can I used neural network to solve equations and get the unknows

I have a system modeled in simulink as given below where there are certain signals (x1,x2,x3,x4,x5). These signals are multiplied by gains(a,b,c,d,e) and then each of them are sent through saturation block and finally added together to get output. Currently I am hand tuning the gain to get desired output. I am thinking to applying neural network to train the system offline using expected output values and corresponding inputs. Note: The outputs are absolute values. I am able to train the neural network. Please find below the Plots of the training. I wonder if I can find out the gains a,d,c,d,e using the neural network using the weights and bias. I need to find the gains offline and use them later.

The implementation is as below:

Also if this approach is not suitable can you suggest me a better way to solve this.

Answer 1

[Greg Heath]

It looks like you would have to set all of learning rates for net.b and net.IW{ i, j ~= I ) to zero.