbacks such as slow modeling speed and low model utilization efficiency, this study proposed an improved just-in-time learning (JITL) strategy, which approached these problems from two perspectives. At the offline stage, the K-means clustering algorithm was used to pre-classify historical data, and thus the scope of similar sample selection was changed from all historical data to historical data with the corresponding mode. At the online stage, a model update strategy was integrated to enhance the model utilization efficiency and consequently improve the modeling speed by reducing the model update frequency. The improved JITL strategy was applied to nonlinear multimode process fault detection with the just-in-time feature analysis (JITFA) algorithm as the model to calculate statistics. The proposed method was applied to a numerical example and a benchmark case and was compared with five different algorithms including JITFA. The simulation results demonstrate the superiority of the proposed strategy and its corresponding fault detection method.