在线压缩KECA的自适应算法在故障检测中的应用

在复杂的大规模工业过程系统中，实时过程监视、优化计算时间和降低运行内存是实现最终产品质量的最关键和最具挑战性的任务，提出一种在线压缩核熵成分分析（online reduced kernel entropy component analysis， ORKECA）的自适应故障检测算法。首先计算训练样本的核矩阵，根据保留的特征值与特征向量选择有代表性的观测值，构造一个符合全局数据信息特征的压缩集，计算监测统计数据的平方预测误差（squared prediction error， SPE），并利用核密度估计确定控制限。对于在线实时采集的数据，计算该数据的统计量并与压缩集的控制限比较，根据过程状态分析核熵成分分析（kernel entropy component analysis， KECA）模型是否需要进行更新，可以有效提高实时监测过程数据的性能。最后，以一个非线性数值案例及TE过程数据对该方法进行仿真数值分析。结果表明，所提的方法具有有效的可行性。

Application of adaptive algorithm of online reduced KECA in fault detection

In complex large-scale industrial process system, real-time process monitoring, computational optimization and reduction of running memory are the most critical and challenging tasks to achieve final product quality, so an adaptive fault detection algorithm for online reduced kernel entropy component analysis (ORKECA) is proposed. First, the kernel matrix of the training samples is calculated. The representative observation values are selected according to the retained eigenvalues and eigenvectors to construct a reduced set that conforms to the global data information characteristics. The square prediction error (SPE) of monitoring statistical data is calculated, and the control limit is determined by kernel density estimation. For the real-time data collected online, the statistic of the data is calculated and compared with the control limit of the reduced set. Whether the kernel entropy component analysis (KECA) model needs to be updated is analyzed according to the process state. This method can improve the performance of real-time monitoring process data. Finally, a non-linear numerical case and TE process data are used to simulate and numerically analyze the method. The simulation results show that the proposed method is effective and feasible.