PhD thesis defense to be held on December 19, 2024, at 14:30 (virtually)
Picture Credit: Sotiris Pelekis
Thesis title: Trustworthy artificial intelligence in smart energy grids: Applications on load forecasting and demand response
Abstract: Artificial intelligence (AI), among other technologies such as big data, internet of things (IoT), and cloud computing are increasingly being used in the energy sector and specifically electrical power and energy systems (EPES), gradually shifting the contemporary research and industry landscape towards the smart grid paradigm. In this context, machine learning (ML) and deep learning (DL) have been revolutionizing the power grid industry with innovative applications ranging from electricity demand forecasting, load balancing and stability control to electricity theft detection, security, data management, grid analytics, and demand side management. The current dissertation deep dives into several applications of ML and DL in the energy domain as well as the smart grid including short-term load forecasting (STLF), demand response (DR). STLF is vital for the effective and economic operation of EPES and energy markets. However, the non-linearity and non-stationarity of electricity demand as well as its dependency on various external factors renders STLF a demanding task that requires accuracy and efficient integration in production systems. On the other hand, DR is a tool for the demand side management of peak demand, and the balance of generation and consumption in the electrical grid. Specifically, the present study highlights three main contribution pillars within the field of contemporary buildings and smart grids. The first one initially includes the development of DeepTSF, an open-source machine learning operations (MLOps) software for time series forecasting, initially meant for energy-related time series forecasting, however also capable of generalizing to other domains. Secondly, it entails the development of FlexDR, an AI application providing flexibility forecasting and corresponding demand response recommendations within flexible energy communities. Note that both services promote trustworthy artificial intelligence (TAI) either through explainable artificial intelligence (XAI) features or enhanced human oversight and human-in-the-loop (HITL) approaches. The second pillar encompasses three case studies that present innovative AI methodologies, while also serving as demonstrators of DeepTSF and FlexDR. Said case studies involve: i) a comparative and explainable assessment of DL models for day-ahead load forecasting (STLF); ii) transfer learning for day-ahead load forecasting (STLF); iii) a complete framework for targeted demand response within flexible energy communities (DR); Ultimately, given the recent necessity for ethical development and application of AI systems in the energy sector among others, we wrap up the technological and methodological ML and DL developments of this dissertation with a proposed methodological framework for ethical and trustworthy AI. Said TAI framework is based on the EU "Ethics Guidelines for Trustworthy AI" and focuses on education, motivation, and the establishment of ethical principles for the development of AI systems in the power systems domain. The current study provides significant contributions to the open-source community of ML, DL, MLOps, data science, energy, and time series forecasting sectors, and can be of significant interest for transmission system operators, utilities, and prosumers at both European and international levels.
Supervisor: Professor Dimitris Askounis
PhD Student: Sotiris Pelekis