Download - Advances in Fermentation Machines
-
7/31/2019 Advances in Fermentation Machines
1/11
kNN-RVM Lazy Learning Approachfor Soft-sensing Modeling of Fed-
Batch processes
Jun Ji, Hai-qing Wang*, Kun Chen,and Dian-cai Yang
-
7/31/2019 Advances in Fermentation Machines
2/11
Definiton of Terms:
Fed-batch processes- inherently difficult to modelowing to non-steady-state operation, small-samplecondition, instinct time-variation and batch-to-batchvariation caused by drifting
Soft-sensing approachesare investigated to establish
an online monitor and control of fed batch; needsdevelopment
-
7/31/2019 Advances in Fermentation Machines
3/11
Types of Soft-sensors:
SOFT-SENSOR CHARACTERISTICS
AND LIMITATIONS
Analysis based methodsregression may fail in small-samplecondition and be sensitive to
measurement error
Artificial neural networks(ANN) based methods
difficulty in determiningnetwork complexity
Kernel based methods effective to deal with small-size samples but areinsufficient to trace the time-varying characteristics of theprocess
-
7/31/2019 Advances in Fermentation Machines
4/11
SOFT-SENSOR CHARACTERISTICS ANDLIMITATIONS
Adaptive kernel learningalgorithm (AKL)
effectively updates the modelwith a two-stage recursivelearning mechanism butinaccurate prediction may stillarise in certain local regionwhere instinct time-variationis occurred
Relevance VectorMachine(RVM) need very few kernelfunctions but may not beapplicable to large data sets
-
7/31/2019 Advances in Fermentation Machines
5/11
Relevance Vector Machine(RVM)
given the dataset of input-targetpairs N
standard formulation (batchculture) is followed and p(t|x) isassumed to be Gaussian
Does not use enough kernelfunctions
Does not follow standardformulation for fed batch whichhave numerous data sets
-
7/31/2019 Advances in Fermentation Machines
6/11
Adaptive Kernel Learning
Algorithm (AKL)
Erroneous in time variation steps
Does not consider time changesand takes data sets at one time
Like RVM, does not use enough
kernel sensors
Have slow response time whichlimits its capabilities to detectsignificant changes occuring in
the culture
-
7/31/2019 Advances in Fermentation Machines
7/11
kNN-RVM Lazy Learning Approach
Distance Vector
Indices
-
7/31/2019 Advances in Fermentation Machines
8/11
Characteristics
Able to use enough kernel values
Has quick response time that enables it to detectsmall significant changes in the culture
Able to deal with numerous data sets
Considers time changes and is able to see trendsthrough time
-
7/31/2019 Advances in Fermentation Machines
9/11
Performance of Fermentation Machines usingRVM, AKL and kNN as soft-sensors
-
7/31/2019 Advances in Fermentation Machines
10/11
Today
Latest fermentation machines use kNN as soft-sensors
RVM and AKL are still used however without thesame precision as that kNN could offer
RVM and AKL are still used for cultures which do not
require highly sensitive measurements to maintaintheir growth
-
7/31/2019 Advances in Fermentation Machines
11/11
REFERENCES
http://www.sciencedirect.com/science/article/pii/S0098135409000076
http://www.scientific.net/AMM.20-23.1185
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=59
30437 http://dynamics.org/~altenber/UH_ICS/EC_REFS/GP_REFS/GEC
CO/2004/31031078.pdf