Electricity price forecasting using a clustering approach

This paper presents a new method to forecast the short term electricity price as a kind of time series. A clustering based forecasting method is introduced. The proposed method contains input-output decomposition and using a simple clustering approach to classify them and then for a new input (a specified number of past time series values), these clusters are sorted according to the probabilities calculated by using the Bayespsila formula. The prediction is then generated using the weighted average of the forecasted outputs of M clusters with highest probabilities.