Voltage-Current (V-I) Trajectory Clustering

The core principle involves plotting instantaneous current against instantaneous voltage over a single complete AC cycle (typically 1/50th or 1/60th of a second). Normalization scales both axes to a standard range (e.g., [-1, 1]), making the resulting V-I trajectory invariant to the absolute magnitude of voltage or current. This ensures a 100-watt incandescent bulb and a 1500-watt resistive heater produce identical straight-line trajectories, while a vacuum cleaner's motor produces a distinctly looped shape regardless of load. The trajectory captures harmonic content, phase shift, and non-linear conduction angles as geometric features.

Once the V-I trajectory is plotted, quantitative descriptors are extracted to feed into clustering algorithms. Key features include:

• Loop Area: The enclosed area of the trajectory, indicating reactive power and hysteresis.
• Asymmetry: Deviation from symmetry about the origin, revealing non-linear rectification.
• Mean Curve Slope: The average derivative, correlating with the load's power factor.
• Self-Intersection Points: Count and location of trajectory crossings, characteristic of phase-controlled devices like dimmers.
• Curvature and Envelope Shape: Statistical moments describing the trajectory's contour.

Extracted trajectory features are projected into a multi-dimensional space where unsupervised algorithms like DBSCAN or K-Means group similar shapes. This automatically creates an appliance taxonomy without manual labeling. Resistive loads (heaters, incandescent bulbs) form a tight cluster near a straight line. Inductive motor loads (fans, compressors) form elliptical clusters. Power electronics (LEDs, variable frequency drives) form distinct clusters with sharp corners and flat tops caused by bridge rectifiers. New unknown devices are classified by their proximity to existing cluster centroids.

V-I trajectory clustering fundamentally requires high-frequency sampling, typically in the kilohertz range (1 kHz to 30 kHz). Standard smart meter data at 1 Hz is insufficient because it only captures RMS values, obliterating the intra-cycle waveform shape. A sampling rate of at least 60 samples per AC cycle (3.6 kHz for a 60 Hz grid) is necessary to resolve the harmonic distortions and transient edges that define the trajectory's unique fingerprint. This necessitates specialized metering hardware or high-resolution embedded sensors.

A critical advantage of V-I trajectory clustering over raw current waveform analysis is its inherent robustness to voltage fluctuations. Because the trajectory is a parametric plot of current versus voltage, a sag or swell in the supply voltage stretches or compresses both axes simultaneously. The normalized shape remains geometrically similar. This decoupling from supply quality makes the method highly reliable in real-world distribution grids where voltage is rarely a perfect sinusoid, preventing misclassification during grid disturbances.

Complex appliances with variable-speed drives or multiple operational states do not produce a single V-I trajectory. A modern inverter-driven heat pump generates a family of trajectories corresponding to different compressor speeds. Clustering must therefore operate in a hierarchical or multi-modal fashion. The system first identifies the appliance class (e.g., 'variable frequency motor') by the broad shape family, then uses a secondary classifier or regression model on trajectory parameters like loop width to estimate the specific operational state or power consumption level.

V-I trajectory clustering serves as the primary feature extraction backbone for modern NILM systems. By mapping the normalized voltage-current locus over one AC cycle, the algorithm creates a unique shape fingerprint that distinguishes appliances with overlapping real power consumption.

• Resistive loads (heaters, incandescent bulbs) produce a straight diagonal line
• Electronic loads (LEDs, computers) exhibit sharp phase-controlled peaks and flat tops
• Motor-driven appliances (refrigerators, HVAC) generate elliptical hysteresis loops

This shape-based classification achieves >95% accuracy in identifying major household appliances from a single sensing point at the main breaker panel.

Utility-grade smart meters with high-frequency sampling capabilities (kHz range) now embed V-I trajectory clustering directly on the meter's microcontroller. This enables edge-based appliance identification without streaming raw waveform data to the cloud.

• Reduces bandwidth requirements by 99% compared to raw waveform transmission
• Provides real-time appliance-level feedback to consumers via in-home displays
• Enables utilities to segment load composition without customer surveys

The meter transmits only the cluster labels and confidence scores, preserving consumer privacy while delivering granular load intelligence to distribution system operators.

V-I trajectory analysis detects incipient mechanical faults in motor-driven equipment by tracking subtle deformations in the elliptical current locus. Changes in the trajectory's area, inclination angle, and asymmetry correlate with specific failure modes.

• Bearing wear introduces high-frequency oscillations along the ellipse perimeter
• Rotor bar breakage creates periodic amplitude modulation visible as trajectory thickness variation
• Misalignment shifts the symmetry axis of the hysteresis loop

This non-invasive technique identifies faults weeks before catastrophic failure, enabling condition-based maintenance scheduling without installing vibration sensors on each machine.

Commercial building energy auditors deploy portable V-I trajectory analyzers to rapidly inventory plug loads across office floors. A single measurement at a power strip captures the composite trajectory, which is then decomposed into individual appliance signatures.

• Identifies phantom loads (vampire electronics) consuming power in standby mode
• Distinguishes ENERGY STAR compliant devices from inefficient equivalents by trajectory shape
• Generates automated audit reports mapping each outlet to specific device categories

This accelerates the auditing process from hours to minutes per floor, enabling large-scale commercial energy efficiency programs with minimal labor cost.

Distribution utilities use V-I trajectory clustering to distinguish EV charging events from other high-power loads on residential feeders. Modern EV chargers produce a distinctive trajectory characterized by a near-unity power factor rectangle with high-frequency switching ripple at the edges.

• Level 1 chargers exhibit a narrow rectangular trajectory with 120V scaling
• Level 2 chargers show a wider rectangle with distinct harmonic injection patterns
• DC fast chargers present a fundamentally different three-phase rectifier signature

This enables utilities to quantify EV adoption rates per distribution transformer without requiring customer self-reporting or dedicated submetering infrastructure.

Safety certification laboratories apply V-I trajectory clustering to identify counterfeit or non-compliant electrical products entering the supply chain. Genuine certified devices exhibit a tightly defined trajectory envelope, while counterfeits deviate due to inferior components.

• Substandard power supplies show abnormal phase-angle control not present in certified units
• Counterfeit circuit breakers fail to replicate the precise magnetic saturation curve of authentic devices
• Non-compliant LED drivers inject harmonic distortion visible as trajectory boundary roughness

This automated screening method processes thousands of units per day, providing a scalable defense against dangerous counterfeit electrical goods.