Fairness-aware User Classification in Power Grids

With the development of intelligent measurement systems, power grids improved the reliability and efficiency according to the vast amount of collected information. Machine learning techniques are increasingly used in smart grids since they are efficient to deal with the huge amount of collected data and extract valuable information. The availability of large-scale data enables the employment of machine learning methods in various tasks in power grids. However, large-scale deployment of machine learning model relies on how trustworthy the model is. While sole pursuit of overall learning performance, may leads to unfair results. Specifically, the model may unintentionally discriminate different subgroups. Machine learning models for smart grids also have fairness concerns. Power consuming users and buildings with different power consumption patterns may be treated with different conditions. To mitigate the unfairness, we propose accuracy parity, equal opportunity and predictive equal-ity regularizers, which can be used for different classification tasks in power grids to mitigate the corresponding performance discrepancy. Experiments on user classification using loading data show that the regularizers are effective at avoiding disparate mistreatment and sometimes can benefit the overall performance with fine tuning weights.