A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Engineering in Energy, School of Environment, Resources and Development

Thesis (Ph.D.) - Asian Institute of Technology, 2021

At present, electricity price forecasts have become a fundamental input to energy utility and company for decision-making mechanisms. Electricity price forecasting focuses on predicting the spot and forward prices in the wholesale electricity market. The electricity price forecasting depends on weather and the intensity of business and everyday activities. These unique characteristics lead to price dynamics and price spikes. For these reasons, electricity price forecasting is a big challenge in the energy market. A variety of methods and ideas have been developed with varying degrees of success. Computational intelligence models or call machine learning techniques gain popularity in recent years because their major strength is the ability to handle complexity and non-linearity, like electricity price volatility. Therefore, this study will focus on modeling electricity price forecasting based on machine learning techniques with real data simulation. In this work, to examine electricity price forecasting, the proposed electricity forecasting models were formulated based on conventional and modern machine learning techniques. The novel residual neural network in electricity price forecasting was first introduced in this study. The simulation results were evaluated using the coverage width-base criterion, which showed that the proposed model could improve forecasting results accurately and reliable. The average forecasting error was reduced by about 13% error with comparing to multilayer perceptron models. Moreover, this study tried to reduce forecasting error by improving electricity demand and renewable energy resource forecasting. The proposed forecasting model cooperated with demand and renewable energy resource forecasts. The results were reduced below 1%, while the benchmark models are around 5% error.