Online Output Estimation for Multimode Process with Dynamic Time-delay

There is a time-delay problem between the correlated process variables of the industrial. It is very important to get accurate time-delay for estimating hard-to-measure variable. The existing methods for solving time-delay need to obtain both easy-to-measure variables and hard-to-measure variable. But the hard-to-measure variable cannot be directly measured due to some technical and cost reasons. To solve this problem, a random forest regression model combining sliding window and maximum information coefficient method is proposed. The on-line estimation of the time-delay between the easy-to-measure variables and the hard-to-measure variable is realized. The process variables are reconstructed by the time-delay parameters. The reconstructed process variables are used to build the soft sensor. Finally, the hard-to-measure variable can be estimated by soft sensor accurately. This method is applied to the process variable estimation of nitric acid produce process by dual-pressure method. The effectiveness of the method is verified by experiment results.