Abstract: Industrial equipment operating at high speeds generates a large amount of dynamic data streams that reflect fluctuations in its production process and operational status. However, due to the complexity of dynamic process big data, it is difficult to achieve a high recognition accuracy for any type of abnormal pattern relying solely on a single type of data feature. The urgent problem to be solved is how to fully integrate, screen, and efficiently utilize information. At present, intelligent technology has become one of the effective methods to solve complex production processes with characteristics such as nonlinearity, multiple inputs, and multiple outputs, due to its advantage of not requiring the establishment of precise mathematical models. Therefore, classifiers used for dynamic process quality anomaly pattern recognition have evolved from early rule-based expert system judgment to the combination of support vector machines, particle swarm optimization algorithms, rough sets, and other technologies. The RS-GA-SVM model proposed in this article is a computationally simple and highly accurate method for identifying quality anomaly patterns, which is beneficial for real-time quality monitoring and fault diagnosis in automated manufacturing industries such as petroleum and chemical.
   In order to effectively reduce the dimensionality of features and improve the recognition accuracy of abnormal patterns in dynamic processes, this paper combines feature reduction with improved support vector machines to identify abnormal patterns in dynamic process quality. Firstly, it uses Monte Carlo simulation to generate a dynamic data stream of the production process, including training samples and test samples. The test samples are mainly used for estimating the performance of SVM. Next, it extracts 16 statistical features and 7 shape features that can characterize quality anomaly patterns from the raw data. Secondly, this article uses rough sets (RS) to reduce the feature set, eliminate redundant and interfering features, and obtain the optimal attribute set. Subsequently, it uses genetic algorithm (GA) to find the optimal parameters for support vector machine (SVM), which can reduce the subjectivity of SVM in parameter selection. Finally, it uses the GA-SVM model to identify quality anomaly patterns, and compares the recognition accuracy with other similar models to verify the effectiveness and applicability of the proposed recognition model.
   The results show that the statistical features and shape characteristics used in this article can combine multiple quality mode features, integrating the classification advantages of different features on different quality modes. Secondly, the 12 features obtained after rough set filtering have strong ability to distinguish whether there are abnormal conditions in dynamic processes, which can effectively reduce the input dimension of support vector machines. Finally, this article uses simulation experiments to compare the recognition accuracy with other models of the same type. The results show that the genetic algorithm optimized support vector machine has significantly higher accuracy in identifying quality anomaly patterns than other models, indicating that the RS-GA-SVM model proposed has good recognition accuracy and robustness, and can effectively monitor dynamic processes.
   In future, due to uncertain factors in the actual production process that may lead to the emergence of new quality anomaly patterns or multiple concurrent quality patterns, it is necessary to conduct in-depth research on these unknown quality anomaly models. Meanwhile, this article only uses the equal frequency interval partitioning method to reduce the combination of serial features, and rough sets also include other attribute reduction methods. Therefore, there is still a lot of research space for feature reduction methods based on rough sets.

Key words: dynamic process, quality anomaly pattern, rough set, support vector machine, genetic algorithm

摘要： 为了有效降低特征维数并提高动态过程异常模式的识别精度,提出基于特征约简与改进支持向量机的动态过程质量异常识别方法。本文首先提取能反映质量异常模式的16个统计特征与7个形状特征,再使用粗糙集(RS)约简特征集合以剔除冗余特征与干扰特征。同时,使用遗传算法(GA)寻找支持向量机(SVM)的最优参数,并采用GA-SVM模型识别质量异常模式。仿真实验表明:粗糙集筛选后得到的12个特征具有较强区分动态过程是否出现异常状况的能力,遗传算法参数寻优后的支持向量机识别质量异常模式的精度明显高于其他同类型的模型,因此,本文提出的RS-GA-SVM模型具有良好的识别精度与稳健性,能够对动态过程进行有效监控。

关键词: 动态过程, 质量异常模式, 粗糙集, 支持向量机, 遗传算法