Quantitative analysis on the oil content of oilfield wastewater based on a convolutional neural network model and ultraviolet transmission spectroscopy

Abstract

Oil content (OC) is one of the important evaluation indicators in oilfield wastewater (OW) treatment. The purpose of this study is to realize online real-time detection of OC in OW by combining ultraviolet spectrophotometry with the convolutional neural network (CNN). In this paper, 80 groups of OW transmission data were measured for model establishment. Three CNN models with different structures are established to generalize the super parametric optimization process of the model. Furthermore, as a common method used in spectroscopy, the synergy interval partial least squares (siPLS) model is built in order to compare its accuracy with the CNN model. The results indicated the CNN model has a better performance than siPLS, in which the CNN model numbered Model 3 has the lowest root mean square error (MSE) of prediction (RMSEP) of 1.606 mg/L. As a consequence, the CNN model can be used in the monitoring of OW. This article will guide a rapid analysis of the OC of OW. HIGHLIGHTS Transmission spectra of the oil wastewater with different oil contents were determined.; Accuracy of the siPLS and CNN models for the prediction of the oil content in the oil wastewater was compared.; The CNN model is more suitable to analyze the oil content of oilfield wastewater than the siPLS model.; The CNN models with different depth structures were compared, and the deepest model exhibited the highest ability of prediction.;

Keywords

• cnn
• deep learning
• oil content
• oilfield wastewater
• uv spectroscopy