Non-linear modeling using fuzzy principal component regression for Vidyaranyapuram sewage treatment plant, Mysore India Ayesha Sulthana, K. C. Latha, Mohammad Imran, Ramya Rathan, R. Sridhar and S. Balasubramanian ABSTRACT Fuzzy principal component regression (FPCR) is proposed to model the non-linear process of sewage treatment plant (STP) data matrix. The dimension reduction of voluminous data was done by principal component analysis (PCA). The PCA score values were partitioned by fuzzy-c-means (FCM) clustering, and a TakagiSugenoKang (TSK) fuzzy model was built based on the FCM functions. The FPCR approach was used to predict the reduction in chemical oxygen demand (COD) and biological oxygen demand (BOD) of treated wastewater of Vidyaranyapuram STP with respect to the relations modeled between fuzzy partitioned PCA scores and target output. The designed FPCR model showed the ability to capture the behavior of non-linear processes of STP. The predicted values of reduction in COD and BOD were analyzed by performing the linear regression analysis. The predicted values for COD and BOD reduction showed positive correlation with the observed data. Ayesha Sulthana (corresponding author) K. C. Latha Ramya Rathan S. Balasubramanian Department of Water and Health JSS University, S.S. Nagar, Mysore-570 015, Karnataka, India E-mail: ayeshasulthanaa@gmail.com R. Sridhar Department of Computer Science, Sri Ramakrishna Mission Vidyalaya, Coimbatore-641020, Tamil Nadu, India Mohammad Imran Department of Information Technology, College of Applied Sciences, Sohar-311, Oman Key words | FPCR, fuzzy-c-means, principal component analysis, TSK fuzzy model INTRODUCTION To overcome the upsurge of water pollution there is a need for effective wastewater treatment; therefore competent modeling and efcient monitoring of wastewater treatment systems are very much essential. The wastewater treatment plant (WWTP) process is a combination of physical, chemi- cal and biological non-linear processes; therefore the WWTP process cannot be modeled by linear statistical approaches (Oliveira-Esquerre et al. ). Understanding the behavior of complex non-linear relations between the process variables, the system parameters, the control inputs and the external perturbations of the WWTP is a dif- cult task. The non-linear behavioral processes are a requirement to develop a model which describes this real- life phenomenon; however, wastewater treatment par- ameters can be employed to construct the model for predicting its performance (Ruicheng & Xulei ). An explicit valid modeling and monitoring technique is an imperative requirement to retain the optimal functioning of wastewater treatment systems. An efcient model intensies the interpretation of biological non-linear pro- cesses and it also constitutes a footing for improved process, operation and control. Statistical data-based modeling approaches like black- boxdo not require a specic mathematical structure of the process to be modeled; the black-box approach has been used to describe the inputoutput non-linear relation- ships of WWTPs (Lee et al. ; Motta et al. ). Articial neural networks (ANNs) are non-linear black- box type models, which have been used to model the exist- ing non-linear relationships between the inuents and to predict the operational parameters of the WWTP (Mjalli et al. ; Mohammed et al. ). ANN models were developed to predict the conduct of a WWTP based on the past information (Hamed et al. ; Farouq et al. ). Optimal coagulant dosage in a drinking water plant was predicted, even in the unexpected conditions such as heavy rain and high turbidity, by ANN models (Sengul & Gormez ). Residual aluminum level in 1040 © IWA Publishing 2014 Water Science & Technology | 70.6 | 2014 doi: 10.2166/wst.2014.333 Downloaded from https://iwaponline.com/wst/article-pdf/70/6/1040/470639/1040.pdf by guest on 30 November 2018