Virtual inertia provision from distribution power systems using machine learning

This chapter focuses on the application of physics-informed neural networks (PINNs), Bayesian physics-informed neural networks (BPINNs), and reinforcement learning (RL) in an inertia provision framework from inverter-based resources in power distribution systems. Conventional generation with synchronous generators is disconnected from the grid to reduce carbon emissions, and thus the synchronous inertia decreases. This is critical for the stability of the system since the generators’ inertia limits the change in system frequency. To compensate for this lack of inertia, inverter-based resources have to mimic this behavior through advanced inverter control. Their dispersion throughout the grid brings new challenges like provision coordination to avoid congestion and handle bidirectional power flow. Thus, in this chapter, we propose an inertia support framework, which utilizes artificial intelligence (AI). First, this chapter revisits the concept of virtual inertia (VI) and possible sources. Then, the inertia support framework for coordinated provision of VI from the power distribution system is detailed. The chapter proceeds with the investigation of PINNs and BPINNs for estimation and modeling of the present inertia, which is purely data driven and presents case studies. Lastly, the chapter discusses a model-free RL coordination function for VI provision from distribution systems. In the results we find benefits in the utilization of these approaches through short computation times, robustness, and reliance solely on measurement data, that is, model-free operation.

Options

Virtual inertia provision from distribution power systems using machine learning