Combined use of unsupervised and supervised learning for daily peak load forecasting

摘要: Abstract In this paper, we have aimed to present a hybrid neural network model for daily electrical peak load forecasting (PLF). Since loads usually follow similar patterns, classification of data improves the accuracy forecasts. Several factors in load, e.g. weather temperature, relative humidity, wind speed and cloud cover, were introduced into order enhance forecast quality. Most attempts literature been intuitive empty justification. proposed novel approach clustering by using self-organizing map. The Davies–Bouldin validity index was determine best clusters. A feed forward (FFNN) has developed each cluster provide PLF. Eight training algorithms also used train FFNNs. Applying principal component analysis (PCA) decreased dimensions network’s inputs led simpler architecture. To evaluate effectiveness (PHM), performed developing FFNN that uses un-clustered data. results proved superiority PHM. Linear regression (LR) models PLF, indicated PHM produces considerably better forecasts than those LR models. Furthermore, show suggested significantly on too.