Artificial intelligence driven demand forecasting: an application to the electricity market

Abstract

Demand forecasting with maximum accuracy is critical to business management in various fields, from finance to marketing. In today’s world, many firms have access to a lot of data that they can use to implement sophisticated models. This was not possible in the past, but it has become a reality with the advent of large-scale data analysis. However, this also requires a distributed thinking approach due to the resource-intensive nature of Deep Learning models. Forecasting power demand is of utmost importance in the energy industry, and various methods and approaches have been employed by electrical companies for predicting electricity demand. This paper proposes a novel multicriteria approach for distributed learning in energy forecasting. We use a Quadratic Goal Programming approach to construct a robust decision rule ensemble that optimizes a pre-defined loss function. Our approach is independent of the loss function’s differentiability and is also model agnostic. This formulation offers interpretability for the decision-maker and demonstrates less proclivity of regression against the mean that affects autoregressive models. Our findings contribute to the field of energy forecasting and highlight the potential of our approach for enhancing decision-making in the energy industry.

Keywords: Demand forecasting, Federated learning, Deep learning, Multiple criteria decision making