For Short term Load Forecasting using ANN,how to choose the number of hidden layer neurons?

In STLF using nntool, I have used 7 parameters namely: maximum temp.,minimum temp.,humidity,day of the week, month number, previous day load and hour number. The output parameter is next day forecast load. The load is forecasted using LM algorithm over the past 10 days data .The dimensions of the input size and output size are 7x240 and 1x240 respectively.200 neurons are used for this problem. Can we take any random number of hidden layer neuron in neural network? Is there any standard and accepted method for selecting the number of hidden layer neurons in ann for load forecasting? Is my number of hidden layer neurons used in this problem wrong as its number is much greater than the number of input neuron(which is 7?

Kindly suggest me any text book for selecting the number of hidden layer and its neurons in neural network?

Answer 1

[Greg Heath]

Greg Heath has developed a two loop approach which has been relatively successful over the years.

 SINCE A SINGLE HIDDEN LAYER IS ALWAYS SUFFICIENT:

 1. The maximum number of hidden nodes that will yield 
    no more unknown weights than training equations is 
    determined.
 2. The outer loop is over number of hidden nodes
 3. The inner loop is over sets of random initial weights.

Although searching over comp.soft-sys.matlab should yield many more examples than ANSWERS, starting with ANSWERS shouldn't hurt.

Hope this helps.

Thank you for formally accepting my answer

Greg