Optimal Power Flow (OPF)

The core goal of OPF is to minimize the total operational cost of generation. The objective function typically models the fuel cost curves of thermal generators as quadratic functions. Modern formulations also incorporate renewable curtailment penalties and emission costs to reflect environmental policy. The solver iteratively adjusts active power outputs (P) and voltage magnitudes (V) to find the global minimum of this cost surface.

OPF enforces the fundamental physics of Kirchhoff's Current Law at every bus. The sum of power injected must equal the sum of power consumed plus losses. This is expressed as a set of non-linear AC power flow equations:

• Active Power (P): Generation = Load + Losses
• Reactive Power (Q): Generation = Load + Losses These constraints ensure the solution is physically realizable and the grid remains in steady-state equilibrium.

OPF ensures the solution is secure by enforcing hard physical limits:

• Thermal Limits: Line flows (MVA) must stay below ratings to prevent sagging or annealing.
• Voltage Bounds: Bus voltages must remain within ANSI C84.1 limits (typically ±5% of nominal).
• Generation Caps: Active and reactive power output of each generator is bounded by its capability curve. These constraints transform the problem from a simple economic dispatch into a security-constrained optimization.

OPF exists in two primary fidelity levels:

• AC-OPF: The full non-linear, non-convex problem. It accurately models voltage magnitudes, reactive power, and losses. It is computationally heavy and solved using interior-point methods.
• DC-OPF: A linearized approximation that ignores reactive power and voltage magnitudes, assuming all voltages are 1.0 pu and angle differences are small. It solves extremely fast using linear programming, making it ideal for real-time electricity markets where speed is critical.

A critical byproduct of the OPF solution is the Locational Marginal Price. The LMP at a bus is the shadow price of the power balance constraint—the marginal cost of supplying one additional megawatt of load at that specific location. LMP decomposes into three components:

• Energy Component: System-wide marginal fuel cost.
• Congestion Component: Cost of binding transmission constraints.
• Loss Component: Marginal cost of resistive losses. LMPs form the basis of modern wholesale electricity markets.

OPF handles contingencies through two philosophies:

• Preventive OPF: The dispatch is constrained so that if any single contingency (N-1) occurs, the system remains within limits without any post-fault control action. This is conservative and robust.
• Corrective OPF (SCOPF): The dispatch allows for short-term post-contingency violations that can be resolved by fast corrective actions (e.g., generator ramping, phase shifter adjustments) within a defined time window. This yields lower operational costs.

An extension of OPF that incorporates N-1 contingency constraints to ensure the system remains stable following the unplanned loss of any single element. While standard OPF minimizes cost under normal conditions, SCOPF co-optimizes the base case and post-contingency states simultaneously.

• Prevents cascading failures by enforcing thermal limits after generator or line outages
• Computationally intensive due to the exponential growth of contingency scenarios
• Often solved using Benders decomposition or iterative contingency filtering

An advanced control methodology that solves a finite-horizon OPF problem at each time step, applying only the first control action before re-optimizing. This receding horizon approach bridges the gap between static OPF snapshots and real-time dynamic grid control.

• Uses a dynamic system model to anticipate future states and enforce voltage constraints
• Inherently handles forecast errors in renewable generation through continuous feedback
• Applied in Volt-VAR control and microgrid frequency regulation

A distributed convex optimization algorithm that decomposes a large-scale OPF problem into smaller regional subproblems solved in parallel. ADMM enables privacy-preserving coordination between utility control centers without sharing sensitive network data.

• Each region solves its local OPF independently, exchanging only boundary voltage information
• Converges to the globally optimal solution under convexity assumptions
• Foundational for transactive energy and multi-area grid coordination

A deep learning paradigm that embeds the power flow equations directly into the neural network loss function, ensuring predictions obey Kirchhoff's laws. PINNs offer a differentiable surrogate for traditional OPF solvers, enabling gradient-based optimization at scale.

• Eliminates the need for labeled training data by using physical residuals as supervision
• Accelerates OPF convergence by providing warm-start initializations for Newton-Raphson solvers
• Naturally handles non-convex AC power flow constraints that challenge conventional methods

An optimization framework that explicitly incorporates the probability distributions of uncertain variables—such as wind speed and solar irradiance—into the OPF formulation. Unlike deterministic OPF, stochastic programming finds dispatch decisions robust across multiple future scenarios.

• Represents uncertainty through a discrete scenario tree generated via Monte Carlo sampling
• Minimizes expected operational cost rather than a single deterministic outcome
• Often coupled with Conditional Value at Risk (CVaR) to penalize extreme tail events

A deep learning architecture designed to operate directly on graph-structured data, making it inherently suited for modeling the arbitrary topology of electrical networks. GNNs learn optimal generator dispatch by treating buses as nodes and transmission lines as edges.

• Captures topological dependencies that convolutional or fully-connected networks miss
• Enables end-to-end OPF learning that generalizes across different grid configurations
• Used for real-time feasibility prediction and constraint screening before invoking full OPF solvers