A robust, resilience machine learning with risk approach: a case study of gas consumption

This research suggests a novel Robust, Resilient machine learning that focuses on the Risk approach (3R) in a hard situation for the first time. A robust stochastic LASSO regression is proposed for predicting gas consumption. This model tries to optimize a new form of LASSO regression by minimizing the expected value of mean and maximum and EVaR of MAD with the penalty of the regression coefficient. The 3R requirements include robustness and resiliency in the mathematical model approach by paying attention to disaster and flexibility and the risk-averse method by considering risk criteria like max function and EVaR determined. The results show that the value of Robust and Resiliency Mean Absolute Deviation with Risk approach (RRMADR) and R-squared of the hybrid natural logarithm function is 26.81% less than the polynomial LASSO model (base model). In addition, the polynomial regression without LASSO performs better than the base model. This subject is because of the penalty of LASSO regression. The conservatism coefficient, confidence level, penalty of LASSO, resiliency coefficient, and scenario probability are analyzed. Decreasing conservatism and resiliency coefficient, increasing confidence level and penalty of LASSO increase RRMADR and decrease R-squared of the model.