Network State-Based Algorithm Selection for Power Flow Management Using Machine Learning

This paper demonstrates that machine learning can be used to create effective algorithm selectors that select between power system control algorithms depending on the state of a network, achieving better performance than always using the same algorithm for every state. Also presented is a novel method for creating algorithm selectors that consider two objectives. The method is used to develop algorithm selectors for power flow management algorithms on versions of the IEEE 14- and 57-bus networks, and a network derived from a real distribution network. The selectors choose from within a diverse set of power flow management algorithms, including those based on constraint satisfaction, optimal power flow, power flow sensitivity factors, and linear programming. The network state-based algorithm selectors offer performance benefits over always using the same power flow management algorithm for every state, in terms of minimizing the number of overloads while also minimizing the curtailment applied to generators.