Pattern sequence-based energy demand forecast using photovoltaic energy records

Considering recent trends in energy technology development, consumer's energy demand could be influenced by the renewable energy supply in any way. A simple extension of pattern sequence-based forecasting (PSF) enables us to predict demand curves based on the correlated bidimensional time-series by using co-occurrence patterns of energy supply and demand. However, prediction accuracy of PSF deeply depends on the clustering result, which is used for pattern matching. In this paper, a promising clustering method based on nonnegative tensor factorization is applied for this task and evaluated experimentally from the viewpoint of prediction accuracy.