Comparison of Decomposition Techniques in Forecasting the Quarterly Numbers of Pole-Mounted Transformer Failures

Pole-mounted transformers form a vital, final link between a utility and customers, and their failure means that power cannot be supplied. Adequate spares must be available in order to reduce duration of outages and undesirable consequences thereof. In this paper, the authors developed the forecasts of the quarterly numbers of failures for these transformers, using additive and multiplicative forecasting methods. Thereafter, the accuracies of the developed models are compared on the basis of a number of measures, including the mean absolute deviation (MAD), mean squared error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE). Firstly, the irregular components for additive model have values that lie about the “y equals to zero” line and for multiplicative model have values lie about the “y equals to 1” line, meaning that the values of the irregular components can be assumed to be 0 in the additive forecasting model and 1 in the multiplicative forecasting model. Secondly, it is found that the residuals of both methods are significantly large in relation to the values of observations of the time series of numbers of quarterly failures, which affects the accuracy of forecasts adversely. Finally, the capabilities of the additive and multiplicative decomposition seem to be comparable, with very similar values for measures of accuracy of forecast error obtained very similar, and one method being better than the other only dependent on the measure of forecast error considered.