28.02.2014 Views

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

10 CHAPTER 1 INTRODUCTION<br />

improved by proper network structure selection. This can be obtained with the<br />

aid <strong>of</strong> the k-fold cross validation method. Furthermore, the implementation <strong>of</strong> the<br />

proposed network architectures, namely the HMLP and modified Elman networks,<br />

is found to improve the learning rate <strong>of</strong> NN prediction; and this facilitates the<br />

real-time implementation <strong>of</strong> the NN based system identification.<br />

3. <strong>The</strong> development <strong>of</strong> real-time <strong>Neural</strong> <strong>Network</strong> based Approximate Predictive<br />

Control (APC) scheme with constraints.<br />

<strong>The</strong> <strong>Neural</strong> <strong>Network</strong> based Approximate Predictive Controller (NNAPC) is designed<br />

and developed using the NN model identified either from the <strong>of</strong>f-line or<br />

on-line system identification algorithms. <strong>The</strong> complexity <strong>of</strong> the proposed controller<br />

algorithm can be handled well enough by the embedded system even with<br />

the implementation <strong>of</strong> the on-line system identification algorithm. <strong>The</strong> control<br />

method is proposed to overcome the demanding computation efforts <strong>of</strong> non-linear<br />

Model Predictive Control (NMPC)’s optimisation procedure. <strong>The</strong> main difference<br />

between NNAPC with linear MPC and other NMPC methods lies in the prediction<br />

operation <strong>of</strong> NNAPC. Instead <strong>of</strong> relying on the prediction from single linear or<br />

non-linear model, the NNAPC controller extracts linear models from the identified<br />

NN model at each sampling time and uses it to predict the future plant response<br />

within the specified time horizon. <strong>The</strong> proposed NNAPC algorithm is able to<br />

achieve satisfactory tracking control performance with different degrees <strong>of</strong> control<br />

autonomy. <strong>The</strong> NNAPC controller is fine tuned for each <strong>of</strong> the control channels<br />

(roll, pitch, yaw and altitude) and the selected tuning parameters are then used for<br />

the testing <strong>of</strong> full autonomous hovering control. <strong>The</strong> NNAPC controller approach<br />

is also tested and proved to be robust enough to handle variation in helicopter<br />

dynamics and demonstrate good disturbance rejection performance to input disturbances.<br />

<strong>The</strong> flight test results indicate the efficiency <strong>of</strong> the NNAPC controller<br />

to achieve autonomous hovering.<br />

4. <strong>The</strong> development <strong>of</strong> a 6 DOF test rig for safe control flight test <strong>of</strong> helicopter UAS.<br />

Attempting to control a remote control helicopter is difficult and careful experi-

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

Saved successfully!

Ooh no, something went wrong!