Geração de perfis de carga sintéticas em redes elétricas inteligentes: uma abordagem baseada em dados com redes generativas

Abstract

Developing sustainable and economically viable methods has become fundamental for improving energy efficiency policies, reducing costs, and mitigating environmental impacts. This has occurred due to factors such as the energy transition, the digitalization of electrical systems, and the diffusion of distributed energy resources (DERs), which have been transforming the paradigm of the electrical sector. In this context, data analysis from electrical networks emerges as a strategy to identify improvement opportunities and optimize resources allocated to these networks. In this way, data analysis has become fundamental in the operation and planning of electrical networks and the entire electrical power system. Based on this perspective, throughout the research conducted in this doctoral thesis, challenges were identified that impact the more detailed analysis of electrical network dynamics, such as the scarcity of available data, security issues, and data privacy, limiting the diagnostic analysis of networks and the development of applications aimed at energy efficiency and sustainability. To mitigate such factors, synthetic data emerges to fill these gaps. Given this, this thesis proposed to investigate the application of data analysis in an electrical energy distribution system on a university campus, aiming to generate synthetic profiles of load curves. Within the scope of this perspective, the Non-linear Independent Components Estimation (NICE) model was adopted with the inclusion of neural networks with convolutional layers. As a result of the methodology adopted, the representation of synthetic profiles was obtained. This approach allows a representation of the electrical consumption patterns of a university campus, making it possible to evaluate the potential of generative models in characterizing and quantifying load profiles. Therefore, the contribution of this thesis lies in the application of data analysis techniques and the implementation of generative models, such as NICE with convolutional layers. These approaches aim to generate synthetic profiles of load curves from data measured in a university electrical grid, acting as an environment for testing and validating the proposed methodology.