DNV GL © 2018 SAFER, SMARTER, GREENERDNV GL © 2018

Data Science and Machine Learning in the Energy industry

1

DNV GL Energy

or..

how to create safe and reliable AI solutions in Energy

DNV GL © 2018

Theo Borst

❑ Dutch national, MEng, Electrical Engineering, Delft University of Technology, 1989

MBA, Webster University, Leiden Netherlands, 2002

❑ Director of Product Management at DNV GL Digital Solutions

Data management & analytics, Data products, Digital transformation

❑ Track record on business management, product management, team management,

consultancy, project management, system architecture & software engineering

❑ Background in Energy, IT, Telecom and software industries

Philips Electronics, Ericsson Telecommunication, Thales, Tensing, DNV GL Energy

❑ Expertise on Digital Grid Operation

SCADA/EMS/DMS, Smart Metering, Subst. automation, Grid analytics

❑ Secretary of CIGRE JWG D2/C2.41 on Advanced Analytics for Improved

Situational Awareness of System Operation.

❑ Delegate to European Technology & Innovation

Platform (ETIP), Smart Networks for Energy Transition (SNET)

2

Business Director Digital Grid

DNV GL Energy

DNV GL © 20183

150+ years

100+countries

100,000customers

12,500employees

MARITIME

Hamburg,DE

DIGITAL

SOLUTIONS

Oslo, N

BUSINESS

ASSURANCE

Milan, IT

ENERGY

Arnhem, NL

OIL & GAS

Oslo, N

DNV GL – a quality assurance and risk management company

5%of revenue spent on R&D

Stiftelsen Det Norske Veritas is a free-standing, autonomous and independent foundation whose purpose is to safeguard life, property and the environment.

DNV GL © 2018

Main trends in Energy

4

DNV GL © 20185

DNV GL Energy Transition Outlook

DNV GL © 20186

Digitalization will have a huge effect on the Energy industry

DNV GL © 2018

What are the ‘Digital’ Technologies we’re talking about..

7

DARQ power

Distributed Ledgers

Artificial Intelligence

Extended Reality

Quantum computing

< / >

{…}

\{js}

connectivity

Internet

Mobile

5G

000111

100110

111011

sensors & data

Proliferation of data

Decreasing cost of sensors

IoT

computers

Large and small

Distributed

Cloud and Edge

DNV GL © 2018

Digital Transformation in Energy

Digital transformation will change business models, create new revenue streams and value

producing opportunities. Two waves:

1. Smartgrid technology

Mostly about adding digital hardware in the grid

(protection, automation, controllers, meters, sensors).

2. The Internet of Energy

Is about emerging digital technologies like

data analytics, AI, cloud, mobile and blockchain.

This will add layers of software and applications on

top of the Smartgrid, bringing scalability and

enabling new business models to grid operators.

For grid operators this translates to:

▪ Moving from reactive to predictive

▪ Improved situational awareness and decision support

▪ Predictive asset maintenance

▪ Data driven system operation

8

DNV GL © 2018

Focusing on Artifical Intelligence and Machine Learning

9

DNV GL © 2018

Machine learning in a nutshell:Building predictive models from historical data, for use on new data

10

Data Algorithm Model

Initial observations

Machine Learning give computers the ability to "learn" with data(i.e., progressively improve performance on a specific task), without being explicitly programmed

New observations

Build model

Use model

in operation

Fitted

model

Predicted

outcome

Fitted

model

DNV GL © 2018

Today, machine learning is everywhere

11

DNV GL © 2018

Renewable Penetration

Technology Advancement

Systems interconnection

Data tsunami

Engineering Resources

Extreme Weather

Increasing costs

Cyber Security

Aging Asset

Challenges for Grid or Power System Operators

12

Outages as a result of aging

assets

Reverse Power Flow

Voltage & Frequency

Keeping up with technological

advancement

More OT/IT integration

Increasing data volume and

the challenges of turning data

into information

Aging workforce and limited

skilled resources

Better use of weather information

Prevent internal & external threats

Cost efficiency optimisation

DNV GL © 2018

Use cases of data analytics in Energy, focusing on grid operation

13

▪ Operational decision support

▪ System situational awareness

▪ Renewable energy generation forecasting analytics

▪ Alarm processing and filtering

▪ Weather caused damage prediction

▪ Peak load management analytics

▪ Outage restoration analytics

▪ System oscillations detection (using PMU data)

▪ Real‐time voltage stability monitoring

▪ Fault location and root cause analysis

▪ Asset health assessment analytics

▪ Predictive asset maintenance analytics

▪ Power quality analytics

▪ Load research analytics

▪ Non-technical loss analytics

▪ Cyber security assessment analytics

GRID OPERATION

DNV GL © 2018

Specific use cases of Data science and Machine learning in Energy

▪ Enhanced distributed resource management.

Machine Learning can help utilities realize the next-generation grid through enhanced distributed resource

management that automatically flows power through the grid to deliver more reliable energy and greater

customer choice. Example: Smart charging for EV

▪ Asset optimization.

Machine learning can be used to develop industry intelligence that will predict the probability of failure.

These algorithms take into account industry-wide early failure rates for equipment, creating a richer

understanding of premature failure risks for enhanced asset maintenance, workflow, and portfolio

management.

▪ Outage management.

Utilities can use analytics-validating models to predict and identify outages. Machine learning and device

automation allow for better resource management, reducing downtime and improving reliability. Self-healing

grids can automatically detect and address vulnerabilities, reducing the likelihood of outages.

▪ Customer engagement.

Utilities are mining data with the aid of machine learning to understand customer behavior and service

needs. Using this data, utilities can provide faster and more intuitive interactive customer service via voice

response, personalization, and service matching.

14

DNV GL © 2018

Grid operation in power utilities

15

Controlling and operating the power grid becomes increasingly complex as a result of:

▪ Increased fluctuations due to renewable infeed.

▪ Market driven system operation (FBMC, EU regulations, )

▪ New threats & risks (cyber/physical attacks, severe weather events)

▪ Aging infrastructure/workforce

• Data analytics• Situational awareness• Predictive analytics • Decision support Will reduce complexity in operation

▪ Data analytics

▪ Data visualisation

▪ Situational awareness

▪ Predictive analytics

▪ Decision support

Will reduce complexity in operation

DNV GL © 2018

Changes in system operation: From (reactive) control room to (proactive) decision support centre

16

ASSET

DATA

POWER

QUALITY

WEATHER

DATA

SMART

METER

DATA

CABLE

DATA

WORKFORCE

DATA

PMU DATA

GIS DATA

DNV GL © 2018

Complexity in system operation: Solving the gap between data and actionable information

17

Visual Analytics

DNV GL © 2018

Digital Twins

18

▪ A Digital Twin is a digital representation of a physical object. It includes the model of the physical object,

data from the object, a unique one-to-one correspondence to the object and the ability to monitor the object

▪ Digital Twin together with Machine Learning (ML) promise the ability to simulate assets, scenarios and

operational conditions through forming a cohesive bond between the physical and digital world and between IT

and OT.

▪ It allows for greater integration of information across the asset lifecycle, while ensuring a single view of the

truth at all times

▪ Through the use of Digital Twin and ML, Utilities may be able to reduce downtime, increase planning,

construction and operational efficiency and drive towards zero unplanned maintenance based on real-time

performance monitoring and AI

DNV GL © 2018

Example of a digital twin saving costs and reducing downtime

19

July 2014

Start of generator drive end bearing model deviation

April 2015Bearing failure and full generator replacement was required: total of 15 days downtime

▪ Planned bearing replacement estimated at only 2 days downtime▪ Estimated preventable revenue loss of £15k. Estimated generator cost £80k.

November 2014

Medium confidence alert raised, but no action was taken

DNV GL © 201820

August 2015

Steady generator NDE bearing model deviation

September 2015 onwardsNew bearing means a new SCM component model was required

▪ Bearing failure, resultant damage and potential downtime (500 hours) avoided▪ Estimated prevented revenue loss of £23k

▪ Actively managed wind farm: condition-based maintenance strategy

▪ SCM applied with interventionSeptember 2015

Bearing inspected and replaced total downtime of 12 hours

Example of a digital twin saving costs and reducing downtime

DNV GL © 201821

PD trigger level, for example at 10 kV

21

Partial Discharge in a 10 kV cable

defect

up to 16 km

Synchronization pulsePD pulse

Server

Raw data Warning level

Smart Cable Guard: Predictive analytics solution for cable failures

DNV GL © 2018

Developing a Machine Learning model to automatically detect potential problems in cables

22

50k graphs

Initial goal: Automatic detection of cable problems through ML.No feasible due to not enough data on cable faults

New goal: Reduce amount of human review from 10 % down to 1% by introducing a ML

Operator analysis in 10 % of the cases(currently at 5000 reviews a year, but growing fast)

Initial

Filtering

No problem

detected

Actually OK

False

Negatives

Actually OK

(just noise)

(potential)

Cable problem

~90%

~0%

~9%

~1%

DNV GL © 2018

Examples of DNV GL data analytics pilot projects and tools

23

Anomaly detection for generator plant Data visualisation for optimal outage planning

DNV GL © 2018

Veracity cloud platform

24

DNV GL © 2018

Example of usage of new data sources that help manage the grid:Application of Satellite Remote Sensing in Energy

25

Asset monitoring during construction or operational phase

Land subsidence and ground movements

creates risks for asset infrastructure

DNV GL © 2018

Example of solutions related to data driven system operation:Promaps Realtime

Allows Real time probabilistic risk calculation of power system, and enables the grid operator to quantify current and future risk levels of the power system, and test actions and strategies on a digital twin of the power system.

▪ Input

– Full model export from SCADA system every 5 minutes

– Statistical failure rates for each component, repair time and reconnection time

– Dynamic weather dependent failure rate for power lines

▪ Output

– Failure probability for each component, branch and system

– Risk for loss of load in MWh/h for each load and the system as whole

▪ Creates the baseline for a consistent probabilistic risk management through out the grid owners value chain. The large result data from Promaps Realtime is well suited for machine learning

▪ Example medium power system, one year calculations: 105 120 system risk calculation, consist of apprx 840 million flow calculations (one calculation every 5 minutes, 7000 flow calculation for each simulations)

DNV GL © 2018

..The other side of the story..

27

DNV GL © 2018

Digital Transformation in Energy: some figures

28

40-50%currently using or

testing data analytics

tools and use cases

20%have data analytics

embedded in their

operational processes

Analytics Deployment

90%expected to be using

data driven business

insights by 2020

70%positive toward

security of storing

data in the cloud

Data Utilisation

>16%estimated savings in

OPEX through

digitalization

15%reduction of losses in

Transmission &

Distribution

Cost Savings

DNV GL © 2018

A recent survey on application of digital technology in Energy

Key Findings

▪ 91% of respondents report that embracing

digital technology is crucial to the future

success of their utilities.

▪ Only 23% of utilities have reached a level of

digital maturity where they are making

capital expenditure decisions based on

predictive analytics.

▪ In the next 3 years, 76% of utilities expect

to be able to align digital strategy with

regulatory policy and fill key digital roles in

their enterprise.

29

https://www.forbes.com/sites/parmyolson/2019/03/04/nearly-half-of-all-ai-startups-are-cashing-in-on-hype/

https://etsinsights.com/reports/building-the-21st-century-digital-grid/

DNV GL © 2018

▪ Master asset data in

different systems

▪ No unique Asset ID

▪ No enterprise wide

datamodel

▪ Meaning no

integrated view on

assets capability

▪ No automated

interfaces

▪ Not future proof for

interrelation static

and dynamic data

▪ Poor data

governance

▪ No master data

management

▪ Insufficient

availability

▪ Data quality issues

▪ Insufficient data

integrity control

▪ Limited data

awareness

▪ System awareness,

no person for

enterprise data

responsibility

▪ Reporting, analysis,

planning,

engineering,

inspection &

maintenance based

on siloed data.

▪ Data generated in

many systems

▪ Integrated

information is not

accessible

▪ Poor process

support

Reality check 1: Current data reality in utilities needs to be improved

30

▪ Many manual

process steps

required.

▪ Use of temporary

datasets (excel,

access)

▪ Manual processing

leads to errors.

▪ Manual processing

is labour intensive.

▪ Leads to suboptimal

decisions.

▪ No end to end

insight, leading to

poor decision

making on assets

life cycle.

▪ Poor insight in total

cost of ownership

and end to end

process costs.

▪ Not flexible and

predictable.

▪ Not future proof, no

basis for condition

based maintenance.

Siloed data Data governancePoor info

provisioning

Manual

processingEfficiency

POOR DATA ➔ POOR DECISIONS

DNV GL © 2018

‘Enterprise-Oriented’

Strategic Long Term Focus

‘Project-Oriented’

Tactical Short Term Focus

SystemVendors

Project Teams

ProjectConsultants

System Integrators

Data Governance

Internal SystemsExperts

Corporate Data Governance: long term focus is needed

DNV GL © 2018

How to set up, monitor and improve your data management

▪ Data Management provides foundation for organizing data effectively

Required capability areas:

32

Governance

Organization

and people

Processes

Process

efficiencyMetrics &

dimensions

Requirement

definition

Architecture, tools &

technologies

Data

standards

DNV GL © 2018

Reality check 2 : Data projects are different

33

In data-driven projects 70% of the time is spent on import,

data preparation, quality management and data improvement!

DNV GL © 2018

Data Quality Engine

34

Production line for data,Sensor system, Operational & IT

TechnologyData Asset Data Use

The Organization

DNV GL Runtime services

Data Quality Engine

Quality status &

cleansed data

Data Quality rules

Data Cleaning Engine

Data cleaning

rules

DNV GL © 2018

Organizational Data Maturity Assessment

35

Maturitylevel

GovernanceOrganization and

peopleProcesses

Process Efficiency

Requirement definition

Metrics and dimensions

Architecture, tools and

technologiesData standards

LEVEL 5 -Optimized

Data management policies governs and drives improvements

Data management board oversees improvement activities

Processes for continuous improvement in place

Processes provides feed-back and feed-forward to support continuous improvement

Baseline established and improvements measured according to requirements

Metrics defines baseline to support continuous improvement

Tools support policy driven continuous improvement cycle

Standard compliance and domain models are subject to continuous improvement

LEVEL 4 –Managed

Policies defined in relation to business objectives

Skillset extended to include risk analysis of quality issues aligned with business objectives

Processes for impact analysis and risk mgmt. in place

Monitoring is performed across enterprise and published as KPI’s and trends

Requirements are linked to business impacts

Metrics are linked to business impacts and risk analysis

Tools are driven by business objectives and include support for root cause analysis and risk mgmt.

Standards are used actively to reduce risk for critical business operations

LEVEL 3 –Defined

Policies defined at enterprise level

Roles and required skills defined at enterprise level

Processes are defined and implemented consistently across enterprise

Defined metrics are monitored in advance of business impact

Requirements defined and communicated at enterprise level

Framework for metrics and dimensions defined at enterprise level

Architecture in place at enterprise level supporting full stack data management

Standards, domain models and semantics used at enterprise level

LEVEL 2 –Repeatable

Local initiatives address the requirement for policies

Locally defined roles and some basic skills

Best practices in place but not used consistently

Generic metrics are monitored at point of impact

Local initiatives define requirements

Metrics are reused locally in projects

Tools and technologies used consistently in selected projects

Industry standards and domain models used selectively across projects

LEVEL 1 -Initial

Only ad-hoc or temporal policies in place

No formally defined roles or skillset

Ad-hoc or reactive responses to quality issues

No baseline and no monitoring of quality issues

Re-engineering used to derive requirements

Project specific metrics

Tools are used ad-hoc per project

Ad-hoc and inconsistent use of standards

Objectives,

Policy, Culture, Awareness, Risks, Capabilities to handle DQ issues

Organization, roles, responsibilities, authority, skillsets

Structured and vetted ways of handling and preventing DQ issues

Measure, monitor and use metrics to mitigate DQ issues

DQ Requirements defined,communicated and acted upon

DQ metrics defined, setup, measured and monitored

DQ Tools forprocessing, analysing and correcting DQ issues with data assets

Use available standards, models,ontologies and taxonomies – a corporate «DQ language»

DNV GL © 2018

Reality check 3: How can we put trust in AI?Especially when it concerns operational technology?

36

DNV GL © 2018

Explainable AI (X-AI) and AI Safety

37

When machines are increasingly responsible for crtical decisions in real-time, we have the responsibility for designing explainable and inherently safe AI systems.

DNV GL © 2018

AI in the real world can be hardExample: image classification

38

DNV GL © 2018

Recommendations

▪ Hire digitally minded people and create a digital team that can help

train existing staff and build a culture of innovation.

▪ Develop a coordinated strategic road map that is centered on how

moving up the digital maturity curve will lead to improved decision-

making processes.

▪ Use digitization and digitalization to take a portfolio-based approach

to solving business challenges. This will allow a utility to invest in

one or more use cases with the potential for strong ROI over a

shorter period of time, and one or two use cases that are more

strategic and require a longer time frame to see a return.

▪ Address the data requirements up front internally and with key

strategic partners.

▪ Develop a communication plan that emphasizes the importance of

digital transformation to employees, customers, policymakers,

vendor partners, and regulators.

39

DNV GL © 2018

How can DNV GL help

▪ DNV GL helps to unlock the power of

Digital Transformation in Energy by

extracting value from the new

available grid data and enable

realizing operational benefits.

▪ We can help to:

– Create data-driven solutions

– Advise and provide trainings on

digital solutions or digital strategy

– Be involved in assessment and

quality assurance of digital

solutions.

40

https://www.dnvgl.com/energy/themes/digitalization

DNV GL © 2018

Example Roadmap

41

Deliverables:

• Roadmap

• Requirement definition

• Use cases

• Business plan with clear ROI

DNV GL © 2018

Summary

▪ The challenges and complexities of power system operations increases

▪ Emerging digital technology enables opportunities capture, store and analyse big datasets and apply AI and ML on them to create data-driven solutions

▪ Essential enablers are data management and governance, as well as the ability create trust in AI solutions

▪ Eco-systems and platforms enable unlocking the value of data by trust, privacy and ease

▪ Industry cooperation on data creates additional benefits previously untapped

42

DNV GL © 2018

SAFER, SMARTER, GREENER

www.dnvgl.com

43

DNV GL helps to unlock the power of Digital Transformation in Energyhttps://www.dnvgl.com/energy/themes/digitalization

Theo.Borst@dnvgl.com

Director of Product Management – DNV GL Digital Solutions