Abstract
Abstract
The widespread deployment of smart meters in modern energy grids introduces not only unprecedented opportunities for high-frequency analytics, but also acute privacy and security risks. With advanced metering infrastructure now capable of recording sub-second data, fine-grained energy profiles can reveal sensitive household activities, enabling both advanced load monitoring and potential misuse. This thesis proposes an integrated framework that advances resilient and privacy-preserving load monitoring and anomaly detection for next-generation smart energy management systems, targeting both operational efficiency and strong consumer data protection.
This study investigates how far high-resolution NILM analytics expose user privacy in CPS and whether formal guarantees (differential privacy) can be realized without erasing operational value. We use two utility probes—device-level disaggregation and subscriber-level anomaly detection—implemented with harmonics-aware features because they both improve accuracy and create leakage channels. The models are lightweight and interpretable to meet real-time budgets; their role is to calibrate the privacy–utility trade-off rather than to be the contribution.
To resolve the inherent privacy–utility trade-off, a novel key-dependent differential privacy protocol is introduced. This combines deterministic, cryptographically seeded noise generation with secure public-key encryption, ensuring that only authorized consumers can recover high-fidelity data while intermediaries observe only noisy ciphertexts. Unlike purely statistical approaches, this design binds privacy protection to cryptographic key control, making privacy budgets enforceable across distributed systems and resistant to passive and active attacks. The resulting approach addresses gaps in existing standards, which often mandate anomaly detection in principle but lack concrete, privacy-conscious methods for harmonics-rich AMI analytics.