Artificial Neural Networks to Determine Ventilation Emissions and Optimum Degasification Strategies for Longwall Mines

1
Artificial Neural Networks to Determine
Ventilation Emissions and Optimum Degasification
Strategies for Longwall Mines
C.Ö. Karacan and G.V.R. Goodman NIOSH,
Pittsburgh Research Laboratory
2
Introduction

• Methane sources are diverse and complex in
• longwall environment
• Ventilation is the primary means of controlling
• methane
• Accurate prediction of ventilation emissions is
• important
• Selection of auxiliary methane control system
• can be a complex task
• Most predictive models require expertise and
• expensive software packages

3
Schematics of Face Ventilation Path
4
Auxiliary Methane Control Measures
5
Objective

• To develop an artificial neural network (ANN)
  based
• prediction and decision tool for
• Predicting methane emissions from US longwall
• mines based on various parameters that may
• have impact on emissions
• Helping the mines in their decisions of
• degasification type choice (N, G, HG, VHG) as
  a
• function of various important parameters that
  
• affect the selection criteria

6
Methodology

• 63 longwall mines from 10 states were analyzed
• between 1985-2005 for emissions and
  degasification
• system used
• Operational parameters
• Geological parameters
• Geographical parameters
• Productivity related parameters

Niosh Research (IC 9067) Longwall census EPA
reports
7
Longwall Mines 1985-2005
State of Mines
Alabama 8
Colorado 4
Illinois 7
Kentucky 4
Maryland 1
Ohio 4
Pennsylvania 10
Utah 4
Virginia 7
West Virginia 14
63 mines from 10 states
8
Methodology

• Variables for ventilation emissions and
  degasification choice

• Formed in 538 data rows as 18 columns

Cut Height
Panel Width
Panel Length Number of Entries
Cut Depth
Face Conveyor Speed Stage Loader Speed Degasification (Yes/No) Coal Production
State
Basin
County
City

Coalbed Name
Lost Desorbed Gas
Residual Gas Content
Total Gas Content Overburden Thickness Seam Height Coal Rank
Sulphur Content
BTU/lb of Coal
Ash Content

Methane Emission Degasification System (N, G, VH, VHG)

9
Schematics of a Longwall Mine
10
Methodology Principle Component Analysis

• Identifying PCs reduces the dimensionality and
  selecting
• appropriate model inputs while retaining much
  of the variance
• In each PC, variables and their weights are
  reported
• Used to determine the relative importance of
  each variable
• on emissions and degasification system used at
  the mines
• In this work, 80 of the total variance was
  selected to remain
• in the data
• PCA revealed that 80 variance is retained in 5
  PCs.

11
(No Transcript)
12
ANN MODELING

• An ANN simulates the cognitive behavior of
  brain.
• Weighted sum of the incoming signals are used
  with an
• activation function.
• A two-layer MLP-ANN type network was designed.
• Input variables from PCA were used
  interchangeably.
• Network parameters then changed.

13
Final Model MLP-ANNVentilation Emission Model

• 9 parameter model with (Total Gas, Panel Width,
  Conveyor,
• Stage L, Seam H, Cut H, Coal Prod, Entries,
  State) was
• determined as best combination.

Ventilation Emission
56 N
38 N
9 inputs
14
Test Result MLP-ANNVentilation Emission Model
R 0.96 R2 0.92
15
Comparison of Emission ANN Model with Statistical
Models
Linear a0a1v1 a2v2.a9v9 ANN
R2 0.54
N-Linear a0a1v1a2v2.
a9v9b1v12b2v22. ANN R2 0.61
ANN R2 0.92
16
Final model for Degasification System Selection
(Classification) Model

• State, Seam H, Cut H, Entries, Panel W, Coal P,
  Total G,
• OB, Emission)

17
Test Result MLP-ANN Degasification System
Selection Model

• Result of test on 81 random, unseen data

Output / Desired Type of Degas (HG) Type of Degas (N) Type of Degas (G) Type of Degas (VHG)
Type of Degas (HG) 21 2 1 0
Type of Degas (N) 0 29 1 0
Type of Degas (G) 1 1 13 0
Type of Degas (VHG) 0 0 0 12

Percent Correct 95.4 90.6 86.7 100
18
Modeling ANN Modeling Modeling of ventilation
emissions from U.S. LW mines and determination of
degasification system (G, HG, VHG, N)
DLLs were generated for MS-Access to distribute
and use the models in any computer without the
need for ANN model builder.
19
Conclusions

• The results of PCA and ANN model search process
  showed
• that ventilation emissions and degasification
  system
• selection could be made by a number of
  variables
• Based on PCA, methane content and mining
  parameters
• are most influential on ventilation emissions
• ANN model of ventilation emissions is more
  accurate than
• statistical models, and may be one of the most
  practical and
• accurate models to predict ventilation
  emissions in US
• longwall mines

20
Conclusions

• Results showed that the degasification systems
  commonly
• used in US longwall mines can be determined
  effectively
• using a classification network.
• The approach and the results suggest that by
  incorporating
• critical stratigraphic features, rather than
  geographical
• information, the models may be applicable to
  other locations
• with different geological layers.