Logo
Project • Stochastic OPF Paper
Back
EC500 Controls & Optimization of Power Systems • Research

Distributionally Robust Stochastic & Probabilistic DC-OPF

A chance-constrained DC Optimal Power Flow study reformulating stochastic OPF as a convex program. I modeled non-Gaussian load uncertainty using Polynomial Chaos Expansion (PCE) and compared risk, feasibility, and cost behavior across multiple uncertainty families.

Chance Constraints PCE SOCP / Convex Power Systems MATLAB / CVX
OPF paper preview

Focus

Robust reliability under uncertainty via chance constraints + PCE.

Overview

This paper explores chance-constrained DC Optimal Power Flow (DC-OPF) under uncertain load demand. Rather than assuming the uncertainty is Gaussian, I studied multiple distributions and built deterministic convex reformulations that still guarantee probabilistic reliability on generator limits and line flows.

Problem

Real demand uncertainty is often bounded or skewed. Gaussian assumptions can underestimate tail risk or misrepresent feasibility when constraints are tight.

Approach

Use Polynomial Chaos Expansion (PCE) with distribution-matched bases (e.g., Legendre for Uniform, Laguerre for Gamma) plus chance constraints to obtain deterministic convex programs.

Methods

The workflow combines uncertainty modeling, PCE coefficient propagation, and a convex optimization layer. A key idea is representing uncertain quantities with orthogonal polynomial bases so constraint probabilities can be bounded or converted into deterministic inequalities.

Uncertainty modeling

• Load modeled as random variable(s) across different distributions

• Canonical / non-canonical PCE depending on the uncertainty family

• Coefficient interpretation: mean/variance and higher-order behavior

Optimization layer

• Chance constraints on generation and line flows

• Deterministic reformulations (moment/CDF based depending on distribution)

• Solved as convex program (SOCP-style constraints in CVX)

Results

Comparing distributions highlights how “risk” changes operational cost and constraint activity. Stricter chance constraints (smaller ε) push solutions toward safer dispatch and may introduce feasibility cliffs depending on uncertainty magnitude and network congestion.

Key observations

• Risk parameter ε trades off cost vs reliability

• Bounded/skewed uncertainty can shift constraint activation patterns

• Larger uncertainty can create infeasibility or sharp cost increases

Tools used

• MATLAB for modeling + experiments

• CVX for convex optimization

• PCE basis matching (Legendre / Laguerre)

Final Paper

You can scroll the PDF directly below. If your browser blocks embedded PDFs, use the “Open PDF” button.

Reflection

Challenges

• Translating probabilistic constraints into usable deterministic forms

• Debugging feasibility issues when ε became too strict

• Matching PCE bases to uncertainty families and validating assumptions

• Keeping the pipeline stable as uncertainty magnitude increased

What I learned

• How chance constraints shape dispatch decisions under uncertainty

• How PCE coefficients map to moments and constraint conservatism

• Convex modeling discipline: “make it solvable” without losing meaning

• Stronger intuition for OPF constraints and congestion behavior