Intelligent Scene-Adaptive Desensitization: A Machine Learning Approach for Dynamic Data Privacy in Virtual Power Plants

Abstract

In the context of virtual power plants (VPPs), the one-size-fits-all approach of traditional static desensitization methods proves inadequate due to the diverse and dynamic operational scenarios encountered. These methods fail to provide the necessary flexibility for varying data privacy requirements across different scenarios. To address this shortcoming, our research introduces a dynamic desensitization method specifically designed for VPPs. Leveraging machine learning for adaptive scene recognition, the method adjusts data privacy levels intelligently according to each unique scenario. A novel similarity utility function and a Gaussian processes-based differential privacy algorithm ensure tailored and efficient privacy protection. Experimental results highlight an 87.5% accuracy in scene recognition, validating our method’s capability to adapt to diverse scenarios effectively. This study contributes to the field by providing a nuanced approach to data protection, effectively addressing the specific needs of complex VPP environments.