 |
Zhongsheng Chen
|
Project DescriptionThe modern process industry is advanced in technology. It has recorded all the huge data of production operation and management decision through DCS, FCS, MES, ERP, LIMS, CRM, SCM, and so on. There is no doubt that the petrochemical industry has entered the Big Data era. Due to the application of computer control system, such as DCS system, it can achieve good system stability control, which make the operation data of the process system is massive but slight in fluctuation, Thus, the data is repeatedly sampled. In addition, as a result of the high cost of data acquisition or the low probability of occurrence, the useful data is limited, such as fault data of production process, disaster data and energy efficiency data. Therefore, industrial operation parameters, operating parameters, control parameters and other data is accumulated daily and monthly. These data have strong nonlinearity, noise, absence and uncertainty, making it difficult to extract effective information and knowledge from a large number of data. This is a small sample size problem in big data. The quality of data-driven models is seriously restricted by the insufficient samples, the unrepresentative samples and ununiformed sample distribution. Hence, it is difficult to realize the optimal operation of the process system. With the development of Big Data, the research of machine learning is rapidly developing, which promotes the research and application of advanced theory and technology. The precision of data-driven models is closely related to the sample size and the sample distribution. Therefore, by expanding the sample size and guaranteeing the distribution and consistency of the samples, it is an important key step to solve the small sample size problem that cannot be ignored in the Big Data era. From the point of view of data-driven modeling, the traditional analysis methods are difficult to adapt to the complicated process system and the complexity of data. As an important data-driven method, neural network technology has drawn more and more attention from researchers in recent years. The research of neural network model structure design, learning speed, generalization ability, stability and robustness become the research focus. Therefore, it is of great theoretical and practical significance to study virtual sample generation and neural network modeling for small sample size problem. ExperiencePublications and PresentationsJournal PublicationsAwards and HonorsEducation
|