Abstract
The electricity market is one of the most dynamically changing sectors on a global scale and it is undergoing remarkable reforms in many areas. In its early stages (e.g., solar and wind), new needs and challenges are emerging that must be addressed. Additionally, new technologies—such as the rapid development of artificial intelligence (AI)—are creating new possibilities that provide excellent tools for market operation. At the same time, new regulatory rules aimed at abolishing market monopolies and creating a unified, liberalized electricity market are generating new values and demands to ensure smooth operation. In order to ensure the smooth operation of the wholesale electricity market, it is imperative to create and use advanced artificial intelligence tools capable of addressing these new challenges. The adoption of state-of-the-art AI mechanisms can create favorable conditions to predict both the electricity demand of end users and the fluctuations in electricity prices with opt ...
The electricity market is one of the most dynamically changing sectors on a global scale and it is undergoing remarkable reforms in many areas. In its early stages (e.g., solar and wind), new needs and challenges are emerging that must be addressed. Additionally, new technologies—such as the rapid development of artificial intelligence (AI)—are creating new possibilities that provide excellent tools for market operation. At the same time, new regulatory rules aimed at abolishing market monopolies and creating a unified, liberalized electricity market are generating new values and demands to ensure smooth operation. In order to ensure the smooth operation of the wholesale electricity market, it is imperative to create and use advanced artificial intelligence tools capable of addressing these new challenges. The adoption of state-of-the-art AI mechanisms can create favorable conditions to predict both the electricity demand of end users and the fluctuations in electricity prices with optimal accuracy. The main objective of this PhD dissertation is to develop advanced and novel algorithms for electricity load and price forecasting, specifically for the operation of modern wholesale market mechanisms. More precisely, the research focuses on developing algorithms and advanced Deep Learning models that can achieve very high short-term forecasting accuracy, serving as effective tools in innovative programs within the modern electricity market. The research aims to cover various aspects of the modern electricity market by examining electricity mechanisms—specifically the Day Ahead Market, Intraday Market, and Balancing Market—and by exploring applications for active consumer participation such as Demand Side Management (which includes peak clipping, valley filling, load shifting, and load conservation) as well as Demand Response, in addition to forecasting the load of electric vehicles (EV Load) and investigating Transfer Learning with particular consideration given to the more challenging case of the day-ahead market. This dissertation innovates in the following key components: Development and creation of advanced and novel Deep Learning models: These models can be used both for electricity load and price forecasting, as well as for other time series scientific tasks, such as problems in the economic and industrial sectors. Implementation of Novel Deep Learning Models for Electricity Demand Forecasting: These models can predict energy demand under various periods and conditions, assisting power producers adjust their generation and optimize operations. Implementation of novel Deep Learning Models for Wholesale Electricity Price Forecasting: This enables energy producers to make decisions regarding large-scale market operations and production, taking into account the forecasted prices. Analysis of the Price Formulation Framework in Wholesale Markets: By using data from different energy markets, trends, market players’ strategies, and their interactions are analyzed to better understand how the final hourly prices are determined. Implementation of Transfer Learning Methods for the Day-Ahead Electricity Market: The developed models perform optimally in forecasting 24 hourly steps ahead, a challenging task in the current literature. Investigation of Demand Side Management and Demand Response Programs: These strategies are examined for their integration into market operations through active consumer participation. In this way, optimization functions to optimize the network operation and cost, covering both production/supply side are used. Development of New Methods for Forecasting the Load of Electric Vehicles: Considering five different datasets, the distinctive patterns of the EV-load curve are detected, isolated, and forecasted separately. Investigation of the new trends emerging in the operation of the modern electricity market through the study and presentation of various studies, which use different datasets and cover the full spectrum of forecasting algorithms.
show more
