Calibration method of multi-parameter compensation for optical dissolved oxygen sensor in seawater based on machine learning algorithm. (October 2022)

• Record Type: Journal Article
• Authors: Zhang, Ying; Zhang, Yingying; Yuan, Da; Zhang, Yunyan; Wu, Bingwei; Feng, Xiandong
• Is Part Of: Deep sea research. Volume 188(2022)
• Publication Date: 2022-10
• Subjects: Calibration model -- Machine learning -- Multi-environmental factor -- Nonlinear coupled mapping -- Optical dissolved oxygen sensor; Oceanography -- Periodicals; Océanographie -- Périodiques; 551.4605
• DOI: 10.1016/j.dsr.2022.103856
• Languages: English
• ISSNs: 0967-0637

Abstract

Abstract: Measurement accuracy of optical dissolved oxygen sensor in actual seawater field is affected by many environmental factors. Traditional sensor calibration method is carried out by combining laboratory calibration experiment with field environmental factor correction. First, the influence of temperature on the sensor measurement results is added to the laboratory calibration experiment and model research. The measurement results are then corrected according to salinity and depth environmental factors in the field measurement. However, the influence of these environmental factors on the dissolved oxygen measurement results presents a multi-parameter nonlinear coupling characteristic. Nonlinear machine learning method is used in this study to investigate coupling mapping relationship between seawater multi-environmental factors and measured data of the sensor. A calibration device for controlling temperature, salinity, hydrostatic pressure, and dissolved oxygen concentration conditions and a matching multi-parameter calibration experimental process is designed and established. Analysis of the coupling relationship between multi-parameter conditions and their influence on the measurement of dissolved oxygen concentration showed that the established multi-parameter compensation calibration model achieves a calibration error of ±1 μmol L −1 . Compared with existing compensation results after independent calibration of multi-parameters, the proposed model demonstrates better accuracy and rationality. The calibration sensor is tested in the laboratory and seawater field to verify the validity of the calibration model. Dissolved oxygen concentration measured by the sensor is basically consistent with reference measurement data and Winkler analysis value. Results showed that the multi-parameter calibration method based on machine learning can meet calibration requirements of the dissolved oxygen sensor in seawater with improved scientific rationality and calibration efficiency. Furthermore, the proposed method can be extended to other in situ measurement sensors in seawater while considering various environmental factor corrections. Highlights: New multi-parameter calibration compensation method for optical dissolved oxygen sensor in seawater. The comprehensive calibration of temperature, salinity and hydrostatic pressure was investigated by the calibration device. The multi-parameter calibration model of the sensor was tested in the laboratory and seawater field.