Applications of AI in distribution system operation

Abstract

The increasing expansion of renewable energies is leading to a reduction in the inertia of the electrical energy grid, as these are usually connected via power electronics. Unlike conventional synchronous generators, these do not offer inherent inertia in changes of active power. Furthermore, many renewables are connected at distribution grid level, which causes bidirectional power flows.

Throughout this project, two tools were developed to ensure secure grid operation and enable the comprehensive utilization of renewables in the distribution grid to support the grid frequency. The use of physics-informed machine learning approaches has proven to be beneficial, as they utilize the increasing flow of data from measurement devices. The first tool estimates the inertia of the grid at transmission grid level in a distributed manner. It enables reliable estimates for inverter-dominated grids and also indicates the trustworthiness of its own estimates. The second tool coordinates the provision of inertial frequency support from the distribution grid using physics-informed reinforcement learning. Unlike most coordination approaches, the developed tool does not require an accurate model of the distribution grid, which enables fast adaptation to new systems based on data.

Tools

The tools are develop in python relying on commonly used machine learning packages.

Contact

Simon Stock

Duration

01.07.2020 to 31.07.2024