Title :
Data partitioning for training a layered perceptron to forecast electric load
Author :
El-Sharkawi, M.A. ; Marks, RJ, II ; Oh, S. ; Brace, C.M.
Author_Institution :
Dept. of Electr. Eng., FT-10, Washington Univ., Seattle, WA, USA
Abstract :
The multi-layered perceptron (MLP) artificial neural network has been shown to be an effective tool for load forecasting. Little attention, though, has been paid to the manner in which data is partitioned prior to training. The manner in which the data is partitioned dictates much of the structure of the corresponding neural network. In many neural network forecasters, a different neural network is used for each day. The authors compare the performance of a daily partitioned neural network and hourly partitioned neural network. In the experiments, the hourly partitioned neural network forecaster has better performance than the daily partitioned neural network forecaster.
Keywords :
feedforward neural nets; learning (artificial intelligence); load forecasting; power engineering computing; daily partitioned neural network; data partitioning; electric load forecasting; hourly partitioned neural network; layered perceptron training; neural network; Airports; Artificial neural networks; Feeds; Load forecasting; Multilayer perceptrons; Neural networks; Neurons; Statistics; Temperature; Training data;
Conference_Titel :
Neural Networks to Power Systems, 1993. ANNPS '93., Proceedings of the Second International Forum on Applications of
Conference_Location :
Yokohama, Japan
Print_ISBN :
0-7803-1217-1
DOI :
10.1109/ANN.1993.264348