Imbalanced Learning of Weighted Extreme Learning Machines Ensemble Algorithm in Wastewater Treatment Plant Fault Diagnosis

Abstract

Fault diagnosis by machine learning techniques is especially crucial to maintain the stability of the water processing flow in wastewater treatment plants (WWTPs). A key factor influencing the accuracy of fault diagnosis lies in the imbalance data distribution between majority classes (i.e., normal situations) and minority classes (i.e., faulty situations), which may cause misjudgments of faults and lead to failure in practical use. This study proposes an ensemble classification method called AdaWELM ensemble algorithm for fault diagnosis in wastewater treatment which builds up individual classifiers by using weighted extreme learning machine (WELM) and combines them with Adaboost ensemble algorithm. The weight matrix in WELM can be updated adaptively along with the iteration of Adaboost learning. The simulations based on 25 benchmark datasets from KEEL repository are arranged at first and the results show the proposed classifier outperforms the other comparative classifiers for most of the evaluated datasets. And then, a practical fault diagnosis model in wastewater treatment plant based on AdaWELM ensemble algorithm is built and the simulations verifies that, comparing with several classifiers, this fault diagnosis model can efficiently improve the performance of WWTP, especially improve the accuracy of faulty class and G-mean value. So we think the proposed AdaWELM ensemble method could be a promising solution for imbalanced data classification in WWTPs.Through experiments, we find that the performance of G-mean value and accuracy of minority class in a case study of WWTPs are improved by comparing with several other related classification algorithms.