Anda di halaman 1dari 7

Exercise 5

Exercise 5 a) The state space representation of the system in question is presented in the following equation:

Where the input is a step function characterized as follows

Given an initial state z(0)=0 the system presents the step response described for the following outputs y_1 and y_2 measured for k=0,1,2,,30.

y_1 = { 0.0 0.2 0.5 1.2 2.2 3.3 4.3 5.4 6.2 6.6 6.9 ... 7.1 7.4 7.5 7.5 7.4 7.5 7.6 7.5 7.5 7.5 ... y_2 = {0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.5 0.2 -0.2 -0.6 ... -1.2 -1.8 -2.7 -3.5 -4.2 -4.5 -4.7 -4.8 -4.7 -4.5 ... -4.4 -4.3 -4.2 -4.0 -4.1 -4.1 -4.2 -4.2 -4.2 -4.2 }
It can be observed that both outputs have a value of 0.0 at the initial time k=0. Substituting the mentioned values at time k=0 in the state space model yields the following equation [ 1(0) ] = [ 1 ] (0) + [ 1 ] (0) 2 2 2(0) [ 0.0 ] = [ 1 ] (0) + [ 1 ] (1) 2 2 0.0 0.0 [ 1] = [ ] 2 0.0 It can be therefore concluded that the systems is strictly proper. b) A numerical approach for the realization of the single-input-multiple-output (SIMO) model was used. The approximate realization algorithm detailed in Section 4.6 of the lecture notes was followed. The first proposal used for the approximation of the SIMO model was the construction of two (single-input-single-output) SISO models. As described in section 4.6.4 of the lecture notes the step response matrix Sk satisfies Sk=0, k=0,-1,-2,, and

Where Di are the Markov parameters which are available from the experimental data provided. The finite block Hankel matrices were therefore defined as follows:

Exercise 5

Following is the code used in Matlab to construct the Hankel matrices. For each of the two available outputs a set of HE, HA, HC, HB matrices was generated.

Following the mentioned algorithm, a singular value decomposition for each of the two Hankel matrices H E1 and HE2 was computed using the svd command in Matlab having as a result the following matrices. 1 = [11 12 ] [ 11 0 22 ] [ 21 0
0 11 ] [ ] 12 12 0 21 ] [ ] 22 22

2 = [21

Exercise 5

The singular Hankel values can be obtained after the decomposition from the diagonal elements of S1 and S2 which are described as follow:
1 = (51.0650 5.4014 0.7731 0.5085 0.4520 0.2258 0.2021 0.1911 0.1685 0.1370 0.1332 0.1249 0.1189 0.0894 0.0142) 2 = (42.2073 12.9642 4.4203 1.1791 0.6416 0.4880 0.3513 0.2912 0.2631 0.2079 0.1887 0.1609 0.1457 0.0202 0.0137)

The matrices S12 and S22 contain the smallest singular values, which dont contribute to the order of the approximate realization a later explanation of the chosen order for the SISO approximations of the system will be given. Once the Hankel matrices for each of the systems has been computed an approximate state space model can be computed from the following equations as detailed in section 4.6 of the lecture notes.
1 = [112 11 1 11 112 ] 2 2 2 = [21 21 2 21 21 ] 1 = [112 11 1 ] 2 2 = [21 21 2 ] 1 1 1 1 1 1

1 = 1 11 112

1 1

2 2 = 2 21 21

The determination of the most reasonable order for the approximate system was chosen after comparing three different realizations with the original step response output data. The following figure shows the step response of three different approximations of order 2, 3 and 4 compared to the original given output y_1. It can be observed that both the third and fourth order models are close approximations of the original system. Since there is no significant improvement from a third to fourth order realization, the third order one is chosen to be a reasonable order for the given model.

Exercise 5

The figure below shows the step response of three different approximations of order 3, 4 and 5 compared to the original given output y_2. For the approximations of this second system it is evident that the third order approximation is significantly not close to the original system. Once again the two following order approximations (4 and 5) are fairly similar to each other reason why the fourth order realization is chosen as the most reasonable option.

The second approach used for the approximation of the given model was the construction of a SIMO model. The algorithm detailed in section 4.6 of the lecture notes was followed. As opposed to the previous proposal, a single set of Hankel matrices was constructed using as the Markov parameters the data from the two outputs. The following Matlab code shows the algorithm used for the construction of the corresponding Hankel matrices.

Exercise 5

The computation of the Hankel matrices from a step response data follow from the following equations by column operations under system equivalence, which is described by post multiplication of an upper diagonal matrix filled with ones.

The construction of the SIMO discrete time state space model is performed with the same equations used for the construction of the SISO approximate models. The singular Hankel values for the SIMO case are defined by

= (66.0009 14.3990 5.3657 1.7842 0.7998 0.6784 0.6409 0.4690 0.3621 0.3265 0.3249 0.2443 0.1931 0.1733 0.1606)

Exercise 5

The following figure shows the impulse response of three different approximate realizations and the experimental step response data. A fourth order realization was selected as the most convenient one for this SIMO model for the same reasons explained in the SISO case model.

c) The proposed SISO realizations were of order 3 and 4 for the first and second output respectively. The SIMO approximate model was of fourth order. In order to identify the common dynamics of the two proposed realizations the eigen values of the three systems were computed and are determined as follow: 1 = (0.9189 + 0.0000 0.8646 + 0.3043 0.8646 0.3043) 2 = (0.8856 + 0.1454 0.8856 0.1454 0.8509 + 0.3917 0.8509 0.3917) = (0.8603 + 0.1505, 0.8603 0.1505, 0.8275 + 0.3826, 0.8275 0.3826) It can be observed that the systems have in common the eigen values corresponding to the real part equal to 0.86. Therefore, the probable order of the common dynamics of the systems is 2.

Exercise 5

d) When the order of the approximated system is too high two main consequences can be observed. The first one, which in fact must be taken into account when following the numerical approximate realization algorithm, is that unwanted noise phenomena might be appearing in the approximated model. Moreover, the design of controllers for models with higher order becomes more complex in analytical and computational terms. The second effect that could be observed during the realization of this exercise is that a higher order system often contains unstable eigen values, which for discrete time state space model can be identified when they are located outside of the unitary disk as can be observed in the following figure.

Anda mungkin juga menyukai