9th WSEAS International Conference on Circuits, Systems, Electronics, Control & Signal Processing
(EN)
The modern methods for power system load prediction are usually based on Artificial Neural
Networks (ANN), which present satisfactory results. However, the estimation of the confidence intervals can
not be applied directly, unlike to the classical forecasting methods. One of the most commonly used methods is
the re-sampling technique, which calculates the respective confidence interval based on the training data set.
The limits of the training set confidence interval are also applied in the case of the real prediction giving
satisfactory but slightly underestimated results. The targets of this paper are: (1) to apply the basic re-sampling
method for the short term forecasting of the next day load in the interconnected Greek power system using an
optimized ANN proving the aforementioned disadvantage and (2) to propose a modified re-sampling technique
using a proper corrective multiplication factor. Finally, the next day load demand of the test set is estimated
using the best ANN structure and the modified confidence intervals.
(EN)
