A Soft-Sensor Software Application for Environmental Data: Examples From Water Resources
Prof. Paul Anderson, Prof. Jamshid Mohammadi , and Prof. Brent Stephens of the Civil, Architectural and Environmental Engineering department will be hosting a seminar featuring Dr. Jun-Jie Zhu. The topic of the seminar will be A Soft-sensor Software Application for Environmental Data: Examples From Water Resources.
Although data acquisition is critical to provide an effective process controls in water and wastewater treatment industries, much of the required information is difficult or expensive to measure by conventional hard sensors. To address this issue, we have been investigating an alternative data acquisition method based on soft sensors, which can be used to predict needed information based on historical data and easily-acquirable real-time information. Advantages of soft sensors include low cost, fast response times, and the ability to work in parallel or integrated with hard sensors to enhance process control reliability. Soft sensors also have other advantages that conventional hard sensors do not have, including the ability to predict future information, the ability to detect measurement errors, and adaptive learning. Proper management of missing data is critical to soft-sensor applications, but current data management methods can affect covariance and correlation (such as replacement methods) and some of them typically require intensive computations (such as imputations). We developed a simple and effective approach, iterated stepwise multiple linear regression (ISMLR), to evaluate and retain appropriate data for use in an MLR prediction model (Zhu and Anderson, 2016; Zhu et al., 2018). From that initial work, a MATLAB-based ISMLR software application (Zhu, 2017) was recently developed, featuring reduced computation time and options that allow users to adjust parameters for different conditions. This presentation features one example for predicting the current day’s influent ammonia (R2 = 0.84) at the MWRDGC Calumet WRP, and one example to predict future UVA (R2 ~ 0.94) in the Illinois Fox River.