MAKHFI.COM
Fascinating World of Neural Nets  
Neural Network Forums
Home PageHome Page : Neural Networks using MATLAB : Problems Modelling Backpropagation Neural Network with Matlab
  You are currently not logged in. You can view the forums, but cannot post messages. | HOME | Log In | Register | Search | Help
Post a Reply on This Topic Post a Reply on This Topic

Author Topic: Problems Modelling Backpropagation Neural Network with Matlab
Paul Banks Posted: 22-Aug-10 17:29
  Edit Edit
 
Email the Author Mail   View Author's Profile Profile  



Hi All!

I'm new in neural network, and it's my first time in this forum. I'm interested in modelling my first neural network for time series forecasting.


I for example, in my first try I'm using a time serie of temperature, I have 650 matrix for train my neural network.

I want use a moving average for forecasting the next term. I know that I can use other values for forecast that like humity, preassure, etc... But it's only for trying the theory, I don't want that will be a good method for forecasting.

I have the values in array called "temperature" and "targets" the value for forecast. All the values are in Kelvin

I want design a neural network 8-8-1, input layer with 8 neurons, a hide layer with 8 neurons and a output layer with only 1 neuron.

I try this with 8 values in "temperature" matrix with 8x650:(x, x-1, x-2, x-3, x-4... x-8)
And I have "target" matrix 1x650:(x+1)

I type:

net = newff([0 400; 0 400; 0 400; 0 400; 0 400; 0 400; 0 400; 0 400], [8 1], {'purelin','purelin'},'traingd')


net.trainParam.goal = 0.01

net.trainParam.epochs = 1000

net = train( net, temperature, target)

but my problem is: my MSE is around 2000 when will be , and I don't met the goal...

I tryed with RMSE, and changing the goal, logsig, etc... but I don't have any success, Any suggestions?

I think is a problem with modelling, Can someone help me?
What is wrong?



Thank you very much!
 

Post a Reply on This Topic Post a Reply on This Topic
 

Copyright © 2001-2003 Pejman Makhfi. All rights Reserved.