基于AMW-SVDD的多模态过程故障检测方法

Multimode process fault detection method based on AMW-SVDD

针对传统SVDD方法对多模态过程故障检测率低的问题,提出了自适应滑动窗口-支持向量数据描述(adaptive moving window-support vector data description, AMW-SVDD)的故障检测方法。首先,使用网格搜索方法建立窗宽序列并获取初始窗宽;其次,应用滑动窗口技术将整体数据以窗宽为标准切分到多个子窗口;然后,利用网格搜索方法和粒子群优化(particle swarm optimization, PSO)算法,得到最优窗宽和由各窗口最优SVDD模型构成的模型序列;最后,使用最优模型序列进行故障检测,并将此方法应用于数值例子及田纳西伊斯曼(TE)数据集。结果表明,与传统故障检测方法如KPCA和SVDD等相比,AMW-SVDD方法可有效捕获过程数据的多模态特性。AMW-SVDD方法通过滑动窗口技术捕获数据的局部特征,同时应用PSO算法优化局部模型,二者结合可以自适应确定窗宽参数,进一步获取最优子模型序列,能够有效解决多模态过程故障检测问题,为提高SVDD方法在多模态过程中的故障检测性能提供了参考。

Aiming at the problem of low fault detection rate in multimode process by traditional SVDD method, a fault detection method of support vector data description based on adaptive moving window(AMW-SVDD) was proposed.Firstly, the grid search method was used to establish the window width sequence and obtain the initial window width.Next, the moving window technology was applied to segment the overall data into multiple sub windows according to the window width.Then, the grid search method and particle swarm optimization(PSO) algorithm were used to get the optimal window width and the model sequence composed of the optimal SVDD model in each window.Finally, the optimal model sequence was used for fault detection, and this method was applied to numerical cases and Tennessee Eastman(TE) data sets.The results show that compared with traditional fault detection methods such as KPCA and SVDD,AMW-SVDD method can effectively capture the multimode characteristics of process data.In AMW-SVDD method, the local features of data are captured by moving window technology, and the local model is optimized by PSO algorithm, the combination of the two could adaptively determine the window width and further obtain the optimal model sequence, so that the problem of fault detection in multimode process can be effectively solved, which provides some reference for improving the fault detection performance of SVDD method in multimode process.