Comparative Study of Optimisation Algorithms for Cloud Electric Vehicle-to-Grid Battery Operations in Microgrids With High Penetration of Solar Photovoltaics

Abstract

This paper proposes optimisation techniques for the operations of cloud electric vehicle-to-grid battery (CEVB) in microgrids with high penetration of solar photovoltaics (PVs). It thoroughly scrutinises popular methods such as gradient descent (GD), as well as three heuristic methods, including pattern search (PS), particle swarm optimisation, and genetic algorithms. Addressing the limitations of these methods, such as local optimality and constraint violations, is achieved through intensive experimentation, utilising stochastic initialisation and a hybrid heuristic-GD multiple-run strategy. These experiments also investigate the effects of heuristic algorithm parameter settings on the optimisation results, identify optimal parameters for each heuristic-GD method, and assess their effectiveness in handling uncertainties in CEVB operational model inputs (solar irradiance, electricity price, and electric vehicle [EV] power demand). Evaluations conducted using actual operational data from an EV charging station in South Australia demonstrate that all proposed methods can achieve global optimal results in fewer than 100 runs with appropriate parameter settings. Scalability tests were conducted to validate the method's feasibility for larger systems, offering valuable insights into computation times as the number of CEVBs grows. The proposed methods demonstrate robustness in addressing uncertainties in electricity prices and EV power demand, ensuring reliable and adaptable performance across various scenarios.