Over the last two decades, significant efforts have been made to develop sophisticated energy consumption predictions via statistical learning algorithms. However, in many scenarios, the data is subject to uncertainty which would affect the performance of these predictions. For example, due to the stochastic nature of the physical world and the limitation of the sensing technologies, we may suspect that some energy or environmental sensors have reported noisy or biased values about the real situation. Therefore, it seems natural to expect that some of the data may be contaminated to some extent when developing the prediction models. In this project, we apply the linear regression method to predict the urban electricity use in the city of Abu Dhabi from a Robust Optimization (RO) perspective. Uncertainties are addressed in the feature, target, and both via a specific manner by constructing appropriate uncertainty sets. We state that adding robustness would not increase the computational cost of the nominal problem, and thus the problem can maintain its tractability. To evaluate the robustness to the linear regression method, we conduct a series of computational experiments.

According to the confidentiality agreement, this project will not be presented in detail in this portfolio.