An adjustable Predictive&Prescriptive method for the RO-based optimal power flow problem.

Area of Science:

• Electrical Engineering
• Operations Research
• Data Science

Background:

• Traditional optimal power flow considering uncertainty (OPF-U) relies on predicted values, but the impact of prediction errors on prescriptive analytics is unclear.
• Existing methods for OPF-U often use statistical or machine learning for prediction, followed by robust optimization (RO).

Purpose of the Study:

• To propose an adjustable framework combining machine learning and RO for the OPF-U problem that accounts for prediction errors.
• To develop a robust fluctuation region with adjustable parameters for improved uncertainty handling in OPF-U.

Main Methods:

• Utilized k-nearest neighbor (k-NN) to identify samples around predicted uncertainty values.
• Constructed a Minimum Volume Ellipsoid (MVE) set (KMV set) from k-NN samples.
• Developed an adjustable robust fluctuation region using a two-term exponential formula from the KMV set.
• Embedded the fluctuation region into a two-stage RO model for solving the OPF-U problem.

Main Results:

• The proposed fluctuation region demonstrated superior robustness and adjustability compared to state-of-the-art box and ellipsoidal sets.
• The two-stage RO model yielded more economical solutions than existing RO models.
• Out-of-sample simulations confirmed reduced computational burden for larger systems with the proposed adjustable Predictive&Prescriptive method.

Conclusions:

• The proposed adjustable framework effectively integrates predictive and prescriptive analytics for OPF-U.
• The method provides a more robust, economical, and computationally efficient approach to handling uncertainty in power flow optimization.