Deep learning based automatic maintenance of soft sensors used in wastewater treatment plants

Soft sensors in wastewater treatment plants are used to estimate the difficult to measure variables, predict the parameters of concern as per the instructions of the pollution control board or monitor the plant and effluent quality. However the soft sensor deterioration is a limitation of such soft sensors which occur due to various reasons like sudden change in process characteristic or involvement of faulty data in designing and updating the soft sensor. The paper overcome this limitation by using a deep learning based method for automatic maintenance. The proposed technique identify and rectify the faulty data involved in designing and updating of soft sensor based on a ten layered deep neural network. The comparative results show that the proposed method reduces the soft sensor deterioration of a total nitrogen soft sensor used in wastewater treatment plants.