Development of new electricity system marginal price forecasting models using artificial intelligence and statistical methods

In the electricity sector, the System Marginal Price (SMP) is a critical determinant in balancing energy demand and supply. SMP values are subject to constant fluctuations due to a variety of complex factors, including changes in energy demand, variations in supply sources, policy regulations, and environmental conditions. These fluctuations make SMP a crucial metric for energy producers, distributors, and consumers, as it reflects the dynamic variables of the energy market. The balance between the supply and demand of energy resources, changes in energy imports and exports, policy interventions, and technological advancements are key influences in the determination of marginal prices. Within this context, SMP holds strategic importance for decision-makers in the electricity sector, providing essential insights that enable energy companies and regulators to better navigate the complexities of the energy market and ensure effective energy management. The primary objective of this study is to develop SMP forecasting models for the Turkish electricity market, incorporating an interface grounded in hyperparametric elasticity. To achieve this, artificial intelligence techniques and advanced statistical methods are employed to produce highly accurate SMP forecasts. In particular, this thesis applies feature selection algorithms, such as Minimum Redundancy Maximum Relevance (mRMR) and Maximum Likelihood Feature Selector (MLFS), to enhance the performance of SMP forecasting models. The dataset utilized in this study spans from January 1, 2021, to September 14, 2023, and is provided by the Energy Exchange Istanbul (EXIST). The results demonstrate that both the Extreme Learning Machine (ELM) and eXtreme Gradient Boosting (XGBoost)-based models showed comparably strong performance, highlighting their potential effectiveness in SMP forecasting. In contrast, the Monte Carlo (MC) method generally produces poor results, indicating its limited effectiveness for SMP forecasting. Keywords: Electric Energy Sector, Machine Learning, System Marginal Price