Machine learning-based load estimation model for a research office
This project addresses the challenge of load forecasting in research offices, where uneven energy usage leads to inefficient load management. Most effective models require extensive data inputs, such as weather and economic variables, which are not always available. Additionally, most models are s...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2024
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/6862/1/3E_1906647_FYP_report_%2D_JIA_CHENG_MOK.pdf http://eprints.utar.edu.my/6862/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This project addresses the challenge of load forecasting in research offices, where uneven energy usage leads to inefficient load management. Most effective models require extensive data inputs, such as weather and economic
variables, which are not always available. Additionally, most models are static and cannot adapt to changing load patterns, limiting the effectiveness in dynamic environments. This study aims to develop a load forecasting model that uses limited data length (1- 2 years), limited variables—only time and power consumption—to achieve compatible accuracy. The selected algorithms include
Catboost, LSTM, GRU, and CNN-BiLSTM, with a focus on incorporating a self-updating feature to improve adaptability. The results show that while deep learning models achieve reasonable accuracy, Catboost outperformed with an RMSE of 0.0819 kW, MAE of 0.0474kW, and MAPE of 0.5127%. The self-updating Catboost model further enhanced performance compared to its static counterpart to capture future dynamic load, with MAPE below 5% across every
month. The developed models are ready for deployment and require only power consumption data for training, making them both robust and adaptable for predicting future load profiles.
|
|---|