Daily load forecasting based on rough sets and relevance vector machine

Abstract

Relevance vector machine (RVM) is a novel kernel methods based on Bayesian frame. It is similar to support vector machine (SVM), but it needs much less learning parameters to adjust and has a better forecast accuracy. However, due to the fact that the RVM is not suitable for the forecast task with large sample dataset, the application of RVM in daily load forecasting is limited. This paper proposes a daily load forecasting model based on rough sets and RVM. This model takes load data and weather information as condition attributes, classifies the historical load dataset into 24 hourly sub-dataset, utilizes the theory of rough sets for each sub-dataset to obtain the reduced condition attribute which is used as the input of the RVM model, thus avoids the blindness in building the load forecast model. Therefore the issue of daily load forecasting is transformed into the issue of small sample forecast. In experiments, this model illustrates a high practicability and forecast accuracy.