Sensor Fusion

Fusion architectures provide graceful degradation rather than catastrophic failure. By cross-validating measurements from multiple sources, the system can detect and isolate faulty sensors.

• N-1 Sensor Failure: The state estimate remains accurate even if a single PMU or meter drops offline.
• Bad Data Detection: Residual analysis automatically flags and rejects grossly erroneous measurements before they corrupt the grid state.
• Example: If a SCADA voltage reading conflicts with a PMU phasor measurement, the Kalman filter weights the higher-precision PMU data more heavily, effectively ignoring the suspect value.

Fusing data requires precise time synchronization and topological awareness. Measurements must be aligned to a common reference frame before they can be combined.

• GPS Time Stamping: PMU data is synchronized to the microsecond via GPS, enabling wide-area correlation of dynamic events.
• Latency Compensation: Stream processing engines buffer and reorder asynchronous SCADA scans (2-4 second intervals) to align with fast PMU streams (30-60 samples per second).
• Topology Processor Integration: The fusion engine dynamically maps measurements to the current electrical bus-branch model, accounting for breaker status changes.

Every fused estimate carries a mathematically rigorous confidence bound. Sensor fusion does not just produce a single value; it outputs a probability distribution.

• Measurement Covariance: Each sensor's noise characteristics are modeled as a covariance matrix, allowing the fusion algorithm to optimally weight high-precision sources.
• Aleatoric vs. Epistemic Uncertainty: The system distinguishes between irreducible sensor noise and model gaps that can be reduced with better calibration.
• Example: A fused voltage estimate of 1.02 pu ± 0.005 pu gives operators a clear understanding of the estimate's reliability for critical switching decisions.

Grid sensors operate at fundamentally different reporting rates. Fusion algorithms must asynchronously assimilate fast and slow measurements into a unified state vector.

• PMU Data: 30-120 samples per second, capturing dynamic oscillations.
• SCADA Data: 2-4 second scan intervals, providing steady-state voltage and power flows.
• Smart Meter Data: 15-minute to hourly intervals, offering granular load profiles at the grid edge.
• Algorithmic Approach: Ensemble Kalman filters and sequential Bayesian updating incorporate each measurement at its native rate, updating the state estimate without waiting for slow scans.

Fusion algorithms exploit the physical laws governing power systems to validate measurements across different modalities. Kirchhoff's laws act as a physics-based consistency check.

• Power Balance Verification: The sum of power injections at a bus must equal the sum of flows on connected lines, providing a hard constraint for data reconciliation.
• Voltage-Angle Consistency: PMU phase angle measurements must be consistent with SCADA power flow measurements across the same transmission corridor.
• Anomaly Detection: A sudden divergence between physically related measurements triggers an alert for sensor malfunction, communication errors, or even cyber intrusion.

No single sensor type provides complete grid visibility. Fusion combines complementary strengths to achieve full observability where individual systems fall short.

• PMUs provide high-resolution dynamic data but have limited deployment coverage due to cost.
• SCADA offers broad coverage but slow scan rates and no phase angle data.
• Smart Meters deliver edge visibility but with significant latency.
• Fusion Outcome: The combined measurement set achieves numerical observability, enabling the state estimator to solve for voltage magnitude and angle at every bus in the network model.