Using artificial intelligence to help data centers navigate to net-zero

Venessa Moffat

8 August 2023

This blog post analyses three key challenges data centres face in cutting their footprint, suggesting potential solutions to overcome them.

As the amount of data we need to host, store and distribute grows, so too does the need to do each of those things in an efficient and sustainable way.

The sector, however, is facing challenges in decarbonising as it already accounts for some 2% of global emissions ahead of its further predicted growth.

A sustainable and more efficient future for data center owner/operators is certainly within reach if they use the right tools.

From the application of AI to the growth of m-commerce, the birth of the metaverse to the burgeoning Internet of Things, the way we live our lives and run our businesses today is driving a data explosion. As the amount of data we need to host, store and distribute grows, so too does the need to do each of those things in an efficient and sustainable way.

Cutting the footprint of data

To meet targets set in the Paris Climate Agreement, countries are desperately trying to gear up legislation to reduce net greenhouse gas emissions by 55% by 2030 – a challenge for a sector which already accounts for some 2% of global emissions ahead of its further predicted growth.

From next year, any data center owner/operator with a turnover of over €150 million will need to find a way of reporting on their sustainability performance in line with the EU’s Corporate Sustainability Reporting Directive (CSRD).

This is estimated to impact around 55,000 businesses. In addition, as of 15 May 2024, companies with data centers over 500kW will have to report on the previous year’s data (starting 15 May 2023) under the European Energy Efficiency Directive, and this is likely to be similar, if not the same metrics, based on the ISO 30134 / EN 50600.

And then there is the energy price spike of the past year which has provided an additional reminder that both planet and profits suffer when scarce resources aren’t used wisely.

How data centers can improve their operational sustainability

There’s a clear imperative for data center operators to reliably measure, deliver and report concrete improvements in operational sustainability in the next 12 months.

To rise to this challenge, I believe three critical shifts need to take place in the near term.

Understanding the metrics that matter

Until now, a main benchmark for data center efficiency has been Power Usage Effectiveness (PUE) – the ratio of the total amount of power delivered to a facility compared to that provided to the computing equipment.

While the quest for lower PUE figures has undoubtedly contributed to greater data center efficiency in the last ten years, future efficiency gains and reporting requirements under CSRD will require more nuanced measurements covering nine areas of resource use and IT equipment.

Of these, two metrics – IT Equipment Energy Efficiency (ITEEsv) and IT Equipment Utilization (ITEUsv) – are both key to future gains in efficiency and sustainability.

In the future, it will be imperative that all operators are able to measure and improve these key metrics from the IT stack.

Allocating one clear owner for data center efficiency

Today, responsibility for a data center will typically fall between two teams with conflicting agendas: a facilities or estates team responsible for managing the energy budget; and an IT team tasked with maintaining a high-performing, high-availability IT estate.

To succeed in improving data center efficiency and reducing energy costs, both teams need to work together with a clear mandate and targets.

To achieve this, organisations need to review their internal organisation and restructure if necessary to incentivise teams to work together towards the same goals.

Readiness to evaluate and adopt tools

A third shift that needs to take place is the readiness to evaluate and adopt tools that can improve resource use, cut carbon emissions and deliver the assured reporting required for CSRD, EED and the respective sustainability standards.

Talking to consultants and partners tasked with advising the biggest businesses on how to manage this challenge, the scale of data gathering, and time required to understand the changes needed is an almost insurmountable challenge as the deadlines for reporting against standards get closer.

New sustainability tech companies are working hard to solve the reporting challenges using AI and machine learning to provide answers and actions for each element of the data center operations.

A sustainable and more efficient future for data center owner/operators is certainly within reach. The demands of sustainability reporting will not be too challenging, provided organisations step back, understand the changes they need to make and collaborate with those who have the knowledge and tools to help them get there.

This article was published in SG Voice on 8 August 2023

Sustainability for data centres: three critical challenges and how to overcome them

DEUTSCHE

Die vierte industrielle Revolution, oder Industrie 4.0, ist gekennzeichnet durch das intelligente Vernetzen von Maschinen, Produkten und Mitarbeitern mit cyber-physikalischen Systemen in intelligenten Fabriken. Die Produktion in intelligenten Fabriken wird hochflexibel. Die Optimierung der Wertschöpfungskette erfolgt darin basierend auf Echtzeit-Anforderungen, Mitarbeiter werden von intelligenten Kontrollsystemen profitieren. In der intelligenten Fabrik wird die Kombination aus Material, Lieferanten, Ausrüstung und qualifizierten Mitarbeitern so organisiert, dass sie Einheiten produzieren, die den Kundenanforderungen in Echtzeit genügen. Die Personalisierung auf Kunden- und Marktsegmente wird durch integrierten Bedarfssignale gewährleistet, um skalierbare Geschäfte zu erzielen. Die intelligente Fabrik erfordert eine flexible und agile vernetzte Produktion.

Flexibilität bedeutet, dass die Betriebsmittel in der Lage sein müssen, eine Vielzahl unterschiedlicher Produktionsprozesse durchzuführen. Die Mitarbeiter müssen über ein sehr breites Spektrum an Qualifikationen und Erfahrungen im Umgang mit verschiedenen Systemen und Methoden verfügen.

Darüber hinaus erfordert die Flexibilität eine sehr hohe Verfügbarkeit der Einsatzmittel. Vernetzt bedeutet, dass alle relevanten Informationen für die Herstellung eines Produkts in Echtzeit zwischen allen Systemen und den am Prozess beteiligten Personen ausgetauscht werden. Tritt beispielsweise bei kritischen Anlagen ein Fehler auf, müssen neue Wege durch die Produktion gefunden werden, um Aufträge termingerecht zu erfüllen. Um das zu erzielen müssen alternative Arbeitspläne und die daraus resultierenden Produktionskosten und -zeiten analysiert werden. Neue Produktionseinheiten entstehen durch die neue Kombination von Ressourcen, Ausrüstung und Personal pro Auftrag. Dies erfordert eine enge Interaktion zwischen Ausrüstung, Ressourcen und Mitarbeitern mit einem intelligenten Produktionssystem, das sowohl die alternativen Routen als auch die Produktionskapazitäten verwaltet. Daher ist die Integration von Daten aus verschiedenen Systemen intern und extern unerlässlich.

Ziel dieses Forschungsprojekts ist es, Techniken zur Realisierung schlanker intelligenter Fabriken zu entwickeln, umzusetzen und zu validieren:

Das erste Projektziel ist die Entwicklung von neuen Algorithmen die den aktuellen Zustand der Fabrik bei der Erstellung des Produktionsplans berücksichtigen.

Das zweite Projektziel ist die Entwicklung einer Methodik für das Lösen von Systemen mit multiplen und teilweise probabilistischen Randbedingungen basierend auf reinforcement learning techniques.

Das dritte Projektziel ist es, eine mathematische Darstellung des genauen Zustands und Verhaltens der Produktionsanlagen zu generieren, die QiO den PARCS score nennt.

In diesem Arbeitspaket werden hochdimensionale teilweise diskrete Optimierungsprobleme mit einer Vielzahl von Nebenbedingungen mit modernen mathematischen Verfahren bearbeitet.

ENGLISH

The fourth industrial revolution, or Industry 4.0, is marked by the intelligent networking of machines, products and employees with cyber-physical systems in Smart Factories. Production in Smart Factories will be highly flexible, highly productive and resource efficient. The optimization of the value chain is based on real-time requirements. At the Smart Factory, the combination of materials, suppliers, equipment and skilled employees is organized to produce units that meet customer demands in real time. Personalization to customer and market segments is ensured by integrated demand signals to achieve scalable business.

The Smart Factory requires flexible and agile networked production. Flexibility means that the operating resources must be able to carry out a variety of different production processes. Employees are required to have a very broad range of qualifications and experience in using different systems and methods. Furthermore, flexibility requires a very high availability of operating resources.

Networked means that all the relevant information for the manufacture of a product is shared in real-time between all of the systems and people involved in the process. This requires information about potential machine breakdown, delays to a supplier order, missing parts, personnel availability or available skills.

If a fault occurs to critical piece of equipment, for instance, new routes through production must be found in order to fulfil orders on time. In order to do this, alternative work schedules and the resulting production costs and times must be analysed. New production units arise through the new combination of resources, equipment and personnel per order. This requires close interaction between equipment, resources and the workforce with a master intelligent production system, which manages the alternative routes as well as production capacity.

As such integration of data from diverse systems internal and external is therefore essential. The aim of this research project is to develop, implement and validate techniques for the realization of Smart Factories:

The first project objective is the development of new algorithms that take into account the current state of the factory when creating the production plan.

The second project objective is the development of a methodology for solving systems with multiple and partially probabilistic constraints based on reinforcement learning techniques.

The third goal of the project is to generate a mathematical representation of the exact condition and behaviour of the production facilities, which QiO calls the PARCS™ score. It is designed to enable engineers within software tools such as the QiO Foresight Engine™.

This work package deals with high-dimensional, partially discrete optimization problems with a variety of constraints using modern mathematical methods. The results of the project proposed here go well beyond the general state of the art, especially in the fields:

Artificial intelligence for the dynamic learning and modeling of production processes

Artificial intelligence for production planning and supply chain management

Traditional approaches to modeling and describing production processes assume a well-known problem and seek a solution to a specific problem. This approach is not suitable for solving such problems in the context of an intelligent factory. The boundary conditions that describe the processes change regularly here and they are mostly of a probabilistic nature. We plan to learn the processes dynamically based on the data produced by a production hall. We call such dat, the voice of the production hall.

Such automated learning is important because constructing a graphical model by a human expert can be cumbersome and time-consuming. If, as is often the case, there is a database of examples, learning techniques can do at least part of the work. This approach should lead to an always current and dynamic view on the production processes. It should also form a mathematical basis for dynamic production planning as it will be described in the next.