Solutions
- Neural Networks (Gensym/NOL)
Neural Networks provide a cost effective
modeling tool, and can extend the capabilities
of traditional statistics, modeling and
control. They can be applied in both linear
and non-linear systems where first principles
modeling is costly or difficult.
Neural Networks provide very flexible
and powerful techniques for data analysis,
and can be used for:
Dynamic and Static Process Modeling
Nonlinear and Adaptive Control
Inferential Predictions
Fault Detection
Time Series Prediction
Multivariate Pattern Recognition
Some of our Neural Networks applications
have been:
Caustic Unit: A prototype application,
using actual historical process data and
Gensym NeurOn-line software, provided a
low risk approach for evaluating the applicability
of neural networks (accuracy, historical
data requirements, etc.) in a corrosive
process involving slurries. The prototype
application showed the prediction accuracy
possible and the extent of historical data
requirement.
Gas Lift Well Analysis: This project
involves the monitoring and analysis of
Crude oil wells using the Gas Lift recovery
method. The networks will analyze the Gas
Manifold and Well behavior for Operating
and Production states, and identify patterns
such as Continuous/Intermittent Flow, Control
Valve problems, Lift Gas circulation, Multipoint
injection, etc., in addition to longer term
analysis of water content, emulsion, skin
formation, etc.
TA Oxidation Reactors: In a multiple
reactor process involving oxidation reactors
arranged in series and parallel, neural
networks are being used to estimate the
product qualities (4CBA and Transparency).
By making this analysis available online
on a continuous
basis, the GMAXC multivariable predictive
controller can use these nonlinear calculations
instead of analyzers/lab samples which
have delays in excess of four hours.
