Predictions in Energy Economics – Tutorial on Household and PV Load Forecasting

The energy industry is facing significant challenges that require high adaptability and the utilization of flexibility. To optimize the integration of renewable energies and harness the potential of electric mobility and heat pumps, reliable predictions are indispensable. In this series of articles, some fundamental machine learning concepts have been introduced, such as appropriate metrics for evaluating prediction quality, various machine learning methods for time series forecasting, and their advantages and disadvantages. This article provides a practical example of forecasting time series using a Jupyter notebook.

Tutorial on Household and PV Load Forecasting

Figure 1: Result of predicting PV load profiles using linear regression and a Random Forest Regressor.

For this tutorial, public data from the pilot operation of the BDL project, which records household and PV load from various buildings, is used. Some features are initially extracted from the time series, which can be derived from temporal information, past values, or their statistical properties. Subsequently, the time series is divided into training and testing data, and the correlation of features in the training dataset is examined. Finally, linear regression and a Random Forest Regressor are applied to make predictions. Metrics for both models are also calculated and compared to assess their performance.