hi everyone,

I have trained MLP neural network and identified my system with this network. for training, I have normalized all input data so the range of data that enters the network is changing from zero to one. for normalization, I have used this formula: {(Input(i)-min(input space))/(max(input space)-min(input space))}. now i want to use this network to control my system. the data that network generates is somehow normal and my system cannot accept these data. What should I do with this data? I have tried the reverse transformation but since system is working online the max and min of it is not known. anybody have any idea?

hi everyone,

I have trained MLP neural network and identified my system with this network. for training, I have normalized all input data so the range of data that enters the network is changing from zero to one. for normalization, I have used this formula: {(Input(i)-min(input space))/(max(input space)-min(input space))}. now i want to use this network to control my system. the data that network generates is somehow normal and my system cannot accept these data. What should I do with this data? I have tried the reverse transformation but since system is working online the max and min of it is not known. anybody have any idea?

 

an MLP (Multilayer Perceptrons) is on the form [math]I-[A-, ..]O[/math], where number of nodes in the layers [math]I \;\; [ \; \geq A, .. \; ] \;\; \geq O[/math]

 

where [math]I[/math] is the Input Layer, [math]O[/math] is the Output Layer, and [math]A, ..[/math] are Hidden Layers

 

Size of the Input = [math]I[/math]

 

Size of the Output = [math]O[/math]

 

.. and in every Node, there is a formula that takes inputs and give an output

 

Here's an example:

image source: neural-forecasting.com

 

You said: "the data that network generates is somehow normal and my system cannot accept these data"

 

Why not re-design the MLP network, you can add a another hidden layer, to reduce the output size

Edited January 10, 201213 yr by khaled