Learning-assisted optimization for transmission switching

Area of Science:

• Optimization
• Machine Learning
• Power Systems Engineering

Background:

• The Direct Current Optimal Transmission Switching (DC-OTS) problem is crucial for optimizing power grid operations and maintaining generation-demand balance, especially with increasing grid variability.
• DC-OTS is a complex, NP-hard mixed-integer programming problem due to binary variables determining transmission line status.

Purpose of the Study:

• To propose a novel machine learning-based procedure to efficiently solve the computationally difficult DC-OTS problem.
• To leverage historical problem instance solutions to accelerate the optimization of new, unseen DC-OTS models.

Main Methods:

• Development of a learning procedure that utilizes solutions from previous DC-OTS instances.
• Application of the proposed method to a real-life power system dataset for numerical experimentation.

Main Results:

• The proposed approach demonstrates a very high success rate in identifying optimal power grid topologies.
• Significant speed-up factors were achieved compared to alternative heuristic methods.
• While optimality guarantees are not provided, the method offers practical efficiency.

Conclusions:

• Machine learning offers a promising avenue for addressing complex optimization challenges in power systems.
• The developed learning procedure effectively accelerates the solution of the DC-OTS problem, enhancing operational efficiency and cost-effectiveness in power grids.