| 초록 |
As artificial intelligence (AI) technology has been developing, the interest in its application to the actual system grows. However, unlike in a virtual system, safety, stability, and consistency with physical and chemical facts should be guaranteed as priorities. I developed safe and stabilization-oriented reinforcement learning for optimal system control. The RL algorithm guarantees stability and constraint satisfaction during neural network training. In addition, I established a neural network model using the actual oil plant data focusing on the physical and chemical consistency. Based on the domain knowledge, we need to guarantee the increase of conversion with increasing temperature for the particular actual oil plant. I have enforced the tendency into the neural network model by constraining the values of the weights in the positive region. This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (NRF-2021R1C1C1004217). |
