Ingredient analysis of biological wastewater using hybrid multi-stream deep learning framework

Abstract

Wastewater treatment and control have received considerable attention. Accurate analyses of ingredients can assist subsequent purification and relevant chemical controls. However, the existing ingredient analysis methods rely on expensive chemical and physical sensors. In addition, maintenance difficulties, and relatively inaccurate analysis results are considerable issues. To overcome these issues, the use of a new and effective deep learning framework has been proposed. The proposed framework considers two types of data: easy-to-measure data, which is preprocessed using a deep neural network module, and past data of a target variable, which is processed using a recurrent neural network. Accordingly, the proposed framework is termed parallel multi-stream deep learning architecture. The proposed multi-stream deep learning framework quantified the relationship between the input and output target variables and enabled time-series analytics. To demonstrate the effectiveness of the proposed framework, its performances are measured using root mean square error (RMSE) and correlation coefficient (CORR). The proposed framework has the lowest RMSE and the highest correlation coefficient in the comparison tests.