From Recognition to Reaction A Cognitive Engine for Closed-Loop Power Quality Management
Keywords:
Power Quality Disturbance Recognition, Empirical Mode Decomposition (EMD) Convolution Neural Networks (CNN), Linear Discriminant Analysis (LDA), Cognitive Engine, Smart Grid MonitoringAbstract
Power Quality Disturbance (PQD) recognition frameworks have achieved exceptional classification accuracy for
example; the EMD-CNN-LDA-KNN framework reported 99.8% accuracy under noiseless conditions and 97.0%
under 10 dB noises but remain decoupled from physical compensation hardware. This paper proposes a Cognitive
Engine (CE) that bridges the gap between intelligent sensing and real-time compensation. The CE integrates the
EMD-CNN-LDA-KNN framework with an Interface Strategy Mapper (ISM) that translates recognized disturbance
types into actionable control signals for a UPQC-SPV compensator. The ISM employs a rule-based mapping table
covering 10 IEEE 1159 disturbance classes with severity levels, producing series and shunt VSI utilization
percentages. The CE is designed for low-latency operation targeting less than one grid cycle to enable closed-loop
"sense-decide-compensate" functionality. Key contributions include: (1) a structured ISM with justified categorical
allocation where voltage-domain disturbances receive high series utilization, current-domain disturbances receive
high shunt utilization, and combined disturbances receive hybrid compensation; (2) LDA-based 2D visualization for
class separation and interpretability; and (3) a complete, reproducible evaluation methodology with explicit
hypotheses for future validation. This work provides a foundational architecture for autonomous, cognitive power
quality management systems, with all quantitative performance claims framed as open empirical questions.
