Short-term electricity load forecasting-A systematic approach from system level to secondary substations

Energy forecasting covers a wide range of prediction problems in the utility industry, such as forecasting demand, generation, price, and power load over time horizons and different power levels. Short-term load forecasting allows the system operator to make important decisions during network management and planning, which represents an economic improvement in the global electrical system, as well as acts as a fundamental component to address challenges imposed by the energy transition. Despite the urgent need, research into the current state-of-the-art in low voltage forecasting, other than at the smart meters level, has not yet been carried out in an in-depth way. We propose a systematic approach from system level to low voltage considering not only the performance of the models but also the applicability, interpretability, and reproducibility of the method/model. An initial benchmark model was compared against improved regression models enhanced by introducing new synthetic explanatory variables and a different regression technique, generalized additive models. The error was reduced by 42%-47% compared to the benchmark model, preserving the interpretability. Additionally, a simple ensemble method was evaluated to determine how it improves accuracy for specific periods, such as weekends, summer vacation, or public holidays, in which modeling is particularly difficult using standalone generalized additive models. The method was applied to the national power load and, for the first time, to all 100,000 secondary substations that integrate the Portugal power grid, rather than to tackle the few open datasets in this area. The approach was used to build a daily forecasting system live called PREDIS (Portuguese acronym for Distributed Prediction) whose outcomes are used to anticipate load peaks and network constraints in the context of the Portugal distribution system operator.

Automated summary

Energy forecasting is crucial for the utility industry, covering a wide range of prediction problems. Short-term load forecasting plays a vital role in network management and planning, addressing challenges in the energy transition. This study proposes a systematic approach considering model performance, applicability, interpretability, and reproducibility, achieving significant improvements compared to benchmark models.