fbpx
Features Hub Opinion

How Verizon deploys machine learning to improve energy efficiency

Mon 4 Nov 2019 | Oliver Prislan

Verizon’s Network Operations team are on the cutting edge of energy efficiency. Techerati spoke to machine learning project leader Oliver Prislan

Over the past few decades, technological advances have enabled telecom service providers to consolidate their network infrastructure. The reduction in equipment size, coupled with an improved ability to bridge larger distances, has also allowed service providers to reduce their data centre footprint.

But with the reduction in data centre sites, the modern facility has also become comparatively supersized and energy hungry, mainly due to the never-ending increase of network traffic. To mitigate this challenge, companies such as Verizon are turning to machine learning and data analytics to improve the energy efficiency of facilities.

Intelligent efficiency

Over the past four years, in Verizon’s Network Operations, a team led by Oliver Prislan have been applying the latest machine learning methods to forge sophisticated models that improve energy efficiency. Reams of complex data are fed to sensor networks (which transmit data to a centralised analytics platform), providing the information that helps the company reduce operational costs. The team is also simultaneously educating society about the impact of energy hungry data centres on the environment with speaking opportunities at seminars and shows, including Data Centre World Frankfurt next week.

Oliver acquired his master’s in electrical engineering (with a focus on process automation) at the University of Wuppertal — a pioneer in the field of neural networks. The ubiquitous machine learning tools used today – Python and TensorFlow – are based on the same principles Oliver studied in the 90s. He cultivated his knowledge in algorithmic research before entering his current career as manager of Verizon’s global network operations.

Combating cooling costs

When the European Union released its Energy Efficiency Directive policy, Verizon tasked Oliver with auditing the company’s data centre energy consumption.

The bulk of any data centre power consumption is based on the energy consumed by a centre’s cooling systems – in fact, cooling systems account for roughly 38 percent of the operating costs. It is estimated that the proper optimisation of a cooling system can slash operating costs by 25 percent.

Armed with the latest sensors, Oliver began to measure how Verizon’s cooling systems interacted with its data centre fleet, taking stock of variables such as ambient temperature, water temperature and the heat generated by IT systems.

“It all builds up from sensor data to wisdom,” he says. “And this full stack is what is really fascinating from my point of view. The knowledge is inside the data, but to learn from it you have to figure out how to analyse it.”

Once the sensors were configured, the team began analysing the data produced using a statistical technique known as ‘time series analysis’ to forge an algorithm that could identify trends or anomalies, such as power consumption, over time.

The next step for Oliver’s team was to anticipate future energy consumption trends using predictive analytics (PA). PA enables anomaly detection in real time. While simple artificial neural network models require just a few lines of code, the challenge lay in the limit of insights generated. Oliver and his team discovered that while they could predict a future problem, getting to the root of the issue remained an enigma.

Precision control

To combat this, Oliver switched his focus from neural networks to combining time series analysis with advanced techniques providing a model based on a set of differential equations.

This model has allowed Verizon to analyse how its data centres are behaving holistically. Armed with this system description, his team is able to control data centre energy consumption far more efficiently.

While it’s still early days, Verizon has successfully forged a model based on a limited set of input variables. The next step for Oliver’s team is to gradually add more variables — refining the algorithm’s predictive power so it can eventually autonomously adjust controls to suit different scenarios.

To this end, his team continues to experiment with different models to attain greater efficiency and control. For example, he’s studying an application that can enable the data centre to intelligently adjust its systems according to constantly changing weather conditions, which will allow the predictive controller to better utilise free air cooling and avoid attrition, such as starting chillers unnecessarily.

“Today, we operate on more or less fixed set points. But the goal is to attain variable set points in the systems that change in real time,” Oliver says.

Success at scale

Scaling out this application is a challenge for any company that runs an array of data centres in disparate locations, often stacked with different cooling systems. Optimising and integrating this diverse infrastructure side-by-side is what Oliver and his team call “Sisyphus work”.

“The challenges are really different from location to location. We try to concentrate on the big things first, identifying the areas with the biggest potential for the biggest gains, then work through our structured list of activities one by one.”

Oliver emphasises the importance of training and collaboration. For instance, it’s essential that data centre engineers on the ground can work independently to deliver success, even though they have to be closely trained on how to set up sensors and ensure that data is being collected accurately. Equally, it is vital that teams to keep the lines of communications open and transparent.

“Ultimately, success depends on cross functional work between data scientists and engineers to solve complex issues. On the one hand there is the technology knowledge required to understand how e.g. a chiller is working, then there is the IT knowledge needed about how to integrate that in the controller.”

“There are endless opportunities to be leveraged. Identification, business case creation, execution and delivery across all teams remain a challenge. Communicating the benefits in a consistent and strategic way is an essential piece to shape the organization and encourage anybody to participate. The effort is worth it both commercially and environmentally.”

Experts featured:

Oliver Prislan

Manager Tx. Planning & Energy
Verizon

Tags:

energy efficiency machine learning
Send us a correction Send us a news tip




Do NOT follow this link or you will be banned from the site!