System and method for exploiting a good starting guess for binding constraints in quadratic programming with an infeasible and inconsistent starting guess for the solution

• US 7,376,471 B2
• Filed: 02/21/2006
• Issued: 05/20/2008
• Est. Priority Date: 02/21/2006
• Status: Active Grant

• Current Assignee: Rtx Corporation
• Original Assignee: United Technologies Corporation (Rtx Corporation)
• Inventors: Rey, Gonzalo, Das, Indraneel
• Primary Examiner(s): VON BUHR, MARIA N
• Application Number:
  US11/358,735
• Publication Number:
  US 20070198446A1
• Time in Patent Office: 819 Days
• Field of Search: 700 28- 34, 700/37, 700/44, 700/52, 700/53, 700/73, 700/74, 703/2, 703/22, 706/21, 708/446
• US Class Current: 700/28
• CPC Class Codes: G05B 13/042 in which a parameter or coefficient is automatically adjusted to optimise the performance

Abstract

The present invention provides an algorithm that does not relax the problem at the very onset, even if x_f is infeasible. Instead, it solves the EQP with the initial guess for the active set without relaxing the problem. If this solution to the first EQP is not optimal, but nevertheless feasible, we can use this as our guess for the feasible point. This has the advantage of being a feasible point that is consistent with the initial active set, whereas the initial guess used in the previous method is not necessarily so.

Claims

1. A method for controlling a multivariable system including the steps of:

a) receiving a plurality of sensor signals indicating current conditions of the system;

b) receiving a plurality of commands;

c) determining the desired dynamic response of the system based upon the commands and the sensor signals;

d) in each of a plurality of time steps, formulating a problem of controlling the system to achieve the desired dynamic response as a solution to a quadratic programming problem;

e) solving the quadratic programming problem in each time step using an iterative algorithm which searches for an optimal active set, wherein the active set comprises a set of constraints that are binding at an optimized solution; and

f) in each subsequent time step of the plurality of time steps;

g) solving the quadratic programming problem based on a final active set of a prior time step of the plurality of time steps to obtain x_f;

h) initializing a search for the optimal active set based on the final active set of the prior time step of the plurality of time steps and based upon the assumption that x_f is feasible.

2. The method of claim 1 wherein said step h) is performed without relaxing any of the active constraints.

3. The method of claim 1 wherein said step h) is performed without relaxing a plurality of the active constraints.

4. The method of claim 3 further including the steps of, in each subsequent time step in said step f):

i) determining that x_f is feasible, but not optimal, after said step g) and using x_f in said step h) based upon the determination that x_f is feasible.

5. The method of claim 1 wherein the constraints include a plurality of critical constraints and a plurality of non-critical constraints, the method further including the steps of, in one of the plurality of subsequent time steps in said step f):

i) determining that x_f is not feasible, after said step g);

k) relaxing at least one of the plurality of non-critical constraints based upon the determination that x_f is not feasible.

6. The method of claim 5 further including the steps of:

l) determining that x_f does not violate the critical constraints; and

m) based upon said step l), using a next set of working constraints as the initial working set, minus the constraints for which an associated Lagrange multiplier is negative.

7. The method of claim 5 further including the steps of:

l) determining that x_f violates at least one of the critical constraints; and

m) based upon said step l), using a projection of x_f onto the plurality of critical constraints, x_p, as a new guess for the solution.

8. The method of claim 7 wherein v is a vector denoting which constraints should be relaxed, and wherein the plurality of non-critical constraints are relaxed in said step k) by introducing an additional variable t, the method further including the steps of:

based upon the determination in said step l) that x_p violates at least one of the critical constraints, altering the vector v and the value of t to maximize the number of constraints for which x_p is binding in the initial active set.

9. A control system comprising:

a desired trajectory generator for creating a desired trajectory;

a linearization module deriving a linearized model about the desired trajectory;

a quadratic programming module in each of a plurality of time steps, formulating a problem of controlling the system to achieve the desired dynamic response as a solution to a quadratic programming problem;

a quadratic programming solver for solving an optimization problem established by the quadratic programming module to generate a profile of optimal controls, the quadratic programming solver solving the quadratic programming problem in each time step using an iterative algorithm which searches for an optimal active set and in each subsequent time step of the plurality of time steps, the quadratic programming solver in each subsequent time step of the plurality of time steps solving the quadratic programming problem based on a final active set of a prior time step of the plurality of time steps to obtain x_f, the quadratic programming solver initializing a search for the optimal active set based on the final active set of the prior time step of the plurality of time steps and based upon the assumption that x_f is feasible.

10. The system of claim 9 wherein the quadratic programming solver initializes the search for the optimal active set without relaxing any of the active constraints.

11. The system of claim 9 wherein the quadratic programming solver, in each subsequent time step, determines that x_f is feasible, but not optimal.

12. The system of claim 9 wherein the constraints include a plurality of critical constraints and a plurality of non-critical constraints, the quadratic programming solver programmed to, in one of the plurality of subsequent time steps:

determine whether x_f is feasible, relax at least one of the plurality of non-critical constraints based upon the determination that x_f is not feasible.

13. The system of claim 12 wherein the quadratic programming solver is programmed to determine whether x_f violates the critical constraints and if not, to use a next set of working constraints as the initial working set, minus the constraints for which an associated Lagrange multiplier is negative.

14. The system of claim 12 wherein the quadratic programming solver is programmed to determine whether x_f violates the critical constraints and if so, use a projection of x_f onto the plurality of critical constraints, x_p, as a new guess for the solution.

15. The system of claim 14 wherein v is a vector denoting which constraints should be relaxed, and wherein the plurality of non-critical constraints are relaxed by introducing an additional variable t, the quadratic programming solver programmed to alter the vector v and the value of t to maximize the number of constraints for which x_p is binding in the initial active set based upon the determination that x_p violates at least one of the critical constraints.