4.2 Radial Basis Function Network Experts

4.2.1 Offine Training: Approach I

 K-means clustering. Figure 4-4- shows the simulation for a RBFN using k-means

clustering for the RBF layer, the RLS algorithm for the output layer training, and online

change detection is performed.







 I I I I I I -
 500 1000 1500 2000 2500 3000
 o 02

 0 01
 0 I I I I I
 500 1000 1500 2000 2500 3000



 1u1
 500 1000 1500 2000 2500 3000
 50 1
 50



 500 1000 1500 2000 2500 3000


Figure 4-4 Results: (a) Flow: desired in blue, predicted in red; (b) MSE criterion, full line
 is for expert 1, dotted line is for expert 2; (c) Winner; (d) Pressure.

 An important comment to be made is that the value of the error criterion is ten

times lower than for the linear model, and it can be seen on the flow graph that the

predicted series is a lot smoother which means that the experts are modeling the data

more accurately, but the segmentation is not performed well because of the "build-up" of

the criterion on expert 1: it only decreases below the criterion on expert when there are

two mandatory breaths in a row, so single mandatory breaths are not detected. There are

two ways this could be improved: decrease the memory depth of the criterion, but then

we would run the risk of having unwarranted switches (as can be seen on Figure 4-5