Summary:

Short-term water usage forecasts are fundamental for the waterworks and sewerage systems as well as for the sewage treatment plants’ optimization. In this research the capability of forecasting the time series of daily household water usage in the farms with implementation of Data Mining methods was evaluated. To prepare the 10-days water usage forecast, exponential smoothing and ARIMA method was used. The source material were the daily amounts of water used for household purposes in the selected farmhouse during 22 months. Exponential smoothing turned out to be the most useful in water usage forecasting, because it includes not only the values, but also the diversification of the future forecasts’ importance. Significant inequality of the daily water usage causes the increase of the forecasts’ errors. The forecasting methods which base on the exponential smoothing algorithms are easy to apply and do not require the assumption of the stationarity of the time series. In the analyzed case relatively good forecast of the daily household water usage was obtained after applying the additive Winters model. On the other hand, ARIMA models allow for the precise forecast of the water usage, providing that the model parameters will be correctly identified and the condition of the time series stationarity will be met. In the case of non-stationary time series, before the analysis the series has to be transformed with e.g. differentiation method. In order to forecast the daily water usage in the farm-house, ARIMA model was used (0,1,2) with two parameters of the moving aver-age. The exponential smoothing as well as the ARIMA model allowed to obtain the similar forecast results, whereas the average value of the 10-days’household water usage forecast in the exponential model was 4,5% higher than the forecast obtained with the ARIMA model.