Abstract
Energy systems are undergoing rapid transformations, driven by the integration of renewable energy sources, energy storage, and micro-grids. The challenges of this increasing complexity, such as uncertainties or time-varying conditions, demand advanced modeling techniques. In response, our research introduces the advanced Heterogeneous Recurrence Analysis to navigate intricate nonlinear dynamics inherent in energy systems. Recurrence Analysis (RA) and Recurrence Network (RN) techniques have gained widespread use in examining the recurrence properties of complex dynamical systems in phase space. However, most existing recurrence analyses focus only on single-state, homogeneous recurrence, which often makes the discovery of system recurrences not fully exploited. While modern Machine Learning (ML) models such as Convolutional Neural Networks and Recurrent Neural Networks offer powerful tools for analyzing complex systems, they often lack in-depth interpretability. Our approach in integrating RA and RN with ML not only broadens the research horizon but also reveals deeper insights into the system's dynamics.
In our study, we first proposed a new Multi-scale Heterogeneous Recurrence Analysis (M-HRA) framework to address system recurrences of multi-scale transitions on an optimized Iterated Function System (IFS). Secondly, we created a novel Heterogeneous Recurrence Network (HRN) structure. We later combined our newly developed M-HRA and HRN with existing ML methods to tackle several challenges facing modern energy systems, including Urban Load and Renewable Energy Prediction, and Wind Turbine Maintenance.
In conclusion, our research stands as a testament to the potential of integrating recurrence analysis, network methodologies, and machine learning in understanding and addressing challenges in complex energy systems. The contributions of this study lie not only in the development of novel methodologies but also in their practical application, setting a benchmark for future endeavors in the realm of energy system analysis and forecasting.