Distributed Optimal Power Flow with Data-Driven Sensitivity Computation

On account of the increasing influx of distributed energy resources into modern power grids, it is essential to develop efficient distributed control and optimization algorithms capable of providing suitable solutions with access to local data alone. This paper uses a distributed optimal power flow (OPF) algorithm based on a gradient projection method, which applies to any arbitrary grid topology, to solve the OPF problem. A multi-variable linear regression method learns the network sensitivities with historical operational data. The use of a data-driven approach avoids the requirement of accurate information on line parameters and network topology. Additionally, introduced curtailment cost factors into the objective cost function encourage the usage of renewable power sources. In conclusion, we show that the solution achieved using data-driven sensitivities provides an average optimality gap of 1.8% to the centralized OPF solution with numerical test results on a modified IEEE 69 bus system.