System Identification of an unmanned (TREX 550) flybarless helicopter using the state-space structure of a Yamaha RMAX helicopter as done in Bernard Mettler's paper. The code optimises these parameters to get best fit using two different algorithms
- Genetic Algorithms- Testing done
- Invasive Weed Optimization- Testing done
- Simulated Annealing
We aim to maximise the Pearson correlation coefficient the actual data and the flight data.
The major changes have been done in the sphere(or cost) function of each of the methods.
-
Open
iwo.m
andsphere.m
(Changes have been made only in that) -
Open
best.mat
This workspace contains inr-the input matrix of 4 inputs recorded along the time steps outr- 10 outputs excluding rfb,c and d because we do not have a way to measure them t- time steps ( required for using lsim function) population - pre defined initial population because random initializations result in NaN. Insphere.m
which is a cost function we input the array of values for the 40 parameters and create a state space model with that. Now we send the rc input values(inr) to the model and simulate the output. we compare the simulated output with the actual output by using correlation coefficient and least squares (it needs to be refined). this function returns a value that must be minimised. -
Run
iwo.m
-
Once iwo is done with 1000 iterations it stops
-
Go to BestSol structure in the workspace and the best array of the 40 parameters can be found under the variable position
-
Copy this array vertically in the variable in the workspace called popul
-
Now run PopulCheck.m to visualize the model
The SA also follows a similar structure
Creator - Navaneeth Krishnan
Maintainer - Saumya Kumaar