PREDICTIVE MODELLING OF POWER CONSUMPTION IN

CLOUD DATA CENTERS USING MACHINE LEARNING

University College of Engineering (A), Osmania University

UNDER THE GUIDANCE OF:

By
G.Sreeja (100523744211)
 CONTENTS
• Motivation
• Aim
• Objectives
• Introduction
• Literature Review
• Software Required
• Proposed Method
• System Architecture
• Advantages
• Disadvantages
• Results
• Conclusion
• Future Scope
• References
MOTIVATION
The rapid growth of cloud computing has resulted in a notable rise in the quantity and size of data
centers across the globe. These centers now consume a substantial amount of electricity. This
increase in energy use places an extensive cost on service providers.
Green computing techniques are becoming increasingly important as people become increasingly
mindful of how energy usage affects climate change. As a result, the search for energy-efficient
cloud data center solutions is a vital aspect of both technological growth and social responsibility,
not only an environmental or financial issue.
AIM
To optimize power efficiency in cloud data centers by integrated machine learning approach.
OBJECTIVES
The objective of this project is:
•To reduce energy usage and emissions for environmental reasons.
•To cut operating expenses for financial reasons.
•To take on the technological challenge of optimizing cloud infrastructure.
•To increase cloud computing energy efficiency through a comprehensive and creative approach.
INTRODUCTION
•The cloud data centers causes substantial electrical energy consumption concerns.
•Energy-efficient solutions are needed to reduce the impact of this energy use on the environment
and the economy.
•Green Cloud Computing aims to minimize data center energy use and carbon footprint.
•This project introduces a new architecture for Energy-Efficient Hybrids that optimize power
consumption in cloud data centers.
•By minimizing energy use and lowering data centers' carbon footprint, this strategy seeks to
balance scientific advancement with environmental protection.
LITERATURE REVIEW
Insufficient Handling of Dynamic Workloads: Existing frameworks and algorithms are unable
to dynamically adjust to evolving server conditions and workloads. Resource allocation has
advanced, but these methods are not always flexible enough to react quickly to changes in
workload intensity, which leads to inefficiencies and higher energy use.
Underutilization of Predictive Analytics: Many current frameworks fall short of realizing the
full promise of predictive analytics in optimizing data center operations. There is a lack of wider
integration of predictive models in operational decision-making.
Limited Focus Made to Wider Environmental Effects: Although cutting energy use is a top
priority, the wider environmental effects such as resource use and carbon emissions are frequently
neglected. While studies say to address this by bringing in a broader ecological perspective, more
thorough attention is still needed in this area.
SOFTWARE REQUIRED
Software:
• Google Colab

Libraries:
• Scikit-learn
• Numpy
• Pandas
PROPOSED METHOD
•The EEH framework incorporates both request scheduling and server consolidation.
•It begins by classifying client requests according to their power and time needs.
•A scheduling algorithm then takes power consumption into account while making decisions.
•A consolidation algorithm is used to determine which servers are overloaded or underloaded,
allowing virtual machines to be strategically moved to optimize server loads.
•To assess the effectiveness of the EEH framework, a machine learning model specifically, linear
regression is used.
•Based on a number of variables, including resource type, time required, and server load, this
model forecasts power consumption.
•The dynamic thresholding aspect of the model improves its practical application in real-world
circumstances by adjusting forecasts depending on real-time data, such as peak hours and server
load.
•This study showcases the potential of the EEH framework in decreasing power usage and
supporting sustainable cloud computing operations by applying the model through rigorous
testing and visualizations.
System Architecture
ADVANTAGES
•It cuts operating expenses which makes it cost effective.
•It predicts power consumption trends and offer useful, actionable insights.
•Proactive energy management techniques in cloud data centers are made possible in large part by
this predictive power.
•A crucial prerequisite for real-world applications is the model's flexibility in responding to
temporal fluctuations, which is further highlighted by the dynamic thresholding.
•It optimizes energy usage and reduces carbon emissions.
DISADVANTAGES
•Even though the model works well with synthetic data, real-world validation is required to verify
its usefulness.
•Furthermore, investigating more complex models and adjusting the model's hyperparameters
might result in even greater performance gains.
RESULTS
• The training Root Mean Squared Error (RMSE) was calculated at 4.940876335571659, which
provides a measure of the average error magnitude in the same units as the power
consumption.
• The test results showed an MSE of 25.104161760187225, suggesting a satisfactory level of
precision in the predictions.
• Furthermore, an R-squared value of 0.9163845109242488 on the test set indicates a strong
correlation between the predicted and actual power consumption values, with the model
explaining over 91% of the variance.
Time Series Analysis
The time series analysis provided deeper insights into the model's predictive capabilities over
time (Fig 1, Fig 2 and Fig 3).
The visualizations of actual vs. predicted power consumption show a close alignment, with the
predicted values mirroring the actual consumption trends.
Fig 1. Actual vs Predicted Power Consumption (Training Set)
Fig 2. Actual vs Predicted Power Consumption (Test Set)
Fig 3. Actual vs Predicted Power Consumption Over Time
CONCLUSION
• The linear regression model, estimates energy usage accurately, as proven by the low MSE
and high R² values acquired during testing, demonstrated the effectiveness of the EEH
framework.
• The time series analysis’s confirms the model's resilience even more, especially in terms of
its ability to represent the periodic patterns of power consumption linked to peak and off-
peak times.
• The model's capacity to adjust to real-time operational situations was highlighted by the
dynamic thresholding technique.
• The use of such predictive models can result in more strategically allocated resources,
cheaper operating costs, and a smaller environmental effect.
FUTURE SCOPE
•To optimize cloud computing environments by focusing on real-world data validation, advanced
predictive models, integration with renewable energy sources.
•To integrate the EEH framework into a broader energy management system that includes real-
time monitoring and automated control mechanisms.
REFERENCES
1. Alarifi, A., Dubey, K., Amoon, M., Altameem, T., Abd El-Samie, F. E., Altameem, A., ... & Nasr, A. A.
(2020). Energy-efficient hybrid framework for green cloud computing. IEEE Access, 8, 115356 -115369.
2. Bindhu, V., & Joe, M. (2019). Green cloud computing solution for operational cost efficiency and
environmental impact reduction. Journal of ISMAC, 1(02), 120-128.

3. Xu, L., Li, C., Li, L., Liu, Y., Yang, Z., & Liu, Y. (2014, June). A virtual data center deployment model
based on green cloud computing. In 2014 IEEE/ACIS 13th International Conference on Computer and
Information Science (ICIS) (pp. 235-240). IEEE.

4. Wadhwa, B., & Verma, A. (2014, August). Energy and carbon efficient VM placement and migration
technique for green cloud datacenters. In 2014 Seventh international conference on contemporary
computing (IC3) (pp. 189-193).
THANK YOU!