Application of a generalized hybrid machine learning model for the prediction of H2S and VOCs removal in a compact trickle bed bioreactor (CTBB).

ABSTRACT

This study presents a generalized hybrid model for predicting H2S and VOCs removal efficiency using a machine learning model K-NN (K - nearest neighbors) and RF (random forest). The approach adopted in this study enabled the (i) identification of odor removal efficiency (K) using a classification model, and (ii) prediction of K <100%, based on inlet concentration, time of day, pH and retention time. Global sensitivity analysis (GSA) was used to test the relationships between the inputs and outputs of the K-NN model. The results from classification model simulation showed high goodness of fit for the classification models to predict the removal of H2S and VOCs (SPEC = 0.94-0.99, SENS = 0.96-0.99). It was shown that the hybrid K-NN model applied for the "Klimzowiec" WWTP, including the pilot plant, can also be applied to the "Urbanowice" WWTP. The hybrid machine learning model enables the development of a universal system for monitoring the removal of H2S and VOCs from WWTP facilities.