Prediction of CO2 capture capability of 0.5 MW MEA demo plant using three different deep learning pipelines

•Machine learning was formulated using plant data obtained from a 1,000-hour operation.•Three DNN pipelines were developed to predict performances of the CO2 capture.•L-DLP showed the optimal performance in terms of accuracy and computational cost.•Result is useful for predicting the CO2 capability from unsteady to steady states.•The approach can be used for precise prediction of other processes at low computing cost. Efficient CO2 capture from off-gas power plant is attracting increased attention due to its significant effects on global warming and climate change. In this study, three deep learning pipelines (DLPs), namely, Spearman DLP (S-DLP), stepwise backward elimination DLP (SBE-DLP), and Lasso DLP (L-DLP), were developed to predict the CO2 capture potential of amine-based capture processes. Raw operating data, obtained under various conditions in a 0.5 MW MEA demo plant, were used for the DLP, which mainly comprised data gathering and cleaning, feature selection, and deep neural network-based prediction. The outliers, which strongly influence the prediction accuracy, were eliminated from the initial raw data. The clean data was then used to predict the CO2 concentration of treated gas and capture rate using the three DLPs. Based on accuracy and computation cost, L-DLP was selected as the best pipeline to predict the CO2 capture rate and CO2 concentration of the treated gas from the 0.5 MW MEA demo plant. The L-DLP was then used to predict the temperature variation in the absorber and stripper, which is a significant control variable to save energy. The deep learning pipeline represents a feasible strategy to predict the CO2 capture potential of the 0.5 MW MEA demo plant, even though the operation of the plant varied from unsteady to steady states under various conditions. The developed pipeline has a great potential for the utilization for other chemical processes with high accuracy in a low computation time.