Monte Carlo Simulation

The core mechanism relies on generating a large number of pseudo-random or quasi-random samples from predefined input probability distributions. For a grid study, this means drawing thousands of scenarios for wind speed (Weibull), solar irradiance (Beta), and load (Gaussian) to represent possible system states. The accuracy depends on the Law of Large Numbers, where the sample mean converges to the true expected value as the number of trials increases.

For each random sample of inputs, a deterministic power flow calculation is executed. This acts as a black-box function: Y = f(X). The simulator solves the non-linear algebraic equations for bus voltages and line currents exactly as it would for a single snapshot, but it does so repeatedly for every scenario. This non-intrusive nature means the simulation code requires no modification to handle uncertainty.

The raw results from thousands of deterministic runs are aggregated into empirical probability density functions (PDFs) and cumulative distribution functions (CDFs). This yields actionable metrics for grid planners:

• Mean and variance of voltage magnitudes at critical buses.
• Probability of violation: P(V < 0.95 p.u.).
• Conditional Value at Risk (CVaR) for thermal overloads on transmission lines.

A critical characteristic is the need to verify that the simulation has run for a sufficient number of iterations. Techniques include monitoring the stabilization of the mean and standard error of key outputs. A common heuristic stops the simulation when the change in the estimated 95th percentile of a line flow drops below a specified tolerance, ensuring computational resources are not wasted while guaranteeing statistical significance.

To accelerate convergence without increasing sample size, advanced implementations use variance reduction:

• Importance Sampling: Biases sampling toward rare failure regions (e.g., extreme load drops) and reweights outputs to correct the bias.
• Stratified Sampling (Latin Hypercube): Divides the probability space into intervals to prevent sample clustering, ensuring the full range of each random variable is explored efficiently.

Standard Monte Carlo assumes independent, identically distributed (i.i.d.) samples, suitable for snapshot risk assessment. For time-sequential studies (e.g., storage dispatch), Markov Chain Monte Carlo (MCMC) or sequential sampling from ARIMA forecast error models is required. This captures the temporal correlation where a low wind hour is statistically more likely to be followed by another low wind hour.

A stratified sampling method that divides each input distribution's cumulative density function into N equal-probability intervals. By sampling exactly once from each interval, LHS enforces full coverage of the input space. This prevents the clustering artifacts common in simple random sampling, often reducing the number of simulations required by a factor of 10 to 100 while maintaining statistical accuracy in probabilistic power flow studies.

A deterministic alternative to pseudo-random number generation that uses low-discrepancy sequences—such as Sobol, Halton, or Faure sequences. Unlike standard Monte Carlo, which converges at a rate of O(1/√N), QMC methods can approach O(1/N) for smooth integrands. In grid applications, QMC provides faster convergence when estimating the statistical moments of bus voltages and line flows under renewable uncertainty.

A variance reduction technique that concentrates computational effort on rare but critical regions of the input space. Instead of sampling from the original distribution, samples are drawn from a biased proposal distribution that over-represents failure events—such as voltage violations or line overloads—and then reweighted to correct the bias. This makes it practical to estimate extremely small probabilities (e.g., 10⁻⁶) without billions of simulations.

A class of algorithms that construct a Markov chain whose stationary distribution equals the target posterior. The Metropolis-Hastings and Gibbs sampling algorithms are foundational examples. In grid state estimation, MCMC enables Bayesian inference over network parameters when closed-form solutions are intractable, allowing engineers to quantify uncertainty in topology identification and bad data detection.

A spectral method that represents a stochastic system's response as a truncated series of orthogonal polynomials in the random input variables. For a given set of orthogonal basis functions—Hermite for Gaussian inputs, Legendre for uniform—the coefficients are computed via stochastic collocation or Galerkin projection. Once constructed, the PCE surrogate yields output statistics (mean, variance, Sobol indices) analytically, without further sampling.

An efficient rare event simulation technique that decomposes a small failure probability into a product of larger conditional probabilities. Starting from the nominal distribution, the algorithm progressively samples towards the failure region by defining intermediate threshold events. Each successive level uses Markov chain Monte Carlo to generate samples conditional on exceeding the previous threshold, making it ideal for estimating Loss of Load Probability and extreme voltage excursion risks.