Machine Learning based Digital Twin Framework for Production Optimization in Petrochemical Industry

•This paper proposes a theoretical framework for digital twin-based production optimization, which integrates industrial IoT data processing and machine learning approaches.•This paper designs a practice loop of information exchange between the physical factory and a virtual digital twin model, as well as digital twin modeling process, system architecture, and model evaluation indices.•This paper provides a concrete solution to time series data processing issues in the petrochemical industry, such as frequency alignment, time lag issues, and high demand for immediacy.•The framework and approach proposed are practiced in the catalytic cracking unit of a petrochemical factory, and the results show the effectiveness of this approach for production optimization.