Distributed Real-Time Control Schemes for Microgrids Considering Uncertain Renewable Sources With Provisions for Resilient Operation

Abstract

This thesis presents the integration and improvement of different technologies for optimal operation of microgrids, where the objective is to minimise cost of supply and maximise their renewable hosting capacity in a reliable and seamless manner. To this end, this work has proposed a hierarchical control framework for coordinating energy transactions between different stakeholders within the microgrid, and different microgrid clusters with distributed renewable resources integrated, in an economical and reliable manner.

The stability of the proposed primary control layer, and corresponding voltage bus, has been investigated using Lyapunov’s direct method to ensure it remains stable under all operating conditions. Meanwhile, the resiliency of the underlying communication network to maintain a fully connected network for the distributed control nodes has been verified mathematically using Graph theory by ensuring the underlying Laplacian matrix for the network always remains singular and without repeated zero eigenvalues, even following failures in any one individual communication node. This is done to ensure that the proposed control system is capable of reliably maintaining electrical service under normal and faulty conditions.

The hierarchical control framework relies on implementation of -based techniques, mainly using Multi Agent Systems implemented within the Java Agent Development Framework environment to simulate a distributed control architecture for achieving real-time control of multiple resources within the microgrid (and clusters of microgrids). Meanwhile, using , accurate forecast models have been developed for both electricity price forecast and wind speed (for distributed renewable resources) within the control architecture for optimising energy transactions between relevant stakeholders within the microgrid.

To verify the accuracy and compare the results of ML-based forecast models, this thesis proposes the use of statistical methods to ensure any differences between different forecast models are due to the model structure rather than randomness in the data. This comparison is carried out in addition to usual statistical comparison of different forecast models in terms of total absolute error, squared error and total cost.

At the hardware level, the hierarchical control system has been implemented using a connected network of Raspberry Pi computers acting as individual distributed control nodes with the microgrid simulated in a real-time simulator environment realised by an OPAL-RT (series OP5700) real-time simulator to test the viability of the control system to respond to a microgrid cluster system in real-time.