Power System Operations
Distribution Grid State Estimation
Network parameter and state estimation from sparse SCADA, AMI, and PMU measurements
Introduction
Electric distribution grids are undergoing rapid transformation with the integration of rooftop solar, electric vehicles, and battery storage. Utilities need accurate real-time visibility into voltage profiles, line flows, and DER (distributed energy resource) injections to maintain reliability and optimize operations.
The challenge: distribution systems have sparse instrumentation compared to transmission networks. SCADA provides measurements at substations, advanced metering infrastructure (AMI) gives customer-level data with delays, and micro-PMUs (μPMU) offer high-resolution phasors at limited locations. Can we reconstruct the full network state and estimate unknown line parameters from this partial data?
Why this matters: Accurate state estimation enables volt-var optimization (reducing losses by 2–5%), hosting capacity analysis for solar/EV interconnection, fault location identification, and proactive voltage regulation—preventing equipment damage and customer complaints.
Electrical Network Science
A distribution feeder operates as a radial network governed by fundamental electrical laws:
- Kirchhoff's Current Law (KCL): Net current injection at each bus equals zero (sum of loads, generation, and line flows)
- Kirchhoff's Voltage Law (KVL): Voltage drops along branches equal sum of impedance drops
- AC power flow: Active and reactive power flows depend on voltage magnitudes, phase angles, and branch impedances
- Load variability: Residential/commercial loads vary hourly, PV generation follows solar irradiance, EVs charge in evening peaks
- Voltage regulation devices: Capacitor banks, load tap changers (LTCs), and voltage regulators dynamically adjust to maintain voltage within ±5% of nominal
Mathematical Model: AC Power Flow Equations
The distribution network is modeled as a directed graph with N buses and L branches:
States (per bus $i$):
$$\begin{aligned}
V_i &= \text{voltage magnitude [pu or V]} \\
\theta_i &= \text{voltage phase angle [rad]} \\
P_i, Q_i &= \text{active and reactive power injection [kW, kVAR]}
\end{aligned}$$
Power Flow Equations (nonlinear algebraic):
$$\begin{aligned}
P_i &= V_i \sum_j V_j (G_{ij} \cos(\theta_i - \theta_j) + B_{ij} \sin(\theta_i - \theta_j)) \\
Q_i &= V_i \sum_j V_j (G_{ij} \sin(\theta_i - \theta_j) - B_{ij} \cos(\theta_i - \theta_j))
\end{aligned}$$
Where:
$$\begin{aligned}
G_{ij} + jB_{ij} &= \text{admittance matrix element (from line impedances)} \\
\sum &\text{ over all buses } j \text{ connected to bus } i
\end{aligned}$$
Branch Flows:
$$\begin{aligned}
P_{ij} &= V_i^2 G_{ij} - V_i V_j (G_{ij} \cos(\theta_i - \theta_j) + B_{ij} \sin(\theta_i - \theta_j)) \\
Q_{ij} &= -V_i^2 B_{ij} - V_i V_j (G_{ij} \sin(\theta_i - \theta_j) - B_{ij} \cos(\theta_i - \theta_j))
\end{aligned}$$
Constraints:
$$\begin{aligned}
V_{\min} &\leq V_i \leq V_{\max} \quad \text{[ANSI C84.1: 0.95 pu} \leq V \leq \text{1.05 pu]} \\
P_{ij}^2 + Q_{ij}^2 &\leq S_{\max,ij}^2 \quad \text{[thermal limits]}
\end{aligned}$$
Unknown Parameters (varies by feeder)
| Parameter |
Description |
Units |
Typical Uncertainty |
| Rij, Xij | Line resistance and reactance | Ω/km | ±10–30% (corrosion, wrong database) |
| Zij = Rij + jXij | Branch impedance | Ω | ±15% (aging, temperature) |
| Pi_DER | Rooftop PV power injection | kW | ±20% (no telemetry at many sites) |
| Qi_load | Reactive power demand | kVAR | ±30% (power factor uncertainty) |
| tap_pos | Transformer tap position | — | ±1 tap (telemetry failure) |
| Vi, θi | Unmeasured bus voltages/angles | pu, rad | Hidden states |
| topology | Switch status (NO/NC) | binary | Inferred from measurements |
State Estimation Challenge
Given measurements from a distribution feeder with 100–500 buses:
- SCADA (RTU): Substation voltage, feeder head active/reactive power, capacitor bank status [4-second refresh, <1% error]
- AMI (smart meters): Customer-level voltage magnitudes, active power [15-minute intervals, ±1% error, 5–30 min latency]
- μPMU: Voltage phasor (magnitude + angle) at 2–5 strategic buses [30–60 Hz, ±0.1% error]
- Pseudo-measurements: Load forecasts from historical data [±20–40% error]
The inverse problem is formulated as weighted least squares with inequality constraints:
Minimize:
$$J(x, \theta) = \sum_k \left[ \frac{(z_{\text{meas},k} - h_k(x, \theta))^2}{\sigma^2_k} \right] + \lambda \cdot ||x_{\text{prior}} - x||^2_P$$
Subject to:
$$\begin{aligned}
g(x, \theta) &= 0 \quad \text{[power flow equations: } P_i, Q_i \text{ balance]} \\
x_{\min} &\leq x \leq x_{\max} \quad \text{[voltage limits, physical bounds]}
\end{aligned}$$
Where:
$$\begin{aligned}
x &= [V_1, \ldots, V_N, \theta_1, \ldots, \theta_N] \quad \text{[states: voltages and angles]} \\
\theta &= [R_{12}, X_{12}, \ldots, P_{\text{DER},5}, \ldots, \text{tap}_{\text{pos}}] \quad \text{[parameters]} \\
z_{\text{meas}} &= [V_{\text{sub}}, P_{\text{sub}}, Q_{\text{sub}}, V_{\text{AMI},i}, P_{\text{AMI},i}, V_{\mu\text{PMU},j}, \theta_{\mu\text{PMU},j}] \\
h_k(x, \theta) &= \text{measurement function (maps states to observations)} \\
\sigma^2_k &= \text{measurement variance (SCADA: 0.01}^2\text{, AMI: 0.02}^2\text{, pseudo: 0.3}^2\text{)}
\end{aligned}$$
Observability issue: With sparse measurements, some buses have no direct observations. The system is weakly observable—we must use power flow constraints and pseudo-measurements (load forecasts) to propagate information through the network.
Simulation Design
Generate synthetic feeder data for IEEE 123-bus or 8500-node test feeders:
- Define base topology: Radial feeder with primary/secondary distribution lines, transformers, regulators, capacitors
- Time-series profiles:
- Residential loads: NREL OpenEI profiles (winter/summer, weekday/weekend)
- Rooftop PV: 30% penetration, 15-minute granularity with cloud transients
- EV charging: Evening peak (6–10 PM), Level 2 chargers (7 kW)
- Voltage regulator actions: LTC operates when V < 0.97 or V > 1.03, ±16 taps
- Add noise and dropouts: AMI: 2% random dropouts per interval, ±1% Gaussian noise; μPMU: occasional GPS sync loss
- Parameter variations: Line impedances drift ±10% (temperature effects), 20% of PV sites have no telemetry
Estimation Methods
1. Weighted Least Squares (WLS)
Classical distribution system state estimation:
- Solve nonlinear optimization via Gauss-Newton or Levenberg-Marquardt
- Iterate until convergence: ||x^(k+1) - x^(k)|| < ε
- Chi-square test for bad data detection (outlier rejection)
- Computational cost: 0.1–1 second per snapshot (suitable for 4-second SCADA cycle)
2. Extended Kalman Filter (EKF)
For dynamic state estimation with time-series AMI data:
$$\begin{aligned}
\text{State vector: } & x = [V_1, \ldots, V_N, \theta_1, \ldots, \theta_N, P_{\text{DER},1}, \ldots, P_{\text{DER},M}] \\
& \text{Augment slowly-varying DER injections and tap positions} \\[1em]
\text{Prediction (quasi-steady):} \\
& x_{k+1|k} = x_{k|k} + w_k \quad \text{[process noise for load/DER changes]} \\[1em]
\text{Update when AMI data arrives:} \\
& K_k = P_{k|k-1} H^T (H P_{k|k-1} H^T + R)^{-1} \\
& x_{k|k} = x_{k|k-1} + K_k (z_k - h(x_{k|k-1})) \\
& P_{k|k} = (I - K_k H) P_{k|k-1} \\[1em]
& \text{Jacobian } H = \frac{\partial h}{\partial x} \text{ (sensitivity of measurements to states)}
\end{aligned}$$
3. Semidefinite Programming (SDP) Relaxation
For globally optimal solutions when WLS gets stuck in local minima:
- Reformulate power flow as convex SDP problem (OPF relaxation)
- Solve via interior-point methods (MOSEK, SeDuMi)
- Check rank-1 solution for exactness
- Computationally expensive (10–60 seconds), used for planning studies
4. Bayesian Estimation with MCMC
For uncertainty quantification in parameter estimation:
- Define prior distributions on line impedances, DER capacities
- Sample posterior p(θ | z) via Markov Chain Monte Carlo (Metropolis-Hastings)
- Obtain credible intervals for hosting capacity analysis
Observability and Sensor Placement
Key results from observability analysis:
Minimum sensor requirements:
- Feeder head: Must measure V, P, Q at substation (mandatory for power balance)
- Voltage measurements: Need at least 1 measurement per 5–10 buses for acceptable error (<2% voltage error)
- μPMU placement: Place at mid-feeder and end-of-line to capture voltage drops and detect high-impedance faults
- Topology observability: Require at least 1 measurement per switch zone to infer topology changes
- DER visibility: Solar inverters with IEEE 1547 telemetry improve state estimation accuracy by 40–60%
Optimal sensor placement: Greedy algorithm maximizes reduction in voltage estimation covariance per sensor installed. Typical result: 5–8 strategically placed μPMUs reduce average voltage error from 3.2% to 0.8% on IEEE 123-bus feeder.
Validation Strategy
| Validation Test |
Method |
Acceptance Criterion |
| Voltage estimation error |
Compare estimated V to true V (ground truth from load flow) |
RMSE < 0.01 pu (1%) |
| Power flow estimation |
Validate branch flows Pij, Qij at instrumented lines |
MAPE < 10% |
| Bad data detection |
Inject 5% gross errors (meter failures), test detection rate |
Detection rate > 95% |
| Topology change detection |
Simulate switch operation, measure time to convergence |
Detect within 2 time steps |
| Computational speed |
Solve 500-bus feeder on edge computing hardware |
< 1 second per snapshot |
Utility Impact
Mechanistic state estimation enables operational improvements:
- Volt-Var Optimization (VVO): Use real-time voltage profiles to coordinate capacitor banks and regulators, reducing losses by 2–5% ($50K–200K annually per substation)
- Hosting Capacity: Accurate impedance estimates improve solar/EV interconnection limits by 15–30% (enables more DER without voltage violations)
- Fault Location: Rapid voltage/current anomaly detection reduces outage duration by 20–40% (SAIDI improvement)
- Asset Health Monitoring: Track transformer loading and line losses to prioritize replacements (avoid $500K+ emergency failures)
- Grid Modernization Planning: Identify circuits requiring reinforcement or AMI sensor upgrades based on estimation uncertainty
Case study: A California utility deployed distribution state estimation across 50 feeders with 80% AMI coverage and 3 μPMUs per feeder. VVO optimization saved 4.2 GWh annually ($420K), and improved voltage regulation reduced customer complaints by 65%.
Try ProcessLM
ProcessLM streamlines grid modeling: describe your feeder topology and available measurements, and it generates the power flow model, configures the state estimator, and validates observability—no PowerFactory or GridLAB-D scripting required.
Request Early Access