Continue to Site

Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

Register Log in

Neural network - prediction. How?

Status
Not open for further replies.
C

ckck20

Member level 1
Joined
May 19, 2002
Messages
33
Helped
0
Reputation
0
Reaction score
0
Trophy points
1,286
Activity points
219
nural network

Hi,
i have a set of measurements(about 1000) about the intensity of the wind and these are taken at intervals of 10min on the same place. I need to predict the intensity of the wind for the next 2 hours(the new 12 measurements).
I don't know how to model this as a neural network. I mean, that i don't know what to use for the training(inputs and targets).
I'd like to hear your ideas and please anything is appreciated!

Thanks in advance
 

neural network prediction

The answere is so tricky
The best answere is to use 2 input nural network the first input is your data taken each 10 minutes. for the second input use the same data with 10 minutes delay. It means you have a Z^-1 feedback from output to input making the nural network able of predition. using a single or double feedback( Z^-1 and Z^-2) is usually enough.
 

    C

    ckck20

    Points: 2
    Helpful Answer Positive Rating
predictions with matlab

Thanks for your answer.
I'm using matlab and a backpropagation neural network. The training is done by giving it as input the last 10 measurements and targeting it to the next(assuming all i want is the n+1 prediction). But it doesn't learn!
Now i'll search about what you said and see what i can do.
 

prediction neural

As I think, the way you do it should work. Taking 10 last measurements instead of just 2, the result should be better. Anyway, you can modify your neural network and test with every kind of input and output.

Another thing you should put attention is the architecture of the neural network. How many hidden layers you're using? How many neurons on each hidden layer? Also make sure your neural network code is correct. You can try this by testing with XOR problem, which is a typical example of backpropagation neural network.
 

predict matlab

Hi, and thanks for your reply.
My code is correct as i see it. I'm using 1 hidden layer of about 10-20 neurons and i've tried using different number of past values. I noticed that i get better results when using 1 hidden layer instead of 2. But the overall prediction is not good in both cases. I've been trying for a day now to find the right combination but till now i haven't found it. Maybe i should use a different approach, but the one i used seems more logical to me. Any suggestions?
 

nural network

ckck20, how's your findings now? Can you get a good result?

As you have around 1000 data, I think you may need to input more instead of just 10...If you input 50 inputs, and predict the next data, it may work...you have to test out all the methods of input and output, beside the architecture of the neural network.
 

Status
Not open for further replies.

Similar threads

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Back
Top