Dimethyl ether/Methanol to Olefins (DMTO) is one of the important unit in coal chemical industry, and the distribution of its reaction products can be regulated and optimized by catalyst carbon deposition. Aiming at the disadvantages of time-consuming and high-cost analysis of traditional catalyst carbon deposition measurement methods, a Just-in-Time Learning (JITL) soft sensor model based on Euclidean Distance similarity and Bayesian Regression local model (EDBR-JITL) is proposed for DMTO catalyst carbon deposition in this paper. In the proposed model, the mapping relationship between process state, operating parameters, and catalyst carbon deposition is established, and catalyst carbon deposition is predicted at sampling time by a linear local model (BR) which is obtained by the ED similarity. The results of an industrial DMTO example show that the mean absolute error (MAE) between the predicted value of EDBR-JITL online and the real value of measurement offline for catalyst carbon deposition is 0.10, the mean absolute percentage error (MAPE) is 1.46%, the coefficient of determination (R 2 ) is 0.92, and the prediction time is only in milliseconds. In summary, catalyst carbon deposition in DMTO process is not only predicted and monitored online accurately and conveniently by the proposed EDBR-JITL method, but also is regulated and optimized as the linear BR local model can return the weight of each input parameter at each sampling time.
AIXIE/DMTO-Catalyst-Carbon-Deposition-EDBR-JITL
Dimethyl ether/Methanol to Olefins (DMTO) is one of the important unit in coal chemical industry, and the distribution of its reaction products can be regulated and optimized by catalyst carbon deposition. Aiming at the disadvantages of time-consuming and high-cost analysis of traditional catalyst carbon deposition measurement methods, a Just-in-Time Learning (JITL) soft sensor model based on Euclidean Distance similarity and Bayesian Regression local model (EDBR-JITL) is proposed for DMTO catalyst carbon deposition in this paper. In the proposed model, the mapping relationship between process state, operating parameters, and catalyst carbon deposition is established, and catalyst carbon deposition is predicted at sampling time by a linear local model (BR) which is obtained by the ED similarity. The results of an industrial DMTO example show that the mean absolute error (MAE) between the predicted value of EDBR-JITL online and the real value of measurement offline for catalyst carbon deposition is 0.10, the mean absolute percentage error (MAPE) is 1.46%, the coefficient of determination (R 2 ) is 0.92, and the prediction time is only in milliseconds. In summary, catalyst carbon deposition in DMTO process is not only predicted and monitored online accurately and conveniently by the proposed EDBR-JITL method, but also is regulated and optimized as the linear BR local model can return the weight of each input parameter at each sampling time.
Python