Bioprocess modeling based on ASM1-SMP model: hybrid modeling approach integrating projection to latent structures (PLS) and artificial neural networks (ANN).

This work presents a new activated sludge model based on ASM1 and soluble microbial product (SMP) kinetics designed to better control fouling and to facilitate integrated simulation of membrane bioreactor for wastewater treatment. The objective is to present a new dynamic mathematical model of activated sludge capable of predicting the formation and degradation kinetics of utilization-associated product (UAP) and biomass-associated products (BAP), operating at different organic load and sludge retention times. Analytical expressions have been developed, based on steady-state ASM1-SMP mass balances, with the inclusion of six additional linear differential equations. The established differential equations are validated using MATLAB and Aquasim. Average deviations (g/L) of the model output ammonia nitrogen (S_NH), nitrate and nitrite concentration (S_NO) and soluble organic matter (SOM) are all below 0.1 g/L. The average values of the results of the deviations between the model simulations ASM1-SMP MATLAB, ASM1-SMP Aquasim and experimental measurements of UAP and BAP are all below 20%, which are 14%, 20% and 21%, for sludge retention time (SRT) of 20, 40 and 60 days respectively. Modeling and predicting SMP using hybrid modeling integrating Project to Latent Structure and an Artificial Neural Network (PLS+ANN) model to correlate them with relevant parameters can significantly improve the output prediction (SMP). The model represented robust predictive performance with an RMSE and R² on independent dataset testing of 0.06 and 0.99 for SRT of 40 days and 0.07, 0.99 for 60 days respectively.