Power systems are rapidly adopting Internet of Things sensors and wireless communication to replace legacy wired measurement infrastructure, promising lower costs and greater flexibility. However, this shift introduces a critical vulnerability: wireless measurements are far more susceptible to impulsive noise and corrupted data, which can severely degrade the accuracy of power system state estimation—the foundational process that allows grid operators to monitor and control the network safely.
Traditional approaches rely on central processing of all sensor data, which is computationally expensive and slow when filtering bad data across thousands of distributed sensors. Researchers have now proposed DGM-WLS, a distributed framework that pushes data quality assurance to the edge—directly at each sensor node—before transmission to the control centre.
The algorithm employs two stages. First, unsupervised machine learning clustering techniques (DBSCAN and Gaussian mixture models) run locally on sensors to identify and remove corrupted measurements in real time. This pre-filtering eliminates bad data before it ever reaches the central fusion centre, reducing communication overhead and computational burden. Second, the central control station constructs weighted least squares matrices from the pre-cleaned data to perform final state estimation.
Simulation testing on the IEEE 30-bus network demonstrates substantial improvements. DGM-WLS achieved 11% better estimation accuracy compared to the industry-standard iterative reweighted least squares algorithm, and 39% better than conventional weighted least squares methods. Critically, the new approach maintains constant convergence in just five iterations regardless of noise levels, and executes 15 times faster than existing robust methods.
The framework's edge-computing architecture aligns with modern utility digitalization strategies, reducing dependence on centralised processing while improving resilience to cyber and physical threats. As wireless IoT deployment accelerates across distribution networks, localized intelligent filtering becomes essential for maintaining accurate situational awareness and reliable grid operation.
