You are on page 1of 3

1-For designing an artificial neural network , first of all, we should provide the

activation function and weight for data. for this purpose, we should train our
neural network.
For input data we have water flow rate, well head temperature, well head
pressure, time and for output data we have gas flow rate for 6 month ago so we
train the ANN and we find the activation function and weight of data after
training the ANN.
Now if we want to forecast the gas rate for next 12 hours, we give the new input
data to the ANN for this 12 hours and we will get output data from ANN.
2-

Water
flo w
rate

Well
head
tempera
ture

F1(x)

W1
Y=WiFi
F2(x)

Well
head
pressure

W2

i=1,2 ,..n

Wn
Fn(x)

Time

3- Conventional statistical techniques are usually based on certain underlying


assumptions. The important pre-processing step is balancing, which is applied

when input vectors are not equally distributed over the output classes. Without
balancing the data, the neural network only learn to predict low gas flow rate.
So the problem is solved though balancing the data by adding new vectors to the
under-sampled classes. This is done by random selection from these classes and
white noise.
4-We design a fuzzy control model for the above mentioned qualifiers,the system
has fuzzy controller:
Rule 1: If the hourly water flow rate is much higher than the daily average rate,
then the output numbers should be decreased to the daily average.
Rule2:If the hourly temperature is near or above the top quartile the impact of
gas flow rate(gfr), then the output numbers should be decreased to between
gfr^1/2 and gfr^1/3.

11

80%

20%

average
Water flow rate

high

Water flow rate

You might also like