Stochastic Differential Equation (SDE)

Every SDE decomposes system dynamics into two fundamental components. The drift term captures the deterministic trend—such as the expected diurnal increase in solar irradiance or the scheduled ramp-up of industrial load. The diffusion term, driven by a Wiener process, injects continuous random fluctuations representing forecast errors and small-scale turbulence. This decomposition allows grid planners to model the mean trajectory of a state variable while simultaneously quantifying the continuous uncertainty band around that trajectory.

SDEs require a specialized calculus because standard Riemann integration fails when integrating with respect to a Wiener process, which has unbounded variation. Itô calculus provides the rigorous mathematical framework, with the Itô integral defined as the limit of non-anticipating step functions. A critical consequence is Itô's Lemma, which acts as the stochastic chain rule, revealing that the derivative of a function of an SDE state includes an extra second-order term absent in ordinary calculus. This is essential for correctly transforming wind speed models into power output distributions.

Many physical quantities in power systems exhibit mean-reversion, naturally modeled by the Ornstein-Uhlenbeck (OU) process. Unlike a pure random walk that can drift to infinity, an OU process exerts a restoring force proportional to the deviation from a long-term mean. This makes it ideal for modeling:

• Wind speed fluctuations around a forecasted mean
• Load deviations that revert to a typical consumption profile
• Electricity spot prices that oscillate around a seasonal average The speed of reversion parameter directly controls how quickly shocks dissipate.

For variables that must remain strictly positive and exhibit multiplicative noise, Geometric Brownian Motion (GBM) is the canonical SDE model. The drift and diffusion scale proportionally with the current state value, ensuring non-negativity. In grid applications, GBM is used to model:

• Long-term load growth trajectories for capacity planning
• Distributed solar adoption rates over multi-year horizons
• Fuel price uncertainty for generation cost modeling The log-normal distribution of GBM states provides a realistic right-skewed profile for these inherently positive quantities.

Analytical solutions exist for only a limited class of SDEs, necessitating numerical simulation. The Euler-Maruyama method extends the deterministic Euler scheme by adding a scaled Gaussian increment at each time step, achieving strong convergence of order 0.5. For systems requiring higher accuracy, the Milstein scheme adds a correction term involving the derivative of the diffusion coefficient, achieving strong order 1.0 convergence. Grid operators use these discretizations to generate thousands of synthetic trajectories for Monte Carlo-based probabilistic power flow analysis.

SDEs capture the continuous random jitter in electrical load that deterministic forecasts miss. A typical model combines a periodic drift representing daily and seasonal cycles with a state-dependent diffusion that increases with load magnitude. This structure generates realistic intra-hour fluctuations that are critical for:

• Regulation reserve sizing to handle minute-to-minute variability
• Frequency stability studies requiring high-resolution load inputs
• Probabilistic voltage profile assessment along distribution feeders The continuous nature of SDEs avoids the artificial discontinuities introduced by discrete-time noise models.

The fundamental continuous-time stochastic process driving SDEs. A Wiener process W(t) has three defining properties: it starts at zero, has independent increments, and each increment W(t) - W(s) follows a normal distribution with mean 0 and variance t-s. In power systems, this models the continuous random fluctuations in renewable generation output and load behavior. The non-differentiable nature of its sample paths necessitates Itô calculus rather than standard calculus for solving SDEs.

A stochastic integral defining the integral of a process with respect to a Wiener process. Unlike the Riemann-Stieltjes integral, the Itô integral evaluates the integrand at the left endpoint of each partition interval, making it a martingale. This non-anticipating property is crucial for financial and power system modeling where future randomness cannot be known. The resulting Itô isometry provides the variance of the integral, essential for quantifying uncertainty propagation in grid simulations.

The stochastic calculus analog of the chain rule. For a function f(X(t), t) where X(t) follows an SDE, Itô's lemma states that df includes an extra second-order term (1/2)(∂²f/∂x²)σ² dt absent in standard calculus. This correction arises because (dW)² = dt in mean square. In power systems, this lemma is essential for transforming SDEs governing raw wind speed into SDEs for power output, correctly accounting for the nonlinear power curve.

A mean-reverting SDE of the form dX(t) = θ(μ - X(t))dt + σ dW(t), where θ controls the speed of reversion to the long-term mean μ. This process is widely used to model electricity spot prices and wind speed variations because it captures the tendency of these quantities to revert to average levels rather than drift arbitrarily. Unlike a pure Wiener process, the Ornstein-Uhlenbeck process has a stationary distribution—a Gaussian with variance σ²/(2θ).

The simplest numerical scheme for approximating SDE solutions, extending the Euler method for ODEs. Given an SDE dX = a(X)dt + b(X)dW, the discretization is X(t+Δt) = X(t) + a(X)Δt + b(X)ΔW, where ΔW ~ N(0, Δt). This method achieves strong convergence order 0.5 and weak order 1.0. In probabilistic power flow, this enables Monte Carlo path generation for load and generation trajectories when analytical solutions are unavailable.

A deterministic partial differential equation governing the time evolution of the probability density function p(x,t) of the solution to an SDE. For an SDE with drift a(x) and diffusion b(x), the Fokker-Planck equation is ∂p/∂t = -∂(ap)/∂x + (1/2)∂²(b²p)/∂x². Solving this PDE provides the complete statistical description of grid state variables over time without Monte Carlo sampling, though it becomes computationally intensive in high dimensions.