基于自编码器的冷轧带材板形数据降维方法

 为实现板带轧制过程的智能制造,对智能化的内涵进行了深入探索。针对具体问题,将无监督学习与强化学习理论用于生产实践具有重要意义。以板带轧制过程中的板形检测数据为研究对象,通过无监督学习理论中的自编码器进行板形基本模式的自动学习,从而降低板形数据的存储与传输量,实现板形分布的抽象表示,为后续板形异常检测、智能预报和智能控制奠定基础。与基于勒让德多项式模式的传统板形数据降维方法相比,此方法可显著提高板形重构精度,实现板形数据的近似无损压缩。

Dimension reduction method of cold rolling strip flatness data based on autoencoder

In order to realize the intelligent manufacturing of plate and strip rolling process, it is necessary to deeply explore the connotation of intelligent manufacturing. For specific problems, it is of great significance to apply the unsupervised learning and the reinforcement learning theory to production practice. The flatness detection data in the process of strip rolling is taken as the research object and the autoencoder in the unsupervised learning theory is used to automatically learn the basic flatness mode, so as to reduce the amount of storage and transmission of flatness data, realize the abstract representation of flatness distribution, and lay the foundation for the flatness anomaly detection, the intelligent prediction and the intelligent control. Compared with the traditional flatness data dimension reduction method based on Legendre polynomial, the accuracy of flatness reconstruction can be significantly improved and the approximate lossless compression of the flatness data can be realized applying the present method.