University of Twente Student Theses

Login

Explainable Forecasting Models For Load Forecasting

Trivedi, A. (2024) Explainable Forecasting Models For Load Forecasting.

[img] PDF
4MB
Abstract:Electricity grids face rising demand from population growth and renew- able energy integration, creating a crucial need for efficient and explainable forecasting models. The research seeks to evaluate the performance and interpretability of Facebook (FB) Prophet and LSTM (Long-Short-Term Mem- ory) under data constraints compared to Transformer-based solutions such as PatchTST (Patch TS Transformer). This study performed exploratory data analysis (EDA) on a decentralized electricity grid in the Netherlands and predicted the power consumption of a company using the above-stated mod- els. The methodology uses accuracy metrics: (i) Mean Square Error (MSE), (ii) Mean Absolute Error (MAE) and SHapley Additive exPlanations (SHAP) for the explainability of models. FbProphet showed efficiency for short- term forecasting from 15-minute to 1-hour intervals, making it suitable for real-time operational decisions. PatchTST performs consistently for longer horizons (>= 6 hours), benefiting long-term planning and resource allocation. LSTM models require further tuning or additional data to improve accuracy. PatchTST or FbProphet can be selected based on the specific forecasting horizons and availability of training data. In future, these models can forecast load profiles inside a given company to gain insights into the utilization of energy within a company. Additionally, it can expand on explainability tools for Transformer-based models
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Awards:Nominee for best paper
Link to this item:https://purl.utwente.nl/essays/100847
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page