Predicting electricity demand using machine learning: Case study of Oslo Airport Gardermoen
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3125886Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
- Master's theses (RealTek) [1723]
Sammendrag
With an increasing need for electricity in society and a more complex energy production mix, future electrical power systems require intelligent systems for efficient resource management. In order to realize the potential offlexible resources in the power grid, robust and accurate prediction methodsare required. This thesis presents a case study of Oslo Airport Gardermoen(OSL) to explore the potential of Long Short-Term Memory (LSTM) machinelearning models in predicting electricity demand, particularly focusing onpeak demand forecasting. The models are trained on data from 2022 and2023, utilizing electricity consumption measurements and exogenous factorsincluding passenger numbers, outdoor temperature, and electricity prices.The models demonstrate high accuracy in demand prediction, particularlyfor peak hours.To improve peak prediction capabilities, the thesis implements two mainstrategies. First, models are trained using four different loss functions: MeanSquared Error (MSE), Mean Absolute Percentage Error (MAPE), NegativeLog Likelihood (NLL), and a new proposed Weighted Mean Squared Error (WMSE). Second, a comprehensive grid search and cross-validation routine is performed to robustly determine the optimal model architectures.The best-performing models are characterized by simple model architectureswith just 1 hidden layer and 64 or 128 units, suggesting that less complexmodels can efficiently capture the patterns of the data. These models achieveadequate MAPE scores, with the lowest being 4.53%.The new proposed WMSE loss function emphasizes peak hours and significantly enhances peak prediction reliability. Additionally, NLL enables probabilistic outputs, offering valuable uncertainty estimations for practical applications. This thesis provides a robust and versatile framework adaptableto various energy systems, enabling the development of optimized LSTMmodels for efficient electricity demand forecasting.The implications of this work extend beyond OSL, offering insights for managing flexible resources for efficient and sustainable power system operation. The thesis highlights the promising potential in using advanced machine learning methods for energy management systems, and demonstratestheir ability in large-scale commercial buildings.