Implementing Machine Learning for Data Center Cooling Optimization
Question
You work as a machine learning specialist for a large software company that has several huge data centers around the world.
Your company has realized they could do a better job managing their data center cooling by implementing a machine learning system to automate the management of the many controls used to control their data center power usage.
The machine learning model needs to input the unlabeled data from building management systems such as chillers, pumps, cooling units, the actual load from systems usage, etc.
You want to run your model to process real-time inferences while also learning from the new inferences. Which combination of SageMaker algorithms and learning techniques should you use for your model to predict settings that optimize cooling on an ongoing basis?
Answers
Explanations
Click on the arrows to vote for the correct answer
A. B. C. D. E.Answer: C.
Option A is incorrect.
A regression algorithm is not the best choice for optimizing unlabeled data; regression algorithms require labeled data.
Option B is incorrect.
The Multilayer Perceptron algorithm is used for speech recognition and translation.
Option C is correct.
Reinforcement learning is used to update your model as new inference observations are encountered continually.
Option D is incorrect.
The Sequence-to-Sequence Neural Network algorithm is used for machine translation and question answering systems.
Option E is incorrect.
The Feedforward Neural Network algorithm is a simple neural network that cannot handle a complex problem like data center power usage effectiveness management.
Reference:
Please see the article titled Transforming Cooling Optimization for Green Data Center via Deep Reinforcement Learning, the Wikipedia article titled Reinforcement learning.
For this scenario, the most appropriate algorithm and learning technique combination for optimizing cooling in a data center is Reinforcement learning (Option C).
Reinforcement learning (RL) is a type of machine learning in which an agent learns to interact with an environment by taking actions and receiving feedback in the form of rewards or penalties. The objective of the agent is to maximize the total reward it receives over time by learning which actions to take in which situations.
In this case, the data center cooling system is the environment, and the agent is the machine learning model. The model can learn from the feedback it receives in the form of temperature and energy usage data and optimize the cooling system by adjusting the settings of the various cooling components such as chillers, pumps, and cooling units.
Reinforcement learning is suitable for this scenario because it can operate in real-time and learn from new data as it becomes available. This makes it possible to continually optimize the cooling system as conditions change in the data center.
Supervised learning using a regression algorithm (Option A) and Supervised learning using a Feedforward Neural Network algorithm (Option E) are not the best choices because they require labeled data, which is not available in this scenario. The data from the building management systems is unlabeled, and the objective is to optimize the cooling system continuously.
Unsupervised learning using a Multilayer Perceptron algorithm (Option B) and Unsupervised learning using a Sequence-to-Sequence Neural Network algorithm (Option D) are not the best choices because they are not designed for reinforcement learning tasks. They are primarily used for clustering, dimensionality reduction, and other unsupervised learning tasks.
Therefore, the best algorithm and learning technique combination for this scenario is Reinforcement learning.
