A solution based on two FNNs (version 2)

This solution bases on a feedforward neural network. The network for prediction of the usage has got five inputs: average temperature for the month (estimated from the data from last year), usage in last december, dwelling type, number of rooms, and the month. Each value is normalized to the range [-1, 1]. If the number of rooms or dwelling type is not available in the input data, this value is estimated by a second feedforward neural network. It's very simmilar to previous solunion, the only difference is the preparation of the training data.