O.L. (Bosun) Anifowoshe

The rise in electricity prices in Europe during the post-pandemic era has fueled inflation and slowed economic recovery. Thus, understanding the critical factors that drive electricity prices is indispensable due to the implications of high electricity prices on the socio-economic landscape across Europe. Knowing the critical factors is necessary for accurate short-term and long-term electricity price forecasting which would allow policymakers such as the European Commission to develop optimal energy policies that address future spikes in electricity prices. Therefore, it is a significant goal for scholars and industry to develop forecasting models, which incorporate critical factors for accurate and precise electricity price forecasting. One of the critical factors that influence electricity prices is natural gas prices. The relationship between electricity and natural gas prices has been extensively studied in literature utilizing several forecasting strategies from legacy time series analysis to contemporary machine learning algorithms. However, the relationship between electricity and other exogenous factors such as gas storage reserves, renewable energy production, etc. has received little to no attention. In this study, a short-term and long-term electricity price forecasting model incorporating natural gas pricing and other critical factors as independent variables is proposed. The proposed model will be composed of a data pre-processing and feature selection module, training and forecasting module, and sensitivity analysis module. Historical electricity and energy data in the Netherlands from 2015 – 2021 will be used to build the model and the data will be taken from publicly available database domains and repositories. Exploratory data analysis and correlation analysis using Pearson correlation will be employed to analyze the impacts of different variables on electricity prices and the correlation between features. Features that are highly correlated with other independent variables will be evaluated and removed from the dataset to minimize the impacts of multicollinearity on the model performance. Next, a backward elimination multiple regression model will be employed for the daily prediction of electricity prices to understand the significance level of each independent variable. Variables with coefficients that have a low level of significance will be removed. This elimination process will result in a robust dataset which will be used to build an effective electricity price forecasting model. A hybrid model utilizing Seasonal Integrated Auto-Regressive Moving Average (SARIMAX) and Long-Short Term Memory (LSTM) neural network time series models will be used for training and building the electricity price forecasting model. The SARIMAX model performs well on non-stationary data with exogenous variables. It is simple to implement and yields better results for short-term forecasting, however, it cannot be used to model nonlinear relationships between the dependent and independent variables. The LSTM model on the other hand is capable of modeling nonlinear relationships between variables utilized in complex time series forecasting. It yields a higher accuracy for long-term forecasting; however, it is a black-box model and it can be difficult to understand the intuition behind the results generated by the model. Two performance metrics, i.e., accuracy (mean absolute percentage error (MAPE), root mean square error (RMSE), and correlation coefficient) and convergence rate are used for the performance evaluation of the proposed models. The trained model will be used to run sensitivities to simulate how unexpected changes in exogenous variables can impact future electricity prices. These results from this analysis will be utilized in decision-making to formulate policy recommendations that can significantly influence the impact of changes to these variables on future electricity prices.