부산대학교 도서관

The monitoring of groundwater levels of hexavalent chromium is a vital task for the US Department of Energy's (DOE) remediation efforts at the Hanford Site, a decommissioned nuclear production facility operated by the Office of Environmental Management. While previous methods have shown promise for accurately modeling contaminants of concern at DOE sites, some contaminants, such as hexavalent chromium, remain a challenge to model due to the high variability and frequency of data collection. Recent Machine Learning (ML) techniques have shown promise to automatically handle these limitations, and regression-based ML models specifically have the potential to overcome these issues due to their ability to detect seasonal trends and patterns unapparent under simple visual inspection. This study focuses on several Autoregressive Integrated Moving Average (ARIMA)-based models, such as traditional ARIMA and Seasonal ARIMA, for modeling hexavalent chromium across 488 wells within the Hanford Site's 100-area from 1997 to 2022. After preprocessing by resampling the data to regular, monthly intervals and performing linear interpolation and normalization, we demonstrate the ARIMA models' predictive capabilities with time-series visualizations on training and testing data, as well as the models' forecasting results through 2024. The collected data sets enhance the ARIMA-based models' understanding of contaminant fate and transport at the Hanford site, which yields more reliable forecasts. There is potential for this data to be used to optimize pump and treat operations in terms of identifying appropriate periods of time and wells to recover the greatest amount of hexavalent chromium from the subsurface.