Short-term hourly price forward curve prediction using neural network and hybrid ARIMA-NN model

Even though the electricity HPFC (Hourly Price Forward Curve) is still surprisingly under-researched the prediction of electricity prices is highly important in order to keep power plants profitable or in order to optimize the electricity purchases based on future customers demand. In this work two methods to model and predict HPFC based on neural networks will be proposed and compared to more common time series approach - specifically ARIMA model. In the first method the neural network is applied to model the price at desired time as a function of some past observations and also to capture the seasonal character of the data. The second method uses hybrid model which consist of an ARIMA model combined with neural network. The ARIMA is used to capture linear patterns in the data. Then the neural network is used to model remaining non-linear residuals. In this case the whole process is done on deseasonalized data set. Both methods provide more accurate predictions than standard time series approach (in this case ARIMA model) and results clearly state that the neural network approach is a valid alternative for forecasting (not just) economic time series.