基于WMNPE间歇过程监测的改进SVDD算法

Improved SVDD algorithm based on WMNPE batch process monitoring

间歇过程数据包含表征过程变化的相关信息和非相关信息,并且呈现高斯与非高斯的多分布等特点.为了更加充分地提取数据的有用信息和处理数据的非高斯性等问题,实现有效的过程监控,提出一种基于WMNPE间歇过程监测的改进SVDD算法.首先运用多向邻域保持嵌入(MNPE)算法来提取低维子流形以实现降维;再使用概率权值策略来提取表征过程变化的相关信息,通过Greedy方法提取低维子流形的特征样本;最后以支持向量数据描述(SVDD)方法建立监控模型进行监控.通过青霉素发酵过程仿真平台验证了所提算法的有效性.

The batch process data contains both the relevant and irrelevant information,which characterizes the process change,and exhibits both the feature of Gaussian and non-Gaussian distribution.In order to extract the useful information of the data more sufficiently and treat the non-Gaussian problem of the data,an improved SVDD algorithm is proposed based on WMNPE batch process monitoring.Firstly,the multiway neighborhood preserving embedding(MNPE)algorithm is used to extract low-dimension submanifold for dimension reduction.Then,the probability weighting strategy is applied to extract the related information of the process change and Greedy method is applied to extract characteristic samples of low-dimension submanifold.Finally,the support vector data description(SVDD)method is used to establish monitoring model and carry out the monitoring.It is verified by means of penicillin fermentation process simulation platform that the algorithm proposed in this article will be valid.