13.07.2015 Views

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Hard Constraints for Prioritized Objective <strong>Nonlinear</strong> MPC 225Table 1. <strong>Model</strong> Parameters for the Simulated Network <strong>of</strong> Pressure TanksV 1 =8V 3 =8c CV 1 =0.25 c 12 =0.02 c 22 =0.06 c 44 =0.06 γ 1 =0.5V 2 =5V 4 =5c CV 2 =0.25 c 14 =0.05 c 34 =0.02 c 32 =0.05 γ 2 =0.36 Closed-Loop ResultsThe performance <strong>of</strong> the proposed control algorithm is tested on the simulatedpressure tank network. Its ability to appropriately h<strong>and</strong>le control objective prioritizationthrough a number <strong>of</strong> reference transitions <strong>and</strong> in the presence <strong>of</strong>disturbance loads is demonstrated. A number <strong>of</strong> control objectives are defined<strong>and</strong> assigned a relative priority. These are summarized in Table 2. Assume thatfor safety concerns, it is important that the pressure in the upstream tanks arekept below a pressure <strong>of</strong> 60 psig. These constraints are given highest priority.A secondary goal is the regulation <strong>of</strong> the pressure in second tank (P 2 ). It isdesirable for this tank pressure to closely track setpoint, <strong>and</strong> thus a setpointconstraint as well as tight upper <strong>and</strong> lower bounds (±2 psigfromsetpoint)areimposed. Note that the lower <strong>and</strong> upper bounds are assigned to be priority 3<strong>and</strong> 4 respectively, while the setpoint constraint is not considered for objectiveprioritization <strong>and</strong> use <strong>of</strong> hard constraints. Subsequent control objectives includea lower bound on the pressure in tank 1 <strong>and</strong> bounds on the pressure in tank 4.The pressure in tank 3 is left unconstrained. All constraints include a 15 minutedelay for enforcement.Table 2. Summary <strong>of</strong> Prioritized Control Objectives (* Note that a hard constraintcorresponding to the setpoint control objective is not used.)Relative Variable Constraint Constraint Relative Variable Constraint ConstraintPriority Constrained Type Value Priority Constrained Type Value1 P 1 UB 60 6 P 4 UB 302 P 3 UB 60 7 P 4 LB 203 P 2 LB 25/20/25 8 P 4 LB 234 P 2 UB 29/24/29 9 ∗ P 2 SP 27/22/275 P 1 LB 55The controller is tuned to with m =2<strong>and</strong>p =20. Each control objective is assigneda weight <strong>of</strong> Γ e = 100 <strong>and</strong> the input movements are not penalized, Γ u =0.These weights are used to determine the trade<strong>of</strong>fs between s<strong>of</strong>t constraint violations<strong>of</strong> the various control objectives for which the hard constrained problemcannot be solved. For this example, at each time step, the appropriate NLP issolved using a stochastic approach followed by a local gradient based search.Specifically, 1000 points in the solution space are considered, the best <strong>of</strong> whichis used as the starting point for the gradient-based solution. This stochastic <strong>and</strong>gradient-based solution process is repeated 3 times <strong>and</strong> the best solution is takenas the optimal control sequence to be implemented.At t = 150 minutes, a setpoint change for the pressure in tank 2 steps fromits initial value <strong>of</strong> 27 psig to 22 psig. The controller recognizes the change <strong>and</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!