Energy management in smart homes with adversary detection and noise mitigation using a moving prediction window scheme

Abstract

This paper presents an energy management algorithm for scheduling a large number of residential households, utilizing the moving prediction window to make it resilient against false data injection attacks and communication noise. We model multiple communities of smart homes, each managed by a local controller, where self-interested residential households engage in a global non-cooperative game. The cost functions of the households are influenced by a constrained aggregated power consumption term across all households from all communities. The interactions among households are modeled through a multi-community aggregative game. To reach a Nash equilibrium, we propose an iterative algorithm wherein local controllers estimate the coupling aggregate term and corresponding Lagrange multiplier for their respective households and collaborate with other controllers via an unreliable communication network to refine the aggregate estimations. Given the vulnerability of the communication network to external intrusions and the potential for internal controllers to behave maliciously, we explore a moving horizon window technique to detect false data injection attacks and mitigate communication noise. In this regard, first, a moving horizon estimator predicts the community’s current behavior based on historical data; second, a residual-based detection mechanism flags an attack when predicted residuals exceed a dynamic threshold; and third, corrupted measurements are discarded, and the average of the predictions is used in the Krasnoselskii-Mann update to reduce the noise impact. Numerical simulations show the effectiveness of the proposed algorithm in increasing the speed of reaching consensus by about 30 percent while managing the energy consumption of households.