如何用matlab仿真elman神经网络

1:20

p1=sin(t)

p2=sin(t)*2

plot(t,p1,'r')

hold

on

plot(t,p2,'b--')

hold

on

t1=ones(1,20)t2=ones(1,20)*2%产生两组向陪尘量，分别为这两波形幅值，作为输出向量

p=[p1

p2

p1

p2]

t=[t1

t2

t1

t2]

Pseq=con2seq(p)%将矩阵形式的训练样本转换芦闭禅为序列的形式

Tseq=con2seq(t)

R=1%输入元素的数目为1

S2=1%输出曾的神态衡经元个数为1

S1=10%中间层有10个神经元

net=newelm([-2,2],[S1,S2],{'tansig','purelin'})

net.trainParam.epochs=100%设定次数

net=train(net,Pseq,Tseq)

y=sim(net,Pseq)

%预测

P=randn(12,2)T=randn(12,2)

threshold=[0

10

10

10

10

10

10

10

10

10

10

10

1]

a=[11

17

23]

for

i=1:3

net=newelm(thresho...

如你所说的，newgrnn是广义rbf,而广义rbf是不迹宽亏需要train的，所以怎么会有误差曲线巧运了？

P = [1 2 3]

T = [2.0 4.1 5.9]

net = newgrnn(P,T)

这个表示这个网络已经固定了。

network

Create custom neural network

newc

Create competitive layer

newcf

Create cascade-forward backpropagation network

newdtdnn

Create distributed time delay neural network

newelm

Create Elman backpropagation network

newff

Create feedforward backpropagation network

newfftd

Create feedforward input-delay backpropagation network

newfit

Create a fitting network

newgrnn

Design generalized regression neural network

newhop

Create Hopfield recurrent network

newlin

Create linear layer

newlind

Design linear layer

newlrn

Create layered-recurrent network

newlvq

Create learning vector quantization network

newnarx

Create feedforward backpropagation network with feedback from output to input

newnarxsp

Create NARX network in series-parallel arrangement

newp

Create perceptron

newpnn

Design probabilistic neural network

newpr

Create a pattern recognition network

newrb

Design radial basis network

newrbe

Design exact radial basis network

newsom

Create self-organizing map

sp2narx

Convert series-parallel NARX network to parallel (feedback) form

如果姿神是design的，一般是不需要train的

如果是crate,一般有误差曲线

---