Forecasting Hourly Electricity Prices using Principal Components

This research considers forecasting 1 day-ahead hourly electricity prices in the Norwegian power market, after a principal component analysis has been applied on the data. As a way of gaining forecast accuracy, combinations of weighted individual forecast were also considered. Comparisons between the accuracy of five distinct models were made. A base model; in which principal component analysis was applied to the dataset comprising all 24 hourly prices, was used as standard to compare with the alternatives, by means of three evaluation criteria. Ultimately, the results were somewhat the same for the peak/off-peak hour model and the 4 intra-day period model: both were outperformed by the base model, mainly for periods of high volatility. The weighted principal components model simply generated poor results. The combination of weighted base model with the weighted peak/off-peak model was the only model which generated better for volatile periods, but this at the expense of losing accuracy in other periods.