Ahere are many steps involved in preprocessing of the input data depending upon the requirement.  I assume that the data is already partitioned in training and test set. Next, as a basic step the data must be normalised either in the range 0-1 or in the range -1 to 1. You should normalise the data using the min and max of the training set and use only those values for normalisation of test set as well. Next depending upon the parameter to be forecast, relevant frequency components in the signal should be retained and irrelevant ones removed by using the FFT algorithm otherwise the system tends to learn irrelvant frequencies as well which will affect the quality of forecast at a later stage. Further, if required, relevant features should be extracted by doing the PCA and the irrelevant dimensions discarded. Hope it helps...

Dear friends

Patrick and Krishna   thanks for you answers .

Of course you have to monitor the power system and to take into account all the parameters that can influence your output (i.e., load consumption), even past values of the output in order to grab the dynamics of the system.

Dealing with neural network data can be somewhat tricky, The process of encoding categorical data and normalizing numeric data is sometimes called data standardization. take a look in this link ... some good ideas to follow ahead:

http://visualstudiomagazine.com/articles/2014/01/01/how-to-standardize-data-for-neural-networks.aspx