Clear all;
x =[]；
t =[]；
x( 1)= 0.22；
k = 4；
n = 900
N = 400
% generated a logical sequence, and the first 100 was deleted.
For i= 1:n
x(I+ 1)= k * x(I)*( 1-x(I))；
end
x( 1:800)= x( 10 1:900)；
% training RBF neural network with the first 400 data of x
For j= 1:N
P( 1:8，j)= X(j:j+7)'；
end
t = X(9:408)； % target data
net 1=newrb(P，T，0.00 1， 1)； % training RBF neural network
n 1 = 300；
% Select 400-700 data of X for RBF neural network test.
For j= 1:N 1
P 1( 1:8，j)= X(j+400:j+7+400)'；
end
t 1 = X(409:708)； % target data
% simulation verification
a=sim(net 1，p 1)； The prediction result of% test data %sim means simulating the network.
e = t 1-a； % one-step prediction error
Mse = sqrt(e*e')/size(e, 2)% mean square error.
% Draw a picture to depict the simulation results.
Figure (2)
plot(X( 1:200))；
Axis ([1200-0.11.1]);
Title ('logistic chaotic sequence');
xlabel(' t ')；
Ylabel ('magnitude');
Figure (3)
plot( 1:300，T 1，' b '， 1:300，a，' r * ')；
H = legend ('chaotic sequence',' RBF neural network one-step prediction value');
Axis ([1300-0.51.5]);
continue
Title ('chaotic sequence and one-step predictive value');
xlabel(' t ')；
Ylabel ('magnitude');
Fig. 4;
plot(e，' b-')；
Axis ([1300-0.02 0.02]);
Title ('prediction error e');
xlabel(' t ')；
Ylabel ('magnitude');
As for K-means clustering:
One of the most basic rbf algorithms, the learning algorithm adopts pseudo inverse function.
%
%
%
%
%
%
%
%
%
%
%
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Clear all;
All closed;
clc
% defines the variables used.
r = 1； % domain radius, set manually
nodenum = 0； The number of% hidden layer nodes is determined by the actual data and r***
input data =[]； % input matrix
input path =“”； % Path to store original input data
node out =[]； % hidden layer input array
netout =[]； % network output array
Weight = []; % output weight matrix, which is the only weight.
input num = 0； Enter the dimension percentage.
output num = 0； % output dimension
center =[]； % cluster center
num train = 0； Number of% learning samples
row = 0； Number of% learning samples
sim row = 0； Percentage of all samples
numtest = 0； Number of% generalization samples
Strength =1; The range used in% normalization is usually 1.
yout =[]； Expected value of% output
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Initialization of% variable
r = 1；
input num = 9；
input path = ' e:\ yinjia \ data \ years un . dat '；
output num = 1；
simrow = 290
Source=load (input path);
% Save a copy of the source to prepare for denormalization.
copysource = source
% normalization processing
Source=normalize (source, intensity);
yout = source(input num+ 1:sim row+input num)；
inputdata=phasespace(source，inputnum，sim row)；
row = 250
numtrain = row
numtest = sim row-row；
% Set the first person input variable as the initial center.
After initialization, the location of% center will not change.
center=inputdata( 1，:)'；
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% started a simple cluster.
The basic idea of% is: given a fixed neighborhood radius, the points within the neighborhood radius of the central point are regarded as
% is a point in the neighborhood; For points outside the radius, the first point that is not within the radius is taken as the new center point.
% design a function iscenter calculates whether a point is the center point.
For step =2: line
if iscenter(inputdata(step，)'，center，r)
center=[center inputdata(step，)']；
nodenum = nodenum+ 1；
End% this end for is center (input data (step,))
End of step =2% this end:row
Cluster completion percentage.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Prepare the forward calculation of neural network.
[center row nodenum]= size(center)；
Initialization of% network
nodeout=zeros(row，nodenum)；
netout=zeros(row，output num)；
weight=zeros(nodenum，output num)；
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Calculation of% network
For step size = 1: line
For step 1= 1:nodenum
nodeout(step，step 1)= GaussRadialBasisFunction(input data(step，:)'， ...
center(:，step 1)，r)；
End% This step ends 1= 1:nodenum
end %this end for step= 1:row
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% uses pseudo-inverses to calculate weights
weight = pinv(node out)* yout( 1:250)；
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Percentage used for simulation
For step= 1:simrow
For step 1= 1:nodenum
nodeout(step，step 1)= GaussRadialBasisFunction(input data(step，:)'， ...
center(:，step 1)，r)；
End% This step ends 1= 1:nodenum
end % this end for step = 1:sim row
mydata = nodeout * weight
% denormalization
mydata=unnomal(mydata，copysource，strength)；
rl data = copy source(input num+ 1:input num+sim row)；
plot(rl data)； Hold on; plot(mydata，' r ')；
% calculate an evaluation function
rmsetest =(norm(rl data(num train+ 1:simrow)-my data(num train+ 1:simrow)) ...
^2/(numtest- 1))^0.5
There are several small programs, you should be able to write them yourself.