Minimum variance control aims to reduce output fluctuations in systems. It uses an ARMAX model to represent system dynamics and derives a control law that minimizes output variance. The approach balances performance with stability and causality constraints.
Generalized minimum variance control expands on this concept. It introduces a control effort weighting factor, allowing engineers to balance output tracking with energy use. This modification improves robustness and reduces noise sensitivity, offering more flexible control strategies.
Minimum Variance Control
Objective function and constraints
Top images from around the web for Objective function and constraints Lagrange multiplier - Wikipedia View original
Is this image relevant?
HESS - Objective functions for information-theoretical monitoring network design: what is “optimal”? View original
Is this image relevant?
Lagrange multiplier - Wikipedia View original
Is this image relevant?
1 of 3
Top images from around the web for Objective function and constraints Lagrange multiplier - Wikipedia View original
Is this image relevant?
HESS - Objective functions for information-theoretical monitoring network design: what is “optimal”? View original
Is this image relevant?
Lagrange multiplier - Wikipedia View original
Is this image relevant?
1 of 3
Minimize output variance J = E [ y 2 ( t + d ) ] J = E[y^2(t+d)] J = E [ y 2 ( t + d )] optimizes control performance by reducing output fluctuations
ARMAX model A ( q − 1 ) y ( t ) = B ( q − 1 ) u ( t − 1 ) + C ( q − 1 ) e ( t ) A(q^{-1})y(t) = B(q^{-1})u(t-1) + C(q^{-1})e(t) A ( q − 1 ) y ( t ) = B ( q − 1 ) u ( t − 1 ) + C ( q − 1 ) e ( t ) represents system dynamics with input, output, and noise terms
Causality ensures control decisions based only on available information prevents impossible future predictions
Stability requirement maintains system equilibrium avoids unstable oscillations or divergence
Control law derivation
Certainty equivalence assumes estimated parameters are true simplifies controller design process
Express future output y ( t + d ) y(t+d) y ( t + d ) using past inputs/outputs separates controllable and uncontrollable terms
Set predictor of y ( t + d ) y(t+d) y ( t + d ) to zero minimizes expected future output variance
Solve for current input u ( t ) u(t) u ( t ) yields control law u ( t ) = − F ( q − 1 ) E ( q − 1 ) B ( q − 1 ) y ( t ) − G ( q − 1 ) E ( q − 1 ) B ( q − 1 ) u ( t − 1 ) u(t) = -\frac{F(q^{-1})}{E(q^{-1})B(q^{-1})}y(t) - \frac{G(q^{-1})}{E(q^{-1})B(q^{-1})}u(t-1) u ( t ) = − E ( q − 1 ) B ( q − 1 ) F ( q − 1 ) y ( t ) − E ( q − 1 ) B ( q − 1 ) G ( q − 1 ) u ( t − 1 )
Polynomial functions F ( q − 1 ) F(q^{-1}) F ( q − 1 ) and G ( q − 1 ) G(q^{-1}) G ( q − 1 ) depend on system model parameters shape control response
Generalized Minimum Variance Control
Generalized minimum variance control
Modified objective J = E [ ( y ( t + d ) − w ( t ) ) 2 + λ u 2 ( t ) ] J = E[(y(t+d) - w(t))^2 + \lambda u^2(t)] J = E [( y ( t + d ) − w ( t ) ) 2 + λ u 2 ( t )] balances output tracking and control effort
λ \lambda λ factor weights control effort importance allows tuning between performance and energy use
Control law incorporates λ \lambda λ adjusts control actions based on effort weighting
Improved robustness handles model uncertainties better (parameter variations, unmodeled dynamics)
Reduced noise sensitivity decreases impact of measurement errors on control decisions
Prevents excessive control limits actuator wear and energy consumption
Impact of control effort weighting
Higher λ \lambda λ reduces control effort conserves energy, increases output variance allows more fluctuations
Lower λ \lambda λ increases control effort more aggressive control, decreases output variance tighter regulation
Larger λ \lambda λ improves stability margins enhances system robustness (gain margin, phase margin)
Smaller λ \lambda λ decreases stability margins reduces tolerance to modeling errors or disturbances
Performance vs robustness trade-off requires balancing control objectives and system constraints
λ \lambda λ selection considers system requirements (energy limitations, output precision, disturbance rejection)
Closed-loop pole analysis reveals λ \lambda λ impact on system dynamics (damping, natural frequency)
Disturbance rejection affected by λ \lambda λ influences system's ability to counteract external perturbations