40
© GSM Association 2014 Mobile Energy Efficiency The GSMA’s Mobile Energy Efficiency Services 5GrEEn Summer School 28 August 2014
© GSM Association 2014
Mobile Energy Efficiency
The GSMA’s Mobile Energy
Efficiency Services
5GrEEn Summer School
28 August 2014
2
The GSMA’s Mobile Energy Efficiency (MEE) offers two
services: MEE Benchmarking and MEE Optimisation
The GSMA’s MEE Benchmarking service is a management tool that
– helps MNOs measure and monitor the relative efficiency of their
networks
– identifies under-performing networks and quantifies potential
efficiency gains, typically 10% to 25% across a MNO’s portfolio
MEE Optimisation (MEEO) is a follow-on service that uses site
audits and equipment trials to
– analyse the costs and benefits of specific actions to reduce
energy and emissions, and gather real performance data; and
– roll out the most attractive solutions.
The service is run in partnership with a third party, e.g. a vendor
MNOs spend $15bn globally on network electricity and diesel fuel
consumption, accounting for about 70 MtCO2
3
Agenda
Introduction to MEE services
MEE Benchmarking – How it works
– Insights
MEE Optimisation
4
40 MNOs have participated in MEE Benchmarking, or
over half of global mobile subscribers
Launch MEE Benchmarking for MNOs to:
– compare multiple networks on a like-for-like basis and
against standard energy KPIs
– help reduce energy consumption, costs and emissions
Incorporate in new ITU SG5 energy and carbon methodologies
Increase participation and develop MEE Benchmarking so that
it becomes increasingly useful to operators
Coordinate with other industry and regulatory stakeholders so
that the benchmarking methodology is adopted as a global
standard by the industry
40 MNO participants, accounting for more than 200 networks across
145 countries and over 50% of global mobile subscribers
(2010)
Ongoing
Ongoing
Objectives Status
(2011)
5
Some of the MEE Benchmarking participants
6
Mexico
Canada
Costa Rica
Brazil
Morocco
Mauritania
Algeria Qatar
South Africa
Australia
Mongolia Kazakhstan
China Japan
Alaska
Greenland
USA
Argentina
Chile Uruguay
Paraguay
Bolivia
Peru
Ecuador
Colombia
Venezuela
Surinam
Fr. Guyana
Guyana
Cuba
Jamaica
Dominic. Rep.
Bahamas
Guatemala
Belize
Honduras Nicaragua
Panama
El Salvador
New Zealand
Papua New Guinea
Indonesia
Malaysia
Philippines
Vietnam
Thailand
Myanmar Laos
Cambodia
Taiwan
South Korea
North Korea Kyrgyzstan
Tajikistan
Uzbekistan
India
Bangladesh
Bhutan Nepal Pakistan
Afghanistan
Turkmenistan
Iran
Russia
Oman
Yemen
U.A.E Saudi
Arabia
Iraq
Turkey
Syria Lebanon
Egypt
Israel
Sudan
Ethiopia
Somalia
Eritrea
Libya
Mali Senegal
Sierra Leone
Liberia
Ivory Coast
Ghana
Burkina Faso
Niger
Guinea Nigeria
Lesotho
Mozambique
Madagascar Botswana
Namibia
Angola
Zambia
Zimbabwe
Tanzania
D. R. of Congo
Congo
Gabon
Cameroon
Chad
Kenya
Uganda
Finland
Sweden
Norway Iceland
Great
Britain Ireland
Spain Portugal
France
Italy
Germany
Poland Ukraine
Belarus
Romania
Greece
Participant in MEE
MEE participants are located in 145 countries
7
There are 6 key benefits of MEE Benchmarking for MNOs
1. A detailed analysis of relative network performance against a large
dataset: potential energy cost and carbon emissions savings of
10% to 25% per annum are typical for underperforming networks
2. Unique “normalisation” analysis enables like-for-like comparison
3. Suggested high level insights to improve efficiency
4. Annual participation to track improvements over time and quantify the
impact of cost reduction initiatives
5. Demonstration of positive action on energy and emissions reduction
to stakeholders
6. Confidentiality: external comparisons are made anonymously
8
MEE Benchmarking methodology compares networks
against 4 KPIs using a unique normalisation methodology
Networks are compared against four Key Performance Indicators
(KPIs)
1. Energy consumption per mobile connection
2. Energy consumption per unit mobile traffic
3. Energy consumption per cell site
4. Energy consumption per unit of mobile revenue
Unique analytical approach allows MNOs to compare their networks
against one another and against their peers on a like-for-like basis
– Variables outside the MNO’s control, e.g. population distribution
and climate, are ‘normalised’ using regression techniques
– Networks can then be compared like-for-like
9
Prior to any “normalisation”, Network A appears
inefficient and Network Q efficient
Mobile network operations electricity and diesel usage per connection, 2013
A B C D E F G H I J K L
kWh per connection
Country
10x
spread
between
best and
worst
Diesel usage Electricity usage Key
Network A inefficient?
Network Q efficient?
Source: MNOs, UN, GSMA data and analysis
M Y X W U V N O P Q R S T
Example: energy per connection, illustrative
10
There is a strong relationship between number of cell
sites per connection and energy per connection
0
5
10
15
20
25
30
35
40
45
0.0 1.0 2.0 3.0
Energy per
connection
(kWh)
Cell sites per thousand connections
Network A
Line of best fit
Example: energy per connection, illustrative
Network Q
11
Normalising for cell sites per connection still shows
Network A to be high energy but only just
0
5
10
15
20
25
30
35
40
45
0.0 1.0 2.0 3.0
Energy per
connection
(kWh)
Cell sites per thousand
connections
Network A Network A actual
Network A expected
Industry average
Post
normalisation
result
Pre
normalisation
result
Line of best fit
Example: energy per connection, illustrative
12
However, it is more meaningful to include other
variables in the normalisation
Energy per connection is normalised using the following four variables:
The regression analysis thus captures the impact of country, market and
technology factors. Other variables are also tested in order to compare
statistical significance, run sensitivity analyses and to verify the results
Example: energy per connection, illustrative
Normalisation Variable Comment
Number of cell sites per
connection
A single measure that accounts for population density,
market share, topology and technology
Number of cooling degree
days per capita (population
weighted)
A measure of temperature that reflects the energy needed
for cooling
Voice traffic per connection A measure of how active average connections are in terms
of voice usage
Data traffic per connection A measure of how active average connections are in terms
of data usage
13
Normalisation against four variables shows a truer
picture: Network A is more efficient than Network Q
kWh per
connection
A B C P E H G N Q X I L
Country
Difference between actual electrical and diesel energy usage per mobile
connection and the expected value, 2013
Network A more efficient than Q
Source: MNOs, UN, GSMA data and analysis
R2 = 76%
J R U M D Y V W O S F T K
Example: energy per connection, illustrative
Mobile operations diesel & electricity usage per connection regressed against:
- Number of cell sites per connection
- Data traffic per connection
- Number of cooling degree days per capita (population weighted)
- Voice traffic per connection
Regression variables
14
Operators receive anonymised comparisons against
other MNOs, with their networks highlighted
Notes on methodology: An effective diesel generator efficiency of 30% has been used to convert from MWh of
diesel to MWh of electricity generated by the diesel generator. This takes into account the average diesel generator
efficiency of 20% plus the diesel-saving benefit MNOs are receiving from using batteries to reduce genset run time,
especially off grid
* This is an ‘average’ across MNO X’s data and relates to confidence in energy consumption data
Source: GSMA, MNOs
MNO X Other Operators Dataset average Key
Mobile operations’ average estimated RAN grid electricity and diesel generated
electricity usage per mobile connection, 2012 (kWh / connection)
Data confidence
MNO X data*: Medium/High
Overall data set: Medium/High
MNO X
Actual data, pre normalisation
15
The benchmarking results are used to quantify
energy cost saving potential: here it is $89m per year
2013
RAN
energy
(GWh)
Elec. cost
($/kWh)
Diesel
cost ($/l)
Est’d
energy
cost ($m)
% saving
to
average
% saving
to top
quartile
Saving to
average
$m
Saving to
top
quartile
$m
Canada 424 0.09 1.18 38 3% 13% 1 5
France 289 0.11 1.37 32 0% 2% 0 0
India 3666 0.14 0.69 313 9% 21% 28 66
Etc. … … … … … … … …
Total 5736 608 37 89
Example: energy per connection, financials - illustrative
It is not possible to determine how much of the circa $90m p.a. is cost-effective using
the MEE Benchmarking analysis. MEE Optimisation service addresses this
16
Agenda
Introduction to MEE services
MEE Benchmarking – How it works
– Insights
MEE Optimisation
17
It is challenging for many operators to measure and
manage energy consumption and costs
Some major operators do not manage energy centrally, although this
is beginning to change
It is difficult for many operators to gather high quality energy
consumption data from their networks
People responsible for managing energy often struggle to obtain
information regarding key drivers of energy consumption from their
networks
Concern by regulators over carbon emissions is starting to impact the
way some operators manage energy
18
The networks with the lowest energy consumption
typically share several of certain characteristics
Energy costs are managed aggressively by a person with relevant
expertise, typically at group level
High quality energy data is available
Electricity prices are high; diesel usage is minimised
Network equipment is relatively new
A degree of network sharing is occurring
Emerging country networks owned by a European operator are more
energy efficient than their competitors
19
We believe many mobile networks can reduce energy
usage by 15-20% with attractive paybacks
The benchmarking shows that after normalising for factors such as temperature, population density, etc., a typical operator would have to reduce its energy consumption by 15-20% to achieve top quartile performance
Not all operators have implemented “easy wins” such as free cooling, using latest generation of a/c equipment, reducing battery cooling, upgrading to more efficient rectifiers, using generator-battery hybrids
Managed energy contracts can circumvent capital constraints. In some countries attractive financing of energy efficiency projects is also available
Innovative solutions such as the dynamic matching of network radio resources with traffic demand can produce dramatic savings
20
Participants in MEE Benchmarking have submitted case
studies, which show the savings potential available
Telstra achieved AU$1.5 million in network energy savings by Installation of ‘economy cycle cooling fans’ Use of high temperature VRLA batteries Increasing the mobile equipment operational temperatures Replacing traditional room air conditioners
Telefónica Uruguay cut €4.2 million off its network energy bill with free cooling, saving 27% of indoor cell site energy consumption on average, and 25% at switch sites
Vodafone Romania found the most cost-effective free cooling solution to be an “economiser”, providing indirect free cooling to buildings which already have air conditioning systems
Digicel Jamaica cut its radio access network costs by 23% with new software, and without additional capex
For more information: www.gsma.com/publicpolicy/mobile-energy-
efficiency/mobile-energy-efficiency-resources/case-studies
21
KPI Comment
Energy per connection Simple KPI with much agreement on
consistent definition of “connection”
Energy per cell site Difficult to define “cell site” to enable like-
for-like comparison
Energy per unit traffic Theoretically good but hard to gather
necessary traffic data
Energy per revenue Too many other factors influence this KPI
but strong commercial implications
The best KPI now is energy per connection; energy per
traffic should be best when the data is higher quality
22
Prior to any normalisation there is a wide range across
all KPIs
Energy per connection kWh, 2010
Energy per cell site MWh, 2010
Energy per unit traffic kWh / GB, 2010
Source: GSMA Mobile Energy Efficiency Benchmarking
0
5
10
15
20
25
30
35
40
45
50
0
10
20
30
40
50
60
70
80
90
100
0
10
20
30
40
50
60
70
23
0
10
20
30
40
50
Max 25% Average 75% Min
0
10
20
30
40
50
Max 25% Average 75% Min
Pre-normalisation Energy per connection kWh, 2010
Post-normalisation Energy per connection kWh, 2010
Normalising for variables such as population
distribution and climate explains 70% of the variation
Source: GSMA Mobile Energy Efficiency Benchmarking
Wide
spread
Narrow
spread
24
Post-normalisation, the results for different KPIs are
consistent across different networks
Key High energy network Average energy network Low energy network
2013 Canada France India Italy Japan Mexico South
Africa
Pre-normalisation
kWh per
Connection High Low Low Low Low Avg High
kWh per
cell site Low Avg Avg Low High High Avg
kWh per
traffic Avg High High High Avg Avg High
Post-normalisation
kWh per
Connection Low Low High Low Avg High High
kWh per
Cell site Avg Low High Low Low High High
kWh per
traffic Avg Low High Low Avg High High
Summary of results by network
Illustrative
25
0
20
40
60
80
100
120
140
2009 2010 2011
0
5
10
15
20
25
30
2009 2010 2011
0
5
10
15
20
25
30
35
40
2009 2010 2011
119 121 25.5
23.4
33.6
19.1
Total industry RAN + Core
Energy
TWh
Total industry RAN + Core
Energy per connection
kWh / connection
Total industry RAN + Core
Energy per unit traffic
kWh / GB
+2% -5%
-19%
The benchmarking output enables us to track industry energy
usage over time
Note: This analysis is based on 65 networks where data for 2010 and 2011 was high quality, with the % changes applied to estimated
global totals and averages. The dataset was equally split between developed and emerging markets but not truly representative as the
growth in connections was lower than the true global total. The definition of energy here is total energy, without an adjustment for
diesel generator efficiency
Source: GSMA
Industry performance 2009-2011
Total network energy usage by the industry grew slowly in 2010 and 2011 but
decreased sharply on a unit basis
126
+4%
24.1
-3%
27.2
-30%
26
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Connections Energy Consumption
Energy Cost CO2
Emissions
South & Central America
East Europe
Middle East
West Europe
Africa
USA & Canada
Asia Pacific
5 bn 120 TWh $13 bn 70 Mt
The MEE methodology enables estimation of the industry’s total
network energy usage and carbon emissions, a global first
If all networks with above average energy consumption are improved to the industry
average, this implies a potential energy cost saving of $1 billion per annum at 2010
prices. Improving to the top quartile saves potentially over $2 billion annually
Source: GSMA. Note costs exclude the capital cost of diesel generators and their maintenance
2010:
Industry performance 2010
The benchmarking output assists in the measurement and management of industry
energy costs and carbon emissions
27
Agenda
Introduction to MEE services
MEE Benchmarking
MEE Optimisation
28
MEE Optimisation was launched in 2011 to help MNOs
identify and implement measures to reduce energy/carbon
Launch the MEE Optimisation (MEEO) service and start
pitching it to MNOs and vendors / system integrators
Deliver first MEEO project and publish case study
Form Technology Partner Group, a group of specialist
technology providers which delivers MEEO projects
Deliver further MEEO projects, increase participation and
encourage more technological innovation
(2011)
Ongoing
Objectives Status
(2011/12)
(2013)
29
Energy saving
measure
Capital cost
($m)
Annual saving
($m)
Payback
(Months)
Upgrade to high
efficiency rectifiers
Implement free air
cooling on cell sites
Install energy saving
features
Adjust air conditioning
temperature set points
Install smart meters &
establish monitoring
Etc.
MEE Optimisation identifies energy saving measures
and estimates the cost benefit for each measure
30
The first MEEO project with Telefónica Germany & NSN
identified €1.8m p.a. of savings & 9-30 month paybacks
Approach:
1. Assemble team of energy experts
from Telefónica Germany, NSN and
the GSMA
2. Choose 20 representative cell sites,
gather and analyse data, and
undertake site visits
3. Assess data gathered and produce
summary report
Results:
RAN savings identified: €1.8m p.a.
in high efficiency rectifiers and
software features
9 to 30 month paybacks
Much of the savings have now been
implemented
31
The second MEEO project with Warid Pakistan showed
US$6.2m of savings with 14-18 month paybacks
Approach:
1. Assemble team from Warid,
Cascadiant and the GSMA
2. Monitor 10 representative cell sites,
and on 4 sites deploy energy/carbon
saving equipment: GE Durathon
battery, Coolsure DC air-con, and
Ballard methanol fuel cell
3. Produce summary report
Results:
Trial showed energy savings of 30%
for the GE battery and 60% for A/C
Savings when rolled out network
wide: US$6.2m and 19,700 TCO2
14 to 18 month paybacks
Implementation has begun
32
More MEE Optimisation projects are being identified
and scoped out
One of the GSMA’s Technology Partners is currently scoping
out a MEE Optimisation project in Latin America
We are currently talking to several other mobile network
operators about MEE Optimisation projects in networks with
room for improvement in energy efficiency
33
Expertise in ESCo projects; distributed energy
generation and renewables solutions; energy
consulting and technical assistance; and energy
management services
Provides turnkey, off-grid or unreliable grid power
solutions for mobile network operators
Designs and manufactures innovative, outdoor
electronic equipment enclosure solutions, which
are thermally managed and modular
The GSMA has formed a group of Technology Partners to
lead MEE Optimisation projects
34
There are many different energy and carbon saving
solutions
For example:
• Dynamic matching of network radio resources with traffic demand
• Free cooling systems which use separate battery coolers
• Fuel cells either as back-up or prime power
Energy efficiency
• Increase free cooling
• Use latest generation of a/c equipment
• Reduce battery cooling and/or increase set point
by using temperature resistant batteries
• Upgrade to more efficient rectifiers
• Activate more energy saving features
Fuel mix
• Reduce diesel consumption, e.g. by
generator-battery hybrids or
renewables
Infrastructure
• Reduce number of indoor versus outdoor sites
• Share more sites (although this has wider
commercial implications)
Measurement
• Install (more) smart meters or sub-
meters to increase data accuracy
STANDARD SOLUTIONS
MORE INNOVATIVE SOLUTIONS
But more innovation is needed!
35
A typical MEEO project approach:
Assemble Team
Analyse Cell Sites
Energy Reduction Proposal
Financing Implement-
ation
Energy
experts from
- operator
- Technology
Provider
- GSMA
Choose 10-20
representative
cell sites
Gather and
analyse
required data
Install trial
equipment on
some sites
Site visits
Assess
information
Summarise
results
Recommend
solutions
Operator
provides
capex
Or sign a
managed
energy
contract
Evaluate
possible
financing
support
Implement
attractive
solutions
3 to 4 months
36
Any questions?
37
22 March 2011, Vice-President of the European Commission Neelie Kroes
on the GSMA’s MEE Benchmarking service:
"...it’s great to see the Mobile sector’s Green Manifesto getting some real teeth
today with 17 new recruits signing up to the GSM Association’s Mobile Energy
Efficiency Network Benchmarking Service...“
For more information on the GSMA’s MEE services,
please talk to me or email me on
www.gsma.com/mee
38
APPENDIX
39
Brief explanation of regression analysis (1)
Source: GSMA
Regression analysis mathematically models the relationship between a dependent
variable (e.g. energy per connection, energy per cell site or energy per unit traffic) and
one or more independent variables
– For RAN energy per connection the independent variables are Number of cell sites per
connection, Number of cooling degree days per capita, Voice traffic per connection and Data
traffic per connection
The regression analysis produces a set of results which enable a mathematic equation
to be written to explain the relationship. The equation for energy per connection in the
analysis run for this presentation is:
RAN energy per connection = 3 + 10X1 + 0.001X2 + 3X3 + 1X4
where X1 is number of cell sites per connection, X2 is number of cooling degree
days, X3 is voice traffic per connection and X4 is data traffic per connection
With the equation, we can calculate the theoretical energy per connection for a
network, using the network’s values for each of the independent variables. Subtracting
the network’s theoretical value from the actual value gives a measure in kWh per
connection of whether the network is over or under-performing versus the theoretical
value. This approach can be extended to multiple networks
Therefore the effect of differing values of independent variables for multiple networks
can be removed, and so networks can be compared like-for-like
40
Brief explanation of regression analysis (2)
The regression analysis also produces statistics, which show amongst other things:
– How well the equation fits the data points: this is denoted by the coefficient of determination R2 which measures how much of the variation in the dependent variable can be explained by the independent variables
– E.g. an R2 of 55% means that approximately 55% of the variation in the dependent variable can be explained by the independent variable
– The remaining 45% can be explained by unknown variables or inherent variability
– The probability that the coefficient of the independent variable is zero, i.e. that the independent variable is useful in explaining the variation in the dependent variable. These probabilities are given by the P-values. A P-value of 3% for the coefficient of the independent variable ‘voice traffic per connection’ means that this coefficient (value +3) has a 3% chance of being zero, i.e. a 3% chance that this independent variable is not useful in explaining the variation in the dependent variable
As the dataset increases we hope to provide a higher R2 and lower P-values, and also to be able to include additional independent variables
Note that regression analysis does not prove causality but instead demonstrates correlation (i.e. that a relationship exists between the dependent and independent variables), and also that we are assuming a linear relationship over the ranges of variables covered in this analysis
Sensitivity analysis is conducted in two ways: running regressions with slightly different independent variables; and re-running the regressions with subsets of the dataset (e.g. developed vs. emerging countries) Source: GSMA
