Utilities and energy traders are often faced with significant challenges when forecasting certain customer segments, in particular industrial operations. For instance, the lack of a sufficiently long history often prevents traditional statistical forecast methodologies from identifying key correlations between certain predictors – even if such correlations are immediately perceivable to a trained human expert. A similar difficulty is faced when a priori knowledge of the prediction target must be included into the forecast model.
In this context we introduce metaScript Master, a powerful yet flexible rule-based scripting language capable of interacting with existing machine learning methods. Initially conceived as post-processing strategy, its scope has grown to encompass the generation of synthetic predictor time series, forecast model or algorithm selection, and the execution of predictions based exclusively on rule-based computations, capable of completely replacing conventional statistical algorithms.
By detecting structural changes, smoothing transitions or introducing previously not observed factors, the implementation of metaScript Master consistently reduced prediction errors across different energy segments including gas consumption, power generation, and district heating, both in short-term and long-term forecast periods.