THERMAL LOAD PREDICTION USING XGBOOST AND LSTM MACHINE LEARNING ALGORITHM IN DATA CENTER ENERGY MANAGEMENT BASED ON DIGITAL TWIN
<p align="justify">In this research, a digital twin technique was developed to make thermal load predictions through real time data on the HVAC system in the data center. Digitization of physical device systems is carried out using IoT (internet of things) technology, through this...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/70307 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | <p align="justify">In this research, a digital twin technique was developed to make thermal load
predictions through real time data on the HVAC system in the data center.
Digitization of physical device systems is carried out using IoT (internet of things)
technology, through this IoT technology digital space is created to represent
predictive models. The goal to be achieved in this study is to obtain predictions of
the thermal load of the data center energy system with good accuracy. through a
data-driven modeling approach and machine learning. The results of the prediction
of thermal load are then analyzed using the heat balance method in order to
determine the ratio of thermal load to performance (cooling capacity) of existing
data center cooling devices.
The development of the digital twin architecture is carried out on existing university
data center objects (CRCS ITB Data Center). Based on a comparison of evaluation
metric scores, the XGBoost algorithm has an advantage in terms of accuracy as
indicated by good RMSE, MSE, MAE and R2 scores. But in terms of training time
the LSTM algorithm has a better training speed value for processing the same
amount of training data.
In the daily thermal load analysis, the predicted thermal load results obtained an
average value of 30.66 kW per hour for October 25, 2022 and 29.88 kW per hour
for October 26, 2022. Thus it can be calculated the value of the balance of heat
load to cooling capacity the nominal installed cooling device units were 40.95%
for PAC 1 and 49.21% for PAC 2. The prediction model for the thermal load of the
server room was obtained using the XGBoost and LSTM machine learning
algorithms with good accuracy. The best RMSE evaluation score uses the XGBoost
algorithm of 0.01 and 0.07 for the LSTM machine learning algorithm.
|
|---|