The field of power system optimization and analysis is moving towards the development of innovative methods that combine physical constraints with machine learning techniques. This direction is driven by the need for more efficient and accurate solutions to complex optimization problems, such as optimal power flow and unit commitment. Researchers are exploring new frameworks that integrate physical laws and constraints into neural network architectures, allowing for end-to-end physics-constrained learning and improving the accuracy and reliability of optimization results. Another area of focus is the development of hybrid methods that combine traditional optimization techniques with machine learning approaches, enabling the solution of large-scale optimization problems with high accuracy and efficiency. Noteworthy papers in this area include:

• A paper that proposes a neural physics power flow solving method based on manifold geometry and gradient flow, which achieves true end-to-end physics-constrained learning.
• A paper that presents a hybrid sequential convex programming framework for solving unbalanced three-phase AC optimal power flow problems, demonstrating high numerical accuracy and computational efficiency.
• A paper that proposes a method to optimize the aggregate flexibility region of distributed energy resources through network reconfiguration, achieving substantial improvements in flexibility.
• A paper that develops a framework for counterfactual explanations in power system optimization, providing transparent and fair operation of decision-making software.