Imbalanced Data Classification And Its Application In Wastewater Treatment System

Fault diagnosis by machine learning techniques is widely applied in wastewater treatment plants.A key factor influencing the accuracy of fault diagnosis lies in the imbalance between the sample data in faulty situations and that in normal situations,which may cause misjudgments of faults and lead to failure in practical use for traditional machine learning techniques.Operation faults in biochemical wastewater treatment process often result in serious issues such as effluent water below quality specification,high operation cost,and secondary environmental pollution,therefore spontaneous and accurate diagnoses are required.Based on wastewater treatment process,several novel algorithms have been proposed from changing the samples distribution,a fast relevance vector machine and weighted extreme learning machine to deal with the imbalanced data problem.The proposed off-line model and on-line model are successfully applied in fault diagnosis in wastewater treatment process.Simulation results verify the effectiveness of the proposed method.Main content of this paper is as follows:First,this paper describes the development of imbalanced data classification and fault diagnosis of wastewater treatment and studies the problem of fault diagnosis in wastewater treatment.The causes of difficulties in imbalanced data classification are analyzed in later.At last,the typical methods and performance index are introduced in detail.Then,the principle of fast relevance vector machine and its classification model are introduced.Focus on the difficulties of fault diagnosis induced by imbalanced characteristics of the process data in wastewater treatment,this study proposes a novel pre-processing method with a fast relevance vector machine reducing the data of majority class samples and the synthetic minority over-sampling technique expanding the minority class samples.The imbalance of datasets is minimized.Then the pre-processed datasets are used to build fault diagnostic model.Simulation experiments showed that this fault diagnostic model has higher measuring precision in minority class with a satisfactory total classification accuracy.It is of practical significance to fault diagnosis in wastewater treatment plant.Finally,according to the needs of working real-time and high classification accuracy of fault diagnosis in wastewater treatment,a novel online fault diagnostic model for wastewater treatment process was proposed,i.e.,the kernel-based weighted extreme learning machine.Based on extreme learning machine theory,weighting scheme was used to resolve the data imbalance and the non-linear mapping of kernel function was used to improve the extent of linear separation for a higher accuracy of classification.The use of extreme learning machine is learning without repeatedly adjust the parameters,which can speed up online learning speed.Simulation experiments showed that this online fault diagnostic model has higher measuring precision,better generalization ability,and faster online updating speed,and meet the requirement of accuracy and spontaneity.Therefore,the proposed method can be applied in real-time on-line fault diagnosis in wastewater treatment process.