Improved Chemical Process Operations through Data-Based Disturbance Models Eastman gas-phase reactor control system Model Predictive Controller
csp
Identify the minimum number of disturbances from data
∆T Fi
FC
u
To
Ti
x + = Ax + Bu + Gw y = Cx + v w ∼ N(0, Q) v ∼ N(0, R)
Minimum variance estimate of G
c
Convex semidefinite optimization
Gas-phase Reactor
Impose positive definite constraints on variances Q, R Implemented on data from Eastman (above and right) and ExxonMobil Extending to nonlinear models Using with other state estimation methods (particle filtering)
J.B. Rawlings (U. Wisconsin)
p(Y) p(Y)
90
9000
45
4500
0.25
0 −0.25 YC
0
0Y
T
0.25 −0.25
Using original covariances
0.025
0 −0.025 YC
0
0Y
T
−0.025 0.025
Using ALS covariances
2006
1/1