This document discusses addressing demand uncertainty in long-term energy planning models. It compares deterministic models using planning reserve margins to stochastic models that capture expected operational costs under different demand scenarios. Capturing multiple demand scenarios changes the optimal capacity mix by accounting for the expected costs of meeting variable demand. Stochastic models also endogenously assess renewable energy capacity credits over multiple time periods rather than fixing credits based on a single peak period. Accounting properly for demand uncertainty and renewable intermittency provides more robust optimal capacity planning outcomes.

1. Addressing demand uncertainty in long-term planning
models
Tim Mertens
ETSAP workshop
17/06/2018

2. 229/06/2018
Demand uncertainty - Why could it be relevant?
Unexpected demand growth cannot be met by sudden
capacity changes.
• Technology lead times make that investment decisions
are made years before the moment of operation.
Optimal investments decisions are based on future
operational cost (OPEX) estimation.
• OPEX estimation depends on the demand (screening
curves).
Risk of ending up with a suboptimal capacity mix to
cover electricity demand.
• How can we deal with demand uncertainty in long-
term planning models?

3. 329/06/2018
Planning reserve margins
Safety margin that ensures an amount of reserve
capacity to be in place.
• Planning reserve margin forces the total installed
capacity to exceed the perceived peak demand.
• Hedge against being inadequate and the
deployment of emergency measures.
Drawback:
• No estimation of the expected OPEX associated with
an uncertain demand is taken into account.
How important is it to capture this expected OPEX?
𝐷 𝑝𝑒𝑎𝑘
(1 + 𝑃𝑅𝑀) ∙ 𝐷 𝑝𝑒𝑎𝑘
Nuclear
Coal
CCGT
OCGT
RES
Load

4. 429/06/2018
Methodology
Two planning model comparisons:
1. Effect of capturing expected OPEX
2. Effect of capturing expected OPEX + endogenous RES capacity credit assessment

5. 529/06/2018
Deterministic model with planning reserve constraint
𝑀𝑖𝑛
𝑔∈𝐺
𝑓𝑔 ∙ 𝐾𝑔 +
𝑡∈𝑇 𝑔∈𝐺
𝑐 𝑔 ∙ 𝑄 𝑔,𝑡 + 𝑉𝑂𝐿𝐿 ∙ 𝐸𝑁𝑆𝑡
s.t.:
• 𝑔∈𝐺 𝑄 𝑔,𝑡 = 𝐷𝑡 − 𝐸𝑁𝑆𝑡 ∀𝑡 ∈ 𝑇
• 𝑄 𝑔,𝑡 ≤ 𝐾𝑔 ∀𝑔 ∈ 𝐺 𝐷𝑖𝑠𝑝, ∀𝑡 ∈ 𝑇
• 𝑄 𝑔,𝑡 ≤ 𝑅𝑃𝑔,𝑡 ∙ 𝐾𝑔 ∀𝑔 ∈ 𝐺 𝑅𝐸𝑆, ∀𝑡 ∈ 𝑇
• 𝑔∈𝐺 𝐷𝑖𝑠𝑝
𝐾𝑔 + 𝑔∈𝐺 𝑅𝐸𝑆
𝑅𝑃𝑔,𝑡 ∙ 𝐾𝑔 ≥ (1 + 𝑃𝑅𝑀) ∙ 𝐷𝑡 ∀𝑡 ∈ 𝑇
Operational cost estimation based on single demand scenario.
PRC ensures sufficient capacity to cover demand.
• No load shedding in optimal solution.
• RES capacity credit depends on production profile in peak residual demand timestep
Fixed costs Operational costs

6. 629/06/2018
A stochastic planning model
𝑀𝑖𝑛
𝑔∈𝐺
𝑓𝑔 ∙ 𝐾𝑔 +
𝜔∈Ω
𝑝 𝜔 ∙
𝑡∈𝑇 𝑔∈𝐺
𝑐 𝑔 ∙ 𝑄 𝑔,𝑡,𝜔 + 𝑉𝑂𝐿𝐿 ∙ 𝐸𝑁𝑆𝑡,𝜔
s.t.:
• 𝑔∈𝐺 𝑄 𝑔,𝑡,𝜔 = 𝐷𝑡,𝜔 − 𝐸𝑁𝑆𝑡,𝜔 ∀𝑡 ∈ 𝑇, ∀𝜔 ∈ Ω
• 𝑄 𝑔,𝑡,𝜔 ≤ 𝐾𝑔 ∀𝑔 ∈ 𝐺 𝐷𝑖𝑠𝑝, ∀𝑡 ∈ 𝑇, ∀𝜔 ∈ Ω
• 𝑄 𝑔,𝑡,𝜔 ≤ 𝑅𝑃𝑔,𝑡 ∙ 𝐾𝑔 ∀𝑔 ∈ 𝐺 𝑅𝐸𝑆, ∀𝑡 ∈ 𝑇, ∀𝜔 ∈ Ω
Fixed costs Expected operational costs
1. No load shedding:
• RES capacity credit depends on production profile in peak residual demand timestep
2. Load shedding
• Endogenous assessment of capacity credit for RES over multiple timesteps
• Depends on value of lost load (VOLL)

7. 729/06/2018
Case study overview
Deterministic – PRC
(reference)
Stochastic - no ENS Stochastic - ENS
Low
dispatchable
CAPEX
D-low S-NoENS-low S-ENS-low
High
dispatchable
CAPEX
D-high S-NoENS-high S-ENS-high
𝑫 𝒕
𝑈𝑅 ∙ 𝐷𝑡
𝑫 𝒕
𝑃𝑅𝑀 ∙ 𝐷𝑡

8. 829/06/2018
Deterministic with PRC
+
A planning reserve constraint favors the technology with the lowest investment cost.
• The cost of possibly operating these OCGTs is not considered
• Only peak technology contributes to planning reserve margin

9. 929/06/2018
Stochastic – No load shedding
S-NoENS-low S-NoENS-high

10. 1029/06/2018
Performance of deterministic capacity mix
Residual load duration curves…

11. 1129/06/2018
Stochastic – load shedding
Tradeoff between load shedding and OCGT investments -> depends on VOLL
Two effects combined:
• Expected operational costs
• Endogenous assessment of RES capacity credit
S-ENS-low S-ENS-high

12. 1229/06/2018
Wrap up
Planning reserve constraints favor the technology with the lowest investment costs, since
operational costs of reserve capacity are not included in the optimization.
Demand uncertainty affects technology choices in long-term planning models.
• Capturing multiple scenarios changes the perception of the expected load duration.
• Planning reserve constraints do not capture this effect.
Capacity credit assessment of RES is important for model result interpretation.
• Planning reserve constraints fix RES capacity credit on single timestep, which might lead to
significant technology biases (especially for high RES penetration).
• Endogenous assessment of RES capacity credit might provide different results.

13. CONNECTING TECHNOLOGICAL INNOVATION TO
DECISION MAKING FOR SUSTAINABILITY
SEE YOU IN BRUSSELS!!
November 28-30