A Trustable Data-Driven Optimal Power Flow Computational Method With Robust Generalization Ability.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Optimization

Background:

• Data-driven optimal power flow (OPF) methods are a recent research focus.
• Current methods struggle with out-of-distribution (OOD) samples, leading to inaccurate and untrustable solutions.
• Assessing the reliability of data-driven OPF solutions is challenging.

Purpose of the Study:

• To improve the generalization ability and trustworthiness of data-driven OPF approaches.
• To address the limitations of current methods in handling OOD data.
• To develop a reliable method for judging the adaptability of data-driven solutions.

Main Methods:

• Embedding inherent patterns of OPF solutions (e.g., load balance constraints) into the data-driven learning process.
• Proposing an adaptability judgment method using a decoder neural network to assess solution trustability.
• Evaluating the method's performance on various power systems.

Main Results:

• The proposed method significantly improves the calculation accuracy for OOD data by an average of 30.19% compared to state-of-the-art techniques.
• The adaptability judgment method enables the data-driven approach to achieve over 98% accuracy on OOD data.
• Other methods demonstrated accuracies ranging from 34.08% to 94.50% on the same OOD test data.

Conclusions:

• The integration of inherent OPF solution patterns enhances data-driven method generalization.
• The proposed adaptability judgment method provides a reliable way to assess the trustability of data-driven OPF solutions.
• This research offers a more accurate and dependable data-driven approach for OPF problems, especially in complex grid scenarios.