Recording at: https://youtu.be/cXWOE7RDO6M
Recent works in renewable energy sources (RES) forecasting, have shown the interest of using spatially distributed time series and assumed that data could be gathered centrally and used, either at the RES power plant level, or at the level of a system operator. However, data is distributed in terms of ownership, limitation in data transfer capabilities and with agents being reluctant to share their data anyway.
Take control of your SAP testing with UiPath Test Suite
Extracting value from data sharing for RES forecasting: Privacy aspects & data monetization
1. Extracting value from data sharing
for RES forecasting
Privacy aspects & data monetization
Ricardo Jorge Bessa, INESC TEC
Liyang Han, DTU
December 2020
ISGAN Academy webinar #25
Recorded webinars available at: https://www.iea-isgan.org/our-work/annex-8/
2. ISGAN in a Nutshell
Created under the auspices of:
the Implementing
Agreement for a
Co-operative
Programme on Smart
Grids
ISGAN IN A NUTSHELL 2
Strategic platform to support high-level government
knowledge transfer and action for the accelerated
development and deployment of smarter, cleaner
electricity grids around the world
International Smart Grid Action Network is
the only global government-to-
government forum on smart grids.
an initiative of the
Clean Energy
Ministerial (CEM)
Annexes
Annex 1
Global
Smart Grid
Inventory Annex 2
Smart Grid
Case
Studies
Annex 3
Benefit-
Cost
Analyses
and
Toolkits
Annex 4
Synthesis
of Insights
for
Decision
Makers
Annex 5
Smart Grid
Internation
al
Research
Facility
Network
Annex 6
Power
T&D
Systems
Annex 7
Smart Grids
Transitions
Annex 8:
ISGAN
Academy
on Smart
Grids
5. Agenda
• Smart4RES in a nutshell
• Motivation for data sharing &
collaborative analytics
• Collaborative learning for RES
forecasting
• Data markets : Basics and
Smart4RES proposal
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 5
6. Smart4RES in a nutshell
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 6
7. Smart4RES in a nutshell
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 7
Smart4RES vision
Achieve outstanding improvement in RES predictability
through a holistic approach, that covers the whole model
and value chain related to RES forecasting
Improvement from collaborative RES forecasting
Potential for improvement with spatial-temporal
approaches up to 20% for 6h ahead for solar energy and
up to 15-20% for wind energy
Lack of privacy guarantees and price-based incentives to share data
8. Smart4RES consortium
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 8
6 countries
12 partners
11/2019-4/2023
Funds: H2020
programme
Budget: 4 Mio€
Duration: 3.5 years
End-users
Universities
Research
Industry
Meteorologists
9. Motivation for data sharing &
collaborative analytics
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 9
10. Data Sharing: Motivation
• Increasing volume of geographically distributed data
• Improvement in forecasting accuracy by this data
• Main barriers
• Data privacy and confidentiality
• Lack of economic signals for sharing (collaborating with) data
• Lack of business cases for collaborative analytics
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 10
11. Collaborative Analytics
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 11
Model
from centralized to distributed learning
Master Node
Model
Worker Node
ΔW1
Aggr(ΔW1+ΔW2+ ΔW3)
Worker Node
ΔW2
Worker Node
ΔW3
Federated
learning as
industry standard
• Exchange raw data → NO
• Exchange model coefficients → NO
(partially)
• Exchange model outputs → YES
• Data divided by features (not by
observations)
DATA SHARING IN
12. Possible Use cases for Data Sharing
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 12
RES Forecasting
Benefit: Improve forecasting skill in minutes to day-ahead
time horizon & exploit heterogenous data sources
Weather Modelling
Benefit: Liberalization of weather data trading → access to
large-scale weather data
Privacy &
monetization
Monetization
Source: Data Basin
How to price
weather
data?
13. Possible Use cases for Data Sharing
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 13
Peer-to-peer (P2P) Trading
Benefit: Secure analytics with personal data → better
decision-making & more trust
Power Transformer Condition
Benefit: Data augmentation (faults, dissolved gas analysis,
sensors) → improved maintenance policies
Utility A
Utility C
Utility B
€
€
enable sharing of energy
transactions and usage data
14. Collaborative learning for RES
forecasting
1/8/2018 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 14
15. RES Collaborative Forecasting
Formulation
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 15
Vector Autoregressive
Model (VAR)
𝒀 = 𝒄 + 𝑩𝒁 + 𝑬
Z: lagged power
observations
1
Y1,tY1,t-1Y1,t-2
…
2
Yt,2Y2,t-1Y2,t-2
…
N
YN,tYN,t-1YN,t-2
…
Example for
2 PV sites 𝑌1,𝑡 𝑌2,𝑡 = 𝑐1 𝑐2 +
𝐵1,1
1 𝐵1,2
1 𝐵1,1
2 𝐵1,2
2
𝐵2,1
1
𝐵2,2
1
𝐵2,1
2
𝐵2,2
2 ∙
𝑌1,𝑡−1
𝑌2,𝑡−1
𝑌1,𝑡−2
𝑌2,𝑡−2
+ 𝐸1,𝑡 𝐸2,𝑡
B: coefficients matrix
(to estimate)
constant terms
(to estimate)
white noiseforecasted power
multivariate linear model
power forecasts for multiple sites as a function
of past power observations from all sites
16. Iterative estimation
RES Collaborative Forecasting
State-of-the-art distributed learning
16
Distributed
coefficients
estimation
= B2=
𝐶𝑜𝑒𝑓 𝑊𝐹1
𝑘+1
= 𝑔(
𝐶𝑜𝑒𝑓 𝑊𝐹2
𝑘+1
= 𝑔(
𝐶𝑜𝑒𝑓 𝑊𝐹3
𝑘+1
= 𝑔(
)
)
)
, ,
, ,
, ,
1
2
3
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
Coef
WF1
Coef
WF2
Coef
WF3
Power for
multiple
locations
lagged
power
observations
Coefficients
Limitation: access to
private/confidential data
Conciliation
k
Conciliation
k
Conciliation
k
.
C. Gonçalves, R.J. Bessa, P. Pinson, “A critical overview of privacy-
preserving approaches for collaborative forecasting,” International Journal of
Forecasting, vol. 37, no. 1, pp. 322-342, Jan-Mar 2021.
Conciliation
k
data
from
WF1
𝑪𝒐𝒆𝒇
𝑾𝑭 𝟏
𝒌 +
data
from
WF2
𝑪𝒐𝒆𝒇
𝑾𝑭 𝟐
𝒌
data
from
WF3
𝑪𝒐𝒆𝒇
𝑾𝑭 𝟑
𝒌+= ℎ
Power
WF1Power
WF2Power
,
Forecast after k iterations
. . .
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
Distributed learning with ADMM (Alternating Direction Method of Multipliers)
17. RES Collaborative Forecasting
Privacy-preserving Protocol
17
==
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
Power for
multiple
locations
Lagged power
observations
Coefficients
(1) Power data encryption with linear algebra (M:
random matrix – unknown but built by all agents)
PowerWF1
Private matrices
M1 M2 M3. . .
Coef
WF1
Coef
WF2
Coef
WF3
.
Encrypted
data
==
PowerWF1
PowerWF2
PowerWF3
data
from
WF2
Enc
data
from
WF3
EncPWF1
EncPWF2
EncPWF3
Enc
data
from
WF1
Enc
data
from
WF2
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
.
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
18. RES Collaborative Forecasting
Privacy-preserving Protocol
18
Encrypted
data
==
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
Power for
multiple
locations
Lagged power
observations
==
PowerWF1
PowerWF2
PowerWF3
data
from
WF2
Enc
data
from
WF3
EncPWF1
EncPWF2
EncPWF3
Enc
data
from
WF1
Enc
data
from
WF2
M1 M2 M3
PowerWF2
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
Coef
WF1
Coef
WF2
Coef
WF3
. .
. . .
Coefficients
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
(1) Power data encryption with linear algebra (M:
random matrix – unknown but built by all agents)
19. RES Collaborative Forecasting
Privacy-preserving Protocol
19
Encrypted
data
==
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
Power for
multiple
locations
Lagged power
observations
==
PowerWF1
PowerWF2
PowerWF3
data
from
WF2
Enc
data
from
WF3
EncPWF1
EncPWF2
EncPWF3
Enc
data
from
WF1
Enc
data
from
WF2
PowerWF1
Q1
Private matrix for encrypting
coefficients from WF1
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
Coef
WF1
Coef
WF2
Coef
WF3
.
M1 M2 M3. . . .
.
Coefficients
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
(2) Coefficients encryption with linear algebra (Q:
random matrix – own by each agent)
20. RES Collaborative Forecasting
Privacy-preserving Protocol
20
Encrypted
data
==
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
Power for
multiple
locations
Lagged power
observations
==
PowerWF1
PowerWF2
PowerWF3
data
from
WF2
Enc
data
from
WF3
EncPWF1
EncPWF2
EncPWF3
Enc
data
from
WF1
Enc
data
from
WF2
Q2
PowerWF2
Private matrix for encrypting
coefficients from WF2
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
Coef
WF1
Coef
WF2
Coef
WF3
M1 M2 M3. . . .
..
Coefficients
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
(2) Coefficients encryption with linear algebra (Q:
random matrix – own by each agent)
21. RES Collaborative Forecasting
Privacy-preserving Protocol
21
Encrypted
data
Distributed
computation
==
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
Power for
multiple
locations
Lagged power
observations
==
PowerWF1
PowerWF2
PowerWF3
EncPWF1
EncPWF2
EncPWF3
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
Enc
data
from
WF3
Enc
data
from
WF1
Enc
data
from
WF2
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
Coef
WF1
Coef
WF2
Coef
WF3
..
Coefficients
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
(3) Distributed computation of coefficients with ADMM
22. RES Collaborative Forecasting
Privacy-preserving Protocol
22
Encrypted
data
Distributed
computation
==
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
data
from
WF3
PowerWF1
PowerWF2
PowerWF3
data
from
WF1
data
from
WF2
Power for
multiple
locations
Lagged power
observations
==
PowerWF1
PowerWF2
PowerWF3
EncPWF1
EncPWF2
EncPWF3
Original coefficients
Enc
Coef
WF2
Enc
Coef
WF3
Q1
Q2
Q3
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
Enc
data
from
WF3
Enc
data
from
WF1
Enc
data
from
WF2
Enc
Coef
WF1
Enc
Coef
WF1
Enc
Coef
WF2
Enc
Coef
WF3
Coef
WF1
Coef
WF2
Coef
WF3
..
.
.
.
Coefficients
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
(4) Obtain original coefficients with Q matrix
(same coefficients with privacy protocol)
23. RES Collaborative Forecasting
Communication Schemes
23
Agents (RES Producers)
Central node (hub)
Centralized
Model
Peer-to-Peer
Model
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
24. RES Collaborative Forecasting
Communication Schemes
24
Step 1.
Agents compute their encrypted coefficients
𝐸𝑛𝑐 𝐶𝑜𝑒𝑓 𝑗
𝑘+1
= 𝑔( ), ,
Conciliation
k
EncPj
Enc
data
from
agent
𝒋
Initialized with zeros
Centralized
Model
Peer-to-Peer
Model
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
25. RES Collaborative Forecasting
Communication Schemes
25
Centralized
Model
Peer-to-Peer
Model
Step 2.
Agents share their encrypted contribution
𝑬𝒏𝒄
𝑪𝒐𝒆𝒇
𝑾𝑭𝒋
𝒌
Enc
data
from
agent
𝒋
.
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
26.
𝑗
RES Collaborative Forecasting
Communication Schemes
26
Centralized
Model
Peer-to-Peer
Model
Step 3.
Computation of
Conciliation
k+1
, …
𝑬 𝒏𝒄
𝑪𝒐𝒆𝒇
𝑾𝑭𝒋
𝒌
Enc
data
from
agent
𝒋
= ℎ .
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
27. RES Collaborative Forecasting
Communication Schemes
27
Centralized
Model
Peer-to-Peer
Model
Step 4.
Agents obtain conciliation matrix and proceed
to Step 1
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING
28. RES Collaborative Forecasting
Results for Évora PV Dataset
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 28
Asynchronous communication: equal failure probabilities are assumed for all agents
❑ Better performance of the P2P scheme
▪ Centralized: if one agent fails the algorithm proceeds without its information
▪ P2P: agent communicates its contribution to some peers → probability of information lost is smaller
❑ Computational performance
▪ Privacy protocol: 65.5s
▪ 0.05s (centralized) and 0.12s (P2P) for model fitting
ÉVORA
44 Domestic PV
29. RES Collaborative Forecasting
Results for Évora PV Dataset
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 29
RMSE improvement of Privacy-VAR over AR (autoregressive) & Analogs search (collaborative w/ privacy)
Some data owners contribute to improve competitors’ forecast without getting the same benefit (error improvement)
↓
Even if privacy is ensured, such agents may be unwilling to collaborate → data monetization (data markets)
31. Data Market Basics
• Market components
• Key players
• Data buyer(s)
• Data seller(s)
• Monetary Values
• Seller’s cost of offering data
• Buyer’s profit
• Data payment
• Market procedure
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 31
32. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s)
• Monetary Values
• Seller’s cost of offering data
• Buyer’s profit 𝐹(𝐷)
• Data payment
• Market procedure
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 32
B
𝐷
𝐹(𝐷)
Buyer
profit
0
33. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data
• Buyer’s profit 𝐹(𝐷)
• Data payment
• Market procedure
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 33
BS
∆𝐷 𝐷
𝐹(𝐷)
profit
BuyerSeller
0
34. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data
• Buyer’s profit 𝐹 𝐷
• Data payment
• Market procedure
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 34
BS
∆𝐷 𝐷
𝐹(𝐷)
profit
BuyerSeller
0
35. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data
• Buyer’s profit 𝐹(𝐷) < 𝐹(𝐷 + ∆𝐷)
• Data payment
• Market procedure
1) Buyer can profit from seller’s data
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 35
BS
∆𝐷 𝐷
𝐹(𝐷)
𝐹 𝐷 + ∆𝐷
∆𝐹
profit
BuyerSeller
0 *Note: the use of data is
not necessarily volumetric!
36. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data C(∆𝐷)
• Buyer’s profit 𝐹(𝐷) < 𝐹(𝐷 + ∆𝐷)
• Data payment
• Market procedure
1) Buyer can profit from seller’s data
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 36
BS
𝐶
∆𝐷 𝐷
𝐹(𝐷)
𝐹 𝐷 + ∆𝐷 *
∆𝐹
𝐶
profit
BuyerSeller
0 *Note: the use of data is
not necessarily volumetric!
37. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data C(∆𝐷)
• Buyer’s profit 𝐹(𝐷) < 𝐹(𝐷 + ∆𝐷)
• Data payment 𝑅 < ∆𝐹 = 𝐹 𝐷 + ∆𝐷 − 𝐹(𝐷)
• Market procedure
1) Buyer can profit from seller’s data
2) Buyer offers seller monetary rewards
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 37
BS
𝑅
𝐶
∆𝐷 𝐷
𝐹(𝐷)
𝐹 𝐷 + ∆𝐷 *
∆𝐹
𝑅
𝐶
profit
BuyerSeller
0
?
*Note: the use of data is
not necessarily volumetric!
38. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data C(∆𝐷)
• Buyer’s profit 𝐹(𝐷) < 𝐹(𝐷 + ∆𝐷)
• Data payment 𝑅 < ∆𝐹 = 𝐹 𝐷 + ∆𝐷 − 𝐹(𝐷)
• Market procedure
1) Buyer can profit from seller’s data
2) Buyer offers seller monetary rewards
3) Seller either rejects or accepts offer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 38
BS
𝑅
𝐶
∆𝐷
If R < 𝐶
×
𝐷
𝐹(𝐷)
𝐹 𝐷 + ∆𝐷 *
∆𝐹
𝑅
𝐶
𝑅
Reject offer
×profit
BuyerSeller
0 *Note: the use of data is
not necessarily volumetric!
39. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data C(∆𝐷) < 𝑅
• Buyer’s profit 𝐹(𝐷) < 𝐹(𝐷 + ∆𝐷)
• Data payment 𝑅 < ∆𝐹 = 𝐹 𝐷 + ∆𝐷 − 𝐹(𝐷)
• Market procedure
1) Buyer can profit from seller’s data
2) Buyer offers seller monetary rewards
3) Seller either rejects or accepts offer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 39
BS
𝑅
𝐶
Requirement: 𝐶 < 𝑅
∆𝐷
√
𝐷
𝐹(𝐷)
𝐹 𝐷 + ∆𝐷 *
∆𝐹
𝑅
𝑅𝐶
Accept offer
√
profit
BuyerSeller
0 *Note: the use of data is
not necessarily volumetric!
40. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data C(∆𝐷) < 𝑅
• Buyer’s profit 𝐹(𝐷) < 𝐹(𝐷 + ∆𝐷)
• Data payment 𝑅 < ∆𝐹 = 𝐹 𝐷 + ∆𝐷 − 𝐹(𝐷)
• Market procedure
1) Buyer can profit from seller’s data
2) Buyer offers seller monetary rewards
3) Seller either rejects or accepts offer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 40
BS
𝑅
𝐶
Requirement: 𝐶 < 𝑅
∆𝐷
(𝐶 includes the value of privacy)
√
𝐷
𝐹(𝐷)
𝐹 𝐷 + ∆𝐷 *
∆𝐹
𝑅
𝑅𝐶
Accept offer
√
profit
BuyerSeller
0 *Note: the use of data is
not necessarily volumetric!
41. Data Market Basics
• Market components
• Key players
• Data buyer(s) B with known data 𝐷
• Data seller(s) S with data ∆𝐷
• Monetary Values
• Seller’s cost of offering data 𝐂(∆𝑫) < 𝑹
• Buyer’s profit 𝑭(𝑫) < 𝑭(𝑫 + ∆𝑫)
• Data payment 𝑹 < ∆𝑭 = 𝑭 𝑫 + ∆𝑫 − 𝑭(𝑫)
• Market procedure
1) Buyer can profit from seller’s data
2) Buyer offers seller monetary rewards
3) Seller either rejects or accepts offer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 41
BS
𝑅
𝐶
Requirement: 𝐶 < 𝑅
∆𝐷 𝐷
𝐹(𝐷)
𝐹 𝐷 + ∆𝐷 *
∆𝐹
𝑅
𝐶 𝑅
Seller Payoff
= 𝑅 − 𝐶 Buyer Payoff
= ∆𝐹 − 𝑅
profit
Seller Buyer
(𝐶 includes the value of privacy)
0 *Note: the use of data is
not necessarily volumetric!
42. B
Data Market Models
BS
B
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 42
𝑅
∆𝐷
Data Seller(s) Data Buyer(s)
Data Market Models:
S
S
𝐶
𝐷
MoneyData
43. B
Data Market Models
BS
B
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 43
Data Market Models:
• Monopolistic Data Seller
Data Seller(s) Data Buyer(s)
MoneyData
44. Data Market Models
BS
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 44
Data Seller(s) Data Buyer(s)
S
S
Data Market Models:
• Monopolistic Data Seller
• Monopolistic Data Buyer
MoneyData
45. B
Data Market Models
B
S
S
S
B
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 45
Data Market Models:
• Monopolistic Data Seller
• Monopolistic Data Buyer
• Peer-to-Peer Multi-Seller Multi-Buyer
Data Seller(s) Data Buyer(s)
MoneyData
46. B
Data Market Models
B
S
S
S
B
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 46
Data Market Models:
• Monopolistic Data Seller
• Monopolistic Data Buyer
• Peer-to-Peer Multi-Seller Multi-Buyer
• Centralized Model
Data Seller(s) Data Buyer(s)
MoneyData
Central
Data
Platform
47. B
Data Market Models
B
S
S
S
B
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 47
Data Market Models:
• Monopolistic Data Seller
• Monopolistic Data Buyer
• Peer-to-Peer Multi-Seller Multi-Buyer
• Centralized Model
Data Seller(s) Data Buyer(s)
MoneyData
Central
Data
Platform
How can this
be applied to
Smart4RES
in an energy
market?
48. B
Data Market Models
B
S
S
S
B
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 48
Data Market Models:
• Monopolistic Data Seller
• Monopolistic Data Buyer
• Peer-to-Peer Multi-Seller Multi-Buyer
• Centralized Model
Data Seller(s) Data Buyer(s)
MoneyData
Central
Data
Platform
Similar to an energy
wholesale market!
How can this
be applied to
Smart4RES
in an energy
market?
55. R
Energy-Data Market
R
C
C
C
G
G
G
Generators End-Customers
Electric
Energy
Pool
R
Retailers
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 55
MoneyData
Energy
e.g., customer load
forecasting
e.g., RES
forecasting
Reduce cost of real-time
energy imbalance
0
Gen(MW)
Fore
-cast
Real
-time
𝐺
𝐺
Generator
Imbalance Cost
= priceimb × ∆𝐺
∆𝐺
0
Load(MW)
Fore
-cast
Real
-time
𝐿
𝐿
Retailer
Imbalance Cost
= priceimb × ∆𝐿
∆𝐿
57. R
Energy-Data Market
R
G
G
G
Generators End-Customers
Gen-Side
Data Pool
Electric
Energy
Pool
R
Retailers
Data Seller Data Buyer Data Seller/Buyer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 57
Data + Money
MoneyData
Energy
C
C
C
Con-Side
Data Pool
X
Privacy
Concerns
e.g., customer load
forecasting
e.g., RES
forecasting
58. R
Re.g., RES
forecasting
Energy-Data Market
G
G
G
Generators
Gen-Side
Data Pool
Data Seller Data Buyer Data Seller/Buyer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 58
Data + Money
MoneyData
Energy
R
Electric
Energy
Pool
End-Customers
Retailers
C
C
C
Con-Side
Data Pool
X
e.g., customer load
forecasting
Privacy
Concerns
Con-Side Data Market
59. R
Re.g., RES
forecasting
Energy-Data Market
G
G
G
Generators
Gen-Side
Data Pool
Data Seller Data Buyer Data Seller/Buyer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 59
Data + Money
MoneyData
Energy
R
Electric
Energy
Pool
End-Customers
Retailers
C
C
C
Con-Side
Data Pool
X
e.g., customer load
forecasting
Privacy
Concerns
Con-Side Data Market
L. Han, J. Kazempour, P. Pinson,
“Monetizing Customer Load Data for
an Energy Retailer: A Cooperative
Game Approach,” arXiv e-prints, 2020.
60. e.g., customer load
forecasting
R
Energy-Data Market
R
R
Data Seller Data Buyer Data Seller/Buyer
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 60
Data + Money
MoneyData
Energy
End-Customers
Retailers
C
C
C
Con-Side
Data Pool
X
Privacy
Concerns
Electric
Energy
Pool
G
G
G
Generators
Gen-Side
Data Pool
e.g., RES
forecasting
Gen-Side Data Market
61. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 61
62. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 62
63. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 63
Problem 1. RES
generation
uncertainty leads to
imbalance costs
64. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 64
Problem 1. RES
generation
uncertainty leads to
imbalance costs
65. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 65
Problem 1. RES
generation
uncertainty leads to
imbalance costs
Step 1) Data from
neighboring plants can help
improve forecasts,
reducing imbalance costs
66. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 66
Problem 1. RES
generation
uncertainty leads to
imbalance costs
Step 1) Data from
neighboring plants can help
improve forecasts,
reducing imbalance costs
Problem 2. Potential
loss of privacy and
competitiveness
67. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 67
Problem 1. RES
generation
uncertainty leads to
imbalance costs
Step 1) Data from
neighboring plants can help
improve forecasts,
reducing imbalance costs
Problem 2. Potential
loss of privacy and
competitiveness
Step 2). Buyer
offers data
payment
68. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 68
Step 2). Buyer
offers data
payment
Problem 2. Potential
loss of privacy and
competitiveness
Problem 1. RES
generation
uncertainty leads to
imbalance costs
Step 3). Seller
accepts
payment
Step 1) Data from
neighboring plants can help
improve forecasts,
reducing imbalance costs
69. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 69
Outcome 1. Higher
profit for data
buyer, lower cost
for energy supply
Problem 1. RES
generation
uncertainty leads to
imbalance costs
Step 1) Data from
neighboring plants can help
improve forecasts,
reducing imbalance costs
Outcome 2. Higher
profit for data seller
Step 2). Buyer
offers data
payment
Problem 2. Potential
loss of privacy and
competitiveness
Step 3). Seller
accepts
payment
70. RES in a Wholesale Market: 2 Agents
Electric
Energy
Pool
Outcome 2. Higher
profit for data seller
Outcome 1. Higher
profit for data
buyer, lower cost
for energy supply
HIGHER TOTAL
SOCIAL WELFARE
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 70
Problem 1. RES
generation
uncertainty leads to
imbalance costs
Step 2). Buyer
offers data
payment
Problem 2. Potential
loss of privacy and
competitiveness
Step 3). Seller
accepts
payment
Step 1) Data from
neighboring plants can help
improve forecasts,
reducing imbalance costs
71. RES in a Wholesale Market: Many Agents
Electric
Energy
Pool
...
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 71
72. RES in a Wholesale Market: Many Agents
Electric
Energy
Pool
...
Payment
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 72
Architecture 1:
Peer-to-Peer Model
73. RES in a Wholesale Market: Many Agents
Electric
Energy
Pool
...
Data
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 73
Architecture 2:
Centralized Model
74. RES in a Wholesale Market: Many Agents
Electric
Energy
Pool
...
Data
Pool
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 74
Architecture 2:
Centralized Model
C. Gonçalves, P. Pinson, R.J. Bessa,
“Towards data markets in renewable
energy forecasting,” IEEE Trans. on
Sust. Energy, In Press, 2020
75. • The cost of privacy is highly individual and difficult to quantify. As a result, the
value of privacy preserving techniques is difficult to quantify as well.
• Data is a unique commodity. The table below compares data and energy*.
Challenges
Market Production and Replication Value to Buyer Pricing
Energy
Produced at a certain cost, non-
replicable
Additive and known Decided a priori
Data
Usually a side-product that is
produced at zero marginal
costs, Replicable at no extra
costs
Combinatorial: the value of a
dataset is dependent on all other
available data.
Dependent on buyer’s
valuation of the dataset with a
certain prediction task
*Concepts from publication:
A. Agarwal, M. Dahleh, and T. Sarkar, “A marketplace for data: An algorithmic solution,” ACM EC 2019 - Proceedings of the 2019
ACM Conference on Economics and Computation, pp. 701–726, 2019.
17/12/2020 EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 75
77. Take-Away Messages &
Smart4RES Ongoing Research
• Collaborative learning improves forecast accuracy, which may yield additional
individual or societal value in the market.
• Monetizing data promotes data exchanging by redistributing the added value, helping to
address concerns about loss of privacy and competitiveness.
Smart4RES is planning to
• Design a suitable marketplace for data trading;
• Develop relevant data market concepts and create prototypes to foster awareness to
the value of data markets;
• Extend the concept to different use cases from the energy sector;
• Collaborate with other domains, such as IoT, blockchain technologies, etc.
77EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING17/12/2020
78. References
• C. Gonçalves, P. Pinson, R.J. Bessa, “Towards data markets in renewable energy forecasting,” IEEE Trans. on
Sust. Energy, In Press, 2020.
• C. Gonçalves, R.J. Bessa, P. Pinson, "Privacy-preserving distributed learning for renewable energy forecasting,"
Working Paper, 2020.
• C. Gonçalves, R.J. Bessa, P. Pinson, “A critical overview of privacy-preserving approaches for collaborative
forecasting,” Int. J. of Forecasting, vol. 37(1), pp. 322-342, Jan-Mar 2021
• L. Han, J. Kazempour, P. Pinson, “Monetizing Customer Load Data for an Energy Retailer: A Cooperative Game
Approach,” arXiv e-prints, 2020.
• A. Agarwal, M. Dahleh, and T. Sarkar, “A marketplace for data: An algorithmic solution,” ACM EC 2019 -
Proceedings of the 2019 ACM Conference on Economics and Computation, pp. 701–726, 2019.
• BESSA, Ricardo Jorge, GONÇALVES, Carla. “Geographically distributed solar power time series [dataset].” 10
September 2020. INESC TEC research data repository.
https://doi.org/10.25747/gywm-9457
78EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING17/12/2020
79. Smart4RES webinar series
EXTRACTING VALUE FROM DATA SHARING FOR RES FORECASTING 79
Season1: Towards a new Standard for the entire RES forecasting value chain
17/12/2020
To stay in touch
https://bit.ly/2Ta
wgn3
80. Thank you
Ricardo Jorge Bessa, INESC TEC ricardo.j.bessa@inesctec.pt
Liyang Han, DTU liyha@elektro.dtu.dk
info@smart4res.eu
For more information www.smart4res.eu
Follow us on our social media
81. COPYRIGHT (except initial slides on ISGAN)
Smart4RES H2020 Project Next Generation Modelling and Forecasting of Variable Renewable
Generation for Large-scale Integration in Energy Systems and Markets
This project has received funding from the European Union’s Horizon 2020 research and innovation
programme under grant agreement No 864337.
DISCLAIMER
The present presentation reflects only the authors’ view. The European Commission or the European
Innovation and Networks Executive Agency (INEA) is not responsible for any use that may be made of the
information it contains.
PROJECT COORDINATOR & CONTACTS
Georges Kariniotakis, ARMINES/MINES Paris Tech, Centre PERSEE, Sophia-Antipolis France.
info@smart4res.eu