Novel Methods for Smart Grid Intrusion Detection System Using Feature Selection Based on Improved Gravitational Search Algorithm

The smart grid architecture, which represents a deep integration of information technology and power systems, brings many conveniences to people. However, due to the highly open communication network and complex information interaction environment, it also faces more security risks. Existing intrusion detection algorithms based on machine learning cannot cope with the increasing features in the Energy Internet. To address this issue, this paper proposes the Improved Gravitational Search Algorithm (IGSA) for feature selection. Our core idea is to utilize IGSA for efficient feature selection, reducing the learning cost of machine learning methods and improving detection accuracy. Furthermore, to enhance the algorithm's global search capability and robustness, a novel elite selection strategy and adaptive mutation strategy are introduced. Experimental results on three public datasets demonstrate that IGSA improves detection accuracy by an average of 11.14% compared to other feature selection methods.