دانشگاه صنعتي اميركبير

Slides:



Advertisements
Similar presentations
دانشگاه صنعتي اميركبير دانشكده مهندسي پزشكي Constraints in MPC کنترل پيش بين-دکتر توحيدخواه.
Advertisements

Properties of State Variables
Electric Drives FEEDBACK LINEARIZED CONTROL Vector control was invented to produce separate flux and torque control as it is implicitely possible.
Visual Recognition Tutorial
280 SYSTEM IDENTIFICATION The System Identification Problem is to estimate a model of a system based on input-output data. Basic Configuration continuous.
INTEGRATED DESIGN OF WASTEWATER TREATMENT PROCESSES USING MODEL PREDICTIVE CONTROL Mario Francisco, Pastora Vega University of Salamanca – Spain European.
Similarity transformations  Suppose that we are given a ss model as in (1).  Now define state vector v(t) that is the same order of x(t), such that the.
Transient and steady state response (cont.)
Digital Control Systems STATE OBSERVERS. State Observers.

Adaptive Signal Processing
Normalised Least Mean-Square Adaptive Filtering
Iterative Methods for Solving Linear Systems Leo Magallon & Morgan Ulloa.
FULL STATE FEEDBAK CONTROL:
REFERENCE INPUTS: The feedback strategies discussed in the previous sections were constructed without consideration of reference inputs. We referred to.
Chapter 20 1 Overall Objectives of Model Predictive Control 1.Prevent violations of input and output constraints. 2.Drive some output variables to their.
1  The Problem: Consider a two class task with ω 1, ω 2   LINEAR CLASSIFIERS.
CONTROL OF MULTIVARIABLE SYSTEM BY MULTIRATE FAST OUTPUT SAMPLING TECHNIQUE B. Bandyopadhyay and Jignesh Solanki Indian Institute of Technology Mumbai,
دانشگاه صنعتي اميركبير
Motivation For analytical design of control systems,
1  Problem: Consider a two class task with ω 1, ω 2   LINEAR CLASSIFIERS.
دانشگاه صنعتي اميركبير دانشكده مهندسي پزشكي Constraints in MPC-2 کنترل پيش بين-دکتر توحيدخواه.
5. Maximum Likelihood –II Prof. Yuille. Stat 231. Fall 2004.
State Observer (Estimator)
Matlab Tutorial for State Space Analysis and System Identification
METHOD OF STEEPEST DESCENT ELE Adaptive Signal Processing1 Week 5.
دانشگاه صنعتي اميركبير دانشكده مهندسي پزشكي استاد درس دكتر فرزاد توحيدخواه بهمن 1389 کنترل پيش بين-دکتر توحيدخواه MPC Stability-2.
دانشگاه صنعتي اميركبير دانشكده مهندسي پزشكي Constraints in MPC-2 کنترل پیش بین- دکتر توحیدخواه.
Lecture 14: Pole placement (Regulator Problem) 1.
(COEN507) LECTURE III SLIDES By M. Abdullahi
FULL STATE FEEDBACK CONTROL:
Chapter 12 Design via State Space <<<4.1>>>
Chapter 8 State Feedback and State Estimators State Estimator In previous section, we have discussed the state feedback, based on the assumption that all.
LINEAR CLASSIFIERS The Problem: Consider a two class task with ω1, ω2.
Presentation at NI Day April 2010 Lillestrøm, Norway
Pole Placement and Decoupling by State Feedback
Chapter 10 Optimal Control Homework 10 Consider again the control system as given before, described by Assuming the linear control law Determine the constants.
Effects of Zeros and Additional Poles
Lec 9. Root Locus Analysis I
Pole Placement and Decoupling by State Feedback
CHE 391 T. F. Edgar Spring 2012.
Overall Objectives of Model Predictive Control
Fredrik Bengtsson, Torsten Wik & Elin Svensson
Random walk initialization for training very deep feedforward networks
Chapter 9 Design via Root Locus <<<4.1>>>
Modern Control Systems (MCS)
Hidden Markov Models Part 2: Algorithms
Solution of Equations by Iteration
Multirate Output Feedback
Chapter 4 Systems of Linear Equations; Matrices
Instructor :Dr. Aamer Iqbal Bhatti
METHOD OF STEEPEST DESCENT
Digital Control Systems (DCS)
Digital and Non-Linear Control
8. Stability, controllability and observability
Homework 9 Refer to the last example.
دانشگاه صنعتي اميركبير
Linear Equations in Linear Algebra
(Systems And Control Engineering) (Deptartment of Electrical Engg.)
Synthesis of SISO Controllers
دانشگاه صنعتي اميركبير
دانشگاه صنعتي اميركبير
Chapter 8 State Feedback and State Estimators State Estimator In previous section, we have discussed the state feedback, based on the assumption that all.
Maths for Signals and Systems Linear Algebra in Engineering Lectures 13 – 14, Tuesday 8th November 2016 DR TANIA STATHAKI READER (ASSOCIATE PROFFESOR)
7-5 Relative Stability.
Chapter 4 Systems of Linear Equations; Matrices
State Feedback.
Linear Equations in Linear Algebra
IntroductionLecture 1: Basic Ideas & Terminology
IntroductionLecture 1: Basic Ideas & Terminology
Presentation transcript:

دانشگاه صنعتي اميركبير دانشكده مهندسي پزشكي استاد درس دكتر فرزاد توحيدخواه بهمن 1387 MPC with Laguerre Functions-2 کنترل پيش بين-دکتر توحيدخواه 2

Example 3 Illustrate that as N increases, for a given Laguerre pole a, the control trajectory converges to the optimal solution generated by using a linear quadratic regulator. Use the same system and design parameters as in Example 2, where the augmented system model is described by the matrices: We use the MATLAB program called ‘dlqr’ to find the state feedback control gain matrix K as: K = [0.8350 0.6113].

The sum of squared errors between the DLQR and DMPC trajectories is:

Example 4. Continue the study from Example 3 Example 4. Continue the study from Example 3.3 with the identical system and design parameters, except that the weight on the control is reduced from R = 1 to R = 0.1. Examine the incremental control trajectory u(ki +m) for this performance specification, and compare the trajectory with the one from DLQR solution.

The optimal control gain vector:

With a smaller weight R = 0 With a smaller weight R = 0.1, the feedback control gain K has increased from that in Example 3. The DLQR control trajectory has faster response speed.

Closed-loop Poles of a DLQR System

In the DLQR design, when we choose Q = CTC, the closed-loop poles are uniquely determined by the weighting parameter R = rω > 0 in the single input and single-output case with their values given by the inside-the-unit-circle zeros of the equation:

In the multi-input and multi-output case:

Verify this property by re-examining solutions from the previous examples 2, 3, 4.

The zeros are 1.5589 ± j0.9563, 0.4661 ± j0.2859 from which the closed-loop poles of the DLQR system are found to be 0.4661 ± j0.2859. We reduced R from 1 to 0.1, The zeros are 1.8228 ± j2.3858, 0.2022 ± j0.2647 from which the closed-loop poles of the DLQR system are found to be 0.2022±j0.2647. Indeed, these poles confirmed with the computation of the actual closed-loop poles through the system matrix A−BKlqr.

When the weight parameter R is reduced, the closed-loop poles move towards the origin of the complex plane, hence the closed-loop response speed is faster, as we have observed from the responses in the examples.

Use of Laguerre Parameters as Tuning Parameters

The choice of a scaling factor a and a smaller N for the Laguerre functions affects the closed-loop performance of the DMPC system, although for a large N the performance converges to the underlying optimal DLQR system. However, when N is small, the potential here is to use a as the performance fine-tuning ‘knob’.

Example 5.

Examine the effect of parameter a and N on the closed-loop performance Examine the effect of parameter a and N on the closed-loop performance. Assume at time ki = 10, x(ki) = [0.1 0.2 0.3]T .

Case A. We fix the parameter N with a large value (N = 8), and vary the pole location α by choosing α = 0, α = 0.4, and α = 0.8. Table 3 shows the comparison results for the cases where N is large while a changes. It is seen from Table 3 that the closed-loop predictive control system is very similar to the underlying DLQR system and the effect of α is quite small when N is large.

Table 3- Closed-loop eigenvalues and feedback gain vector when N = 8 and α varies from 0 to 0.8

Case B. For the second case, we fix N with a smaller value (N = 2), and vary the parameter α. Table 4 summarizes the comparison results. It is seen that with a smaller N, the closed-loop poles and gain of the predictive control system are functions of the Laguerre pole α.

Table 4. Closed-loop eigenvalues and feedback gain vector when N = 2 and α varies from 0 to 0.9

We observe that the parameters α and N affect the closed-loop control performance, particularly when N is small. This is particularly useful in the situation when the optimal DLQR system does not provide us with satisfactory performance, and these additional tuning parameters help us with fine tuning of the closed-loop performance.

Constrained Control Using Laguerre Functions

With parameterization of the control signal trajectory using Laguerre functions, we have the flexibility to choose the locations of the future constraints. This could potentially reduce the number of constraints within the prediction horizon, and hence the on-line computational load for large-scale systems. In addition, because of the existing exponential decay factor in the Laguerre functions, the difference of the control signal is ensured to converge to zero after the transient period. Thus, it is sufficient in the majority of cases that the constraints are imposed in the transient period of the response. This in turn will reduce the number of constraints.

1 Constraints on the Difference of the Control Variable

Example 6. Suppose that a continuous-time system is described by the transfer function

Optimal solution without constraints. Key: line (1) DMPC; line (2) DLQR

It is seen from this figure that with this choice of α and N, without constraints, the responses are almost identical. Incidentally, if the parameter α = 0, which corresponds to the traditional approach, then N is required to be approximately 40 in order to achieve similar performance. To ensure that the constraints are satisfied for the whole response, the first 15 samples of u are put into the constrained solution.

Optimal solution with constraints

The constrained results within one optimization window. Although there are 30 constraints imposed, there are only three active constraints, which can be found through visual inspection of Figure 3.10a. The first two samples and the 17th sample are the activated constraints in the solution. By comparing Figure 3.10 with Figure 3.9 it can be seen that activated constraints resulted in little performance deterioration.

2 Constraints on the Amplitudes of the Control Signal

Example 7

Optimal solution with constraints on u Optimal solution with constraints on u. Key: line (1) without constraints; line (2) with constraints

It is seen from Figure b that some of the constraints on the control signal were not met. By studying the multipliers, it is noticed that there were eight active constraints and the multipliers could not converge because the active constraints were in conflict. The values of the positive multipliers are listed after 88 iterations in the calculation as:

Numerically, the Mact matrix has row dependence, and as a result, the constraints are compromised (see Figure 12). One comment relates to the number of constraints used in this example. Although there are 192 constraints used in the optimization, there are only 8 active constraints. The rest of the constraints are inactive and have no effect on the optimal solution. Because of receding horizon control, where the first sample of the optimal control is implemented, so from this point of view, only the constraint on the first sample had an effect in this example.

Stability Analysis