The paper is devoted to building up time series models to forecast the power consumption of a mine. The results discussed are obtained using various linear filter models and artificial neural network. The wavelet transform of the raw time series is shown to be an efficient technique to increase the forecasting accuracy.