This paper presents a robust optimization approach for resilient energy forecasting, addressing the challenge of missing features in operational settings. The authors propose a robust regression model that is optimally resilient against missing features at test time, considering both point and probabilistic forecasting. The model is formulated as a robust optimization problem, and three solution methods are developed, leading to Linear Programming (LP) problems with varying degrees of tractability and conservativeness. The proposed methods are validated on prevalent energy forecasting applications, including electricity price, load, wind production, and solar production. The results demonstrate that the robust optimization approach outperforms imputation-based models and exhibits similar performance to retraining without the missing features, while maintaining computational practicality. The paper also highlights the importance of feature uncertainty in energy forecasting and introduces a novel approach to handle missing features through robust optimization. The proposed methods are evaluated on real-world data, and the results show that the robust optimization approach provides better performance in terms of accuracy and robustness compared to traditional methods. The paper concludes that the proposed robust optimization approach is a promising solution for resilient energy forecasting in industrial applications.This paper presents a robust optimization approach for resilient energy forecasting, addressing the challenge of missing features in operational settings. The authors propose a robust regression model that is optimally resilient against missing features at test time, considering both point and probabilistic forecasting. The model is formulated as a robust optimization problem, and three solution methods are developed, leading to Linear Programming (LP) problems with varying degrees of tractability and conservativeness. The proposed methods are validated on prevalent energy forecasting applications, including electricity price, load, wind production, and solar production. The results demonstrate that the robust optimization approach outperforms imputation-based models and exhibits similar performance to retraining without the missing features, while maintaining computational practicality. The paper also highlights the importance of feature uncertainty in energy forecasting and introduces a novel approach to handle missing features through robust optimization. The proposed methods are evaluated on real-world data, and the results show that the robust optimization approach provides better performance in terms of accuracy and robustness compared to traditional methods. The paper concludes that the proposed robust optimization approach is a promising solution for resilient energy forecasting in industrial applications.