Machine learning techniques for accurate prediction and improved matching of renewable energy production, storage, and consumption

Abstract

In recent decades, the electricity industry has undergone considerable transformation, characterized by the expansion of competitive electricity markets, integration of renewable energy sources, and the deployment of digital systems. These developments have contributed to increased complexity and variability within power systems, highlighting the need for more accurate, adaptive, and intelligent methods for managing electricity generation, storage, and consumption. This thesis focuses on addressing these challenges, with a particular emphasis on advancing New Zealand’s renewable energy ambitions through the application of machine learning (ML) and digital twin technologies for optimizing renewable energy forecasting and management in both industrial and residential contexts. In the industrial domain, the research develops real-time solar power monitoring systems and predictive models employing convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory (LSTM) networks. These models are designed to enhance forecasting precision and system efficiency by accounting for the inherent variability in renewable energy generation. Complementing this, fuzzy logic-based control strategies are implemented to manage electricity consumption adaptively in industrial processes, including applications within the meat processing industry. The incorporation of digital twin technologies further facilitates real-time monitoring, predictive maintenance, and agile response to evolving operational conditions, thereby supporting system resilience and efficiency. In residential and community energy systems, the thesis introduces an innovative adaptive energy management framework combining streaming machine learning (SML) with a fractal-structured microgrid architecture. This framework utilizes incremental learning algorithms and digital twin principles to enable real-time adaptive control of energy flows, local trading of electricity, and dynamic adjustments in response to fluctuating supply and demand. The approach aims to enhance battery management, support local energy autonomy, and potentially reduce dependency on centralized grids. Comparative evaluations of centralized, distributed, and hybrid battery storage configurations provide insights into the trade-offs between grid interaction, storage longevity, and operational flexibility, offering practical solutions for optimizing residential microgrids. In conclusion, this thesis explores a combined, case-study based approach to enhancing renewable energy utilization through the combined application of advanced machine learning and digital twin technologies. By addressing both industrial and residential energy systems, the research contributes to the broader vision of developing reliable, efficient, and sustainable power infrastructures. The findings align with New Zealand’s commitment to reducing carbon emissions and advancing renewable energy technologies, offering case-study evidence and potential approaches to support the transition toward a low-carbon energy future.