【摘要】 针对苯氯化反应精馏过程中测量参数存在的高维度、非线性以及噪声干扰的问题,将遗传算法引入到加权邻域保持嵌入(WNPE)的参数选择中,提出了一种故障检测方法。该方法利用改进的遗传算法优选邻域个数和约减维数的参数,再利用WNPE对原始数据进行非线性降维,提取低维流形特征,建立监控统计模型进行故障检测。用Aspen Plus建立苯氯化反应精馏模型,并设置4种故障进行仿真研究,仿真结果表明,所提算法能够有效地检测出故障的发生,其检测精度明显优于PCA、KPCA方法。更多还原

【Abstract】 Aiming at high dimension, nonlinearity and noise existing in benzene chlorination reactive distillation process, Genetic Algorithm(GA) was introduced into the parameter selection of Weighted Neighborhood Preserving Embedding(WNPE), and fault detection method is proposed. In this approach, the numbers of neighborhoods and simplified dimensions were selected by the improved genetic algorithm. The WNPE is used to reduce the nonlinear dimension of original data and extract the low dimensional manifold features, and then the fault detection is carried out. Aspen Plus, a chemical process simulation software, was used to establish the Benzene Chlorination reactive distillation model, and four kinds of fault were set up and simulated. The simulation results indicate that the proposed method can detect fault effectively and have a superior accuracy to PCA and KPCA method.更多还原