Department of Mathematics

QED Environmental Systems Inc.

Proposed Project for the MSU Industrial Math Students


The domestic solid waste landfill market collects and stores large amounts of data that result from the operation of landfill gas (LFG) collection systems. Data is currently collected from the numerous gas extraction points (well heads) often as frequently as once a week, with a typical landfill gas system containing up to several hundred well heads. This LFG data has time and spatial characteristics and parallel data elements, such as weather conditions, that coordinate in time and apply to the entire site (site meta-data).

QED has created a database that is populated with historical LFG data from several landfills. The data sets vary, but some span multiple years of monthly or bi-weekly collected data. The database also has site meta-data obtained from various weather reporting services. In the spring of 2013 a team of MTH 844 students worked with data sets from the database to test the use of neural network modeling as a possible approach to developing predictive capabilities and automated control recommendations for field system adjustments. This team focused on the development of a stable neural network model, which was accomplished once the data was subjected to time consuming manual data “cleaning”. Team findings suggest that the value of any future neural approaches will be highly dependent on data characterization and data exclusion. The development of well-defined mathematical analysis procedures to speed and architect pre-process steps is an essential component of any future neural analysis. Data mining and three dimensional visualization are the focus of the current problem proposal.

Description of Data Sets:

  1. Currently collected data - data collected by field technicians that visit the well heads and take data on a number of parameters (such as CH4 concentration, applied vacuum, etc.).
  2. Static site/position data - data on various fixed parameters relating to each data collection point (well head), such as GPS location (relative XY position), well depth, elevation, screen length, etc.
  3. Site "meta-data" – other, time-based and site-wide data, such as barometric pressure (and other weather and temperature data) taken from nearby weather stations.

Problem 1: Mining and Characterizing Large Landfill Gas Datasets

Develop data characterization tools that enable pre-processing of large landfill gas data sets in order to group like behaving data points and flexibly identify and sequester data elements as “outliers”, discover potential macro drivers causing groupings, data groupings with statistically significant higher “noise” than the overall data set, etc. Visual data characterization may be needed (see examples of “good” and “bad” data sets, below) and expanded into more rigorous numerical methods.

Problem 2: Three-Dimensional Visualization of Landfill Gas Data

Use 3D and time based data visualization tools to discover hidden data relationships. Are “good” and “bad” data sections spatially and/or time bounded? Complex data relationships are often discovered through graphical maps, such as in this example – 2011 Japan earthquake data visualization –

Top of Page