Neural network model ensembles for building-level electricity load forecasts

• A practical forecasting model for day-ahead building-level load forecasting.
• We propose a simple, automated way to select model parameter values.
• Model outperforms current best model for a set of operational C&I sites by up to 50%.
• Our model enables a dynamic trade-off between different factors influencing load.
• It also enables a dynamic trade-off between model parameter values.

The future power grid is expected to provide unprecedented flexibility in how energy is generated, distributed, and managed, which increasingly necessitates an ability to perform accurate short-term small-scale electricity load and generation forecasting, e.g., at the level of individual buildings or sites. In this paper, we present a novel building-level neural network-based ensemble model for day-ahead electricity load forecasting and show that it outperforms the previously established best performing model, SARIMA, by up to 50%, in the context of load data from half a dozen operational commercial and industrial sites. In addition, we show a straightforward, automated way to select model parameters, making our model practical for use in real deployments.