Sensor Data Analysis and Visualization from the Wells at the SRS F-Area using Machine Learning / Deep Learning The traditional method of groundwater monitoring for contaminant transport dynamics based on periodic collection and analysis of samples is not suitable for forecasting and proactive measurement to ensure regulatory compliance. Department of Energy – Office of Environmental Management (DOE-EM) has taken the initiative to develop a new paradigm for long term monitoring…

">

Artificial Intelligence for EM Problem Set (Soil and Groundwater)

Sensor Data Analysis and Visualization from the Wells at the SRS F-Area using Machine Learning / Deep Learning

The traditional method of groundwater monitoring for contaminant transport dynamics based on periodic collection and analysis of samples is not suitable for forecasting and proactive measurement to ensure regulatory compliance. Department of Energy – Office of Environmental Management (DOE-EM) has taken the initiative to develop a new paradigm for long term monitoring of groundwater, which can be automated and suitable for applying Artificial Intelligence.

The overall objective of this project is to determine the most optimal well locations to place Aqua TROLL 500 sensors in the Savannah River Site (SRS) F-Area. The well locations should be chosen with the aim of capturing the overall contamination and groundwater movement.

Objectives:

  • Data Exploration and Pre-processing such as Time series exploration, Resampling, and Interpolation.
  • Spatial Estimation.
  • Optimization algorithm implementation.
  • Majority Voting.

Benefits:

  • Identification of a subset of wells that captures the overall plume dynamic in order of contribution.
  • Capacity to handle multivariate information (multiple analytes and timeframes) through majority voting.
  • Development of a spatial estimation method that is independent from the optimization and therefore interchangeable to suit the needs of the data.

Accomplishments:

  • Developed a suite of preprocessing and visualization tools to understand the pattern of the data for processing.
  • Identified key analytes for the sensor placement optimization by finding the correlation between in-situ sensors and contaminant.
  • Developed a spatial estimation pipeline to optimize its parameters based on the given input.
  • Selection of 13 well locations by LBNL that appeared in the recommended top 20 and another two that appeared in the top 22 to install new Aqua TROLL 500 sensors.
  • Developed and deployed an open-source python package called pyLEnM to analyze contaminated groundwater datasets including sensor placement optimization tools.