Forecasting Nord Pool day-ahead prices with Python

Tarjei Kristiansen

Abstract


This paper presents a Nord Pool forecast model for hourly day-ahead prices, utilizing the Python software. The model is an autoregressive model based on [1] and the data spans the period from 2004 to 2011. The targets (i.e. dependent variables) are the hourly day-ahead prices for a certain hour during the day while the features (i.e. independent variables) are the prices for the same hour the previous two days and the previous week, the minimum price for the previous day, four weekday dummy variables, including the demand and wind for the actual hour. We test the model in a simple linear regression framework with cross-validation. Next, we utilize regularized regressions including Ridge and Lasso.  Finally, we utilize a Keras neural network. The models are evaluated with the mean absolute percentage error (MAPE) criterion, R-square and scatterplots. The results demonstrate that the models perform well and could add value for a market player.

Full Text:

PDF