Comparison of adaptive neuro-fuzzy inference systems (ANFIS) and support vector regression (SVR) for data-driven modelling of aerobic granular sludge reactors

Maintaining stable operation of aerobic granular sludge (AGS) reactors is a challenge due to the high sensitivity of the biomass to a wide array of parameters, and the frequent changes in influent characteristics. The application of artificial intelligence in AGS modelling has shown promising results but is still at its early stages. This work investigated, for the first time, the capabilities of two artificial intelligence algorithms for the development of predictive models for AGS reactors based on influent characteristics and operational conditions: adaptive neuro-fuzzy inference system (ANFIS), and support vector regression (SVR). The model structure adopted a two-stage modular approach. The models predicted the performance of the reactors for the unseen data with an average R², nRMSE, sMAPE, and MASE of 91 %, 0.21, 0.06, and 0.22 for ANFIS, and 99 %, 0.07, 0.006, and 0.01 for SVR. The results showed that SVR provides more accurate predictions than ANFIS, in terms of prediction errors, and correctly predicting different types of failures. However, ANFIS provided better generalization ability than SVR. The results of this study showed the potential of artificial intelligence for the development of predictive models for the AGS process and provided insight into the selection of the appropriate algorithms for these models.