Energy Consumption Modelling for 5G Radio Base Stations with Machine Learning
(2023) EITM01 20231Department of Electrical and Information Technology
- Abstract
- With an ongoing energy crisis in Europe and an ever-evolving political climate where the environmental impacts of industries are regarded as a top priority, the topic of energy efficiency is high on everyone’s agenda. Mathematical optimization of energy consumption requires a model of the problem at hand. In this thesis linear regression is compared with the gradient boosted trees method and a neural network to see how well they are able to predict energy consumption from field data of 5G radio base stations.
The main focus of this thesis has been investigating different architectures of the machine learning models and which features result in the best predictive ability of the models. The models have been trained using field data from... (More) - With an ongoing energy crisis in Europe and an ever-evolving political climate where the environmental impacts of industries are regarded as a top priority, the topic of energy efficiency is high on everyone’s agenda. Mathematical optimization of energy consumption requires a model of the problem at hand. In this thesis linear regression is compared with the gradient boosted trees method and a neural network to see how well they are able to predict energy consumption from field data of 5G radio base stations.
The main focus of this thesis has been investigating different architectures of the machine learning models and which features result in the best predictive ability of the models. The models have been trained using field data from deployed radio base station products. This added a complex data processing layer to the work done in this thesis.
The features that resulted in the best performance were, Number of antennas, Configured Max Transmitter Power, Frequency, Bandwidth, IoT capability, Product type, MAC volume, Throughput volume, PRB utilization, RB symbol utilization and MicroSleepTime.
The findings were that the ANN model had the best predictive ability and it, along with the Gradient Boosted Trees model, performed better than the linear regression model, especially for radio products with more complex configurations. The average errors of the models were 9.47 · 10−3 EUs for the ANN model, 10.3 · 10−3 EUs for the Gradient Boosted Trees model and 13.5 · 10−3 EUs for the linear regression model. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9133984
- author
- Ringwald, Daniel LU and Larsson, Daniel LU
- supervisor
- organization
- course
- EITM01 20231
- year
- 2023
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Machine learning, energy modelling, radio base station, 5G, neural network, gradient boosted trees
- report number
- LU/LTH-EIT 2023-953
- language
- English
- id
- 9133984
- date added to LUP
- 2023-10-13 14:16:17
- date last changed
- 2023-10-13 14:16:17
@misc{9133984,
abstract = {{With an ongoing energy crisis in Europe and an ever-evolving political climate where the environmental impacts of industries are regarded as a top priority, the topic of energy efficiency is high on everyone’s agenda. Mathematical optimization of energy consumption requires a model of the problem at hand. In this thesis linear regression is compared with the gradient boosted trees method and a neural network to see how well they are able to predict energy consumption from field data of 5G radio base stations.
The main focus of this thesis has been investigating different architectures of the machine learning models and which features result in the best predictive ability of the models. The models have been trained using field data from deployed radio base station products. This added a complex data processing layer to the work done in this thesis.
The features that resulted in the best performance were, Number of antennas, Configured Max Transmitter Power, Frequency, Bandwidth, IoT capability, Product type, MAC volume, Throughput volume, PRB utilization, RB symbol utilization and MicroSleepTime.
The findings were that the ANN model had the best predictive ability and it, along with the Gradient Boosted Trees model, performed better than the linear regression model, especially for radio products with more complex configurations. The average errors of the models were 9.47 · 10−3 EUs for the ANN model, 10.3 · 10−3 EUs for the Gradient Boosted Trees model and 13.5 · 10−3 EUs for the linear regression model.}},
author = {{Ringwald, Daniel and Larsson, Daniel}},
language = {{eng}},
note = {{Student Paper}},
title = {{Energy Consumption Modelling for 5G Radio Base Stations with Machine Learning}},
year = {{2023}},
}