A day ahead energy load forecasting: Machine learning based model application on an Italian large enterprise

Abstract

Electric energy costs reduction is a critical aspect for industrial enterprise management. Short-term load forecast is a very important activity both for enterprises and for electric grid manager. Applying a short-term load forecasting method, enterprises can cut energy costs. Furthermore, such an application contributes to the reduction of grid manager interventions to minimize imbalance problems. In this context, industrial sites able to self-produce more than their energy need, have to adopt suitable load forecasting systems both to control energy consumption and to limit dispatching burden due to the feed of power into the grid. Correlation between industrial site energy consumption and industrial productions has encouraged the authors to develop a methodology that provide short-term electric load forecasting, based on machine learning, applicable in a generalized manner using available production plan data. To develop such a complex model, a tool composed of several parts has been implemented. Forecasting model structure is composed of 2 parts, one for prediction and one for imbalance calculation. Neural networks have been used in prediction phases because of their possibility to manage large dataset and to find nonlinear correlation between available variables. Application of developed methodology on real industrial gathered data has provided important results. Forecasting method, although calculated imbalances have reached high values, has led to get around 28% saving on balancing costs compared to enterprise previously applied forecasting method.