Skip to main content
Research Q&A
66 main questions · 539+ sub-questions · Linked to thesis evidence
Problem → Framework → Results → Policy
↑↓ Navigate, Esc Clear filters
B3Layer 5 / 10hard~5 min read
What renders traditional optimization methods inadequate?
Curse of dimensionality (~10⁶-10⁷ variables), scenario explosion (3²⁶ scenarios), non-convex objectives, modeling limitations (10-15% approximation errors).
Key Takeaway
Traditional optimization is computationally intractable for realistic energy transition planning.
Detailed Explanation
The traditional optimization methods are computationally intractable in realistic energy transition settings: Curse of Dimensionality: MILP formulations for multi-regional (7), multi-technology (26), annual resolution (26 years), multi-objective (5 objectives) problems generate 446-dimensional state spaces and 119-dimensional action spaces with binary investment decisions, creating mixed-integer programs with ~10⁶-10⁷ variables and constraints that have prohibitive solve times (weeks to months). Scenario Explosion: Stochastic programming under uncertainty (demand growth ±20%, technology costs ±30%, resource availability ±15%) requires discretizing continuous uncertainty spaces into scenario trees, causing exponential growth in problem size (2 uncertainties×3 discretization levels×26 years = 3²⁶≈10¹² scenarios). Non-Convex Objectives: Real world energy planning involves non-convex phenomena (learning-by-doing with increasing returns to scale, network effects in infrastructure deployment) that violate convexity assumptions required for guaranteed global optimality in traditional methods. Modeling Limitations: MILP requires linear approximations of inherently non-linear processes (S-curve technology diffusion, climate-driven hydropower variability), introducing approximations errors of 10-15%.
Why It Matters
Research Graph — Related Questions
B3 What renders traditional optimization me…
Sub-questions (10)
B3.1Layer 5 / 10hard~2 min read
What is the curse of dimensionality?
~10⁶-10⁷ variables, prohibitive solve times.
Key Takeaway
High-dim intractable.
Detailed Explanation
MILP for 7 regions×26 techs×26 years generates ~10⁶-10⁷ variables with weeks-to-months solve times.
Why It Matters
B3.2Layer 5 / 10hard~2 min read
What is scenario explosion?
3²⁶≈10¹² scenarios for stochastic programming.
Key Takeaway
Scenario tree intractable.
Detailed Explanation
Stochastic programming under uncertainty causes exponential scenario growth (3²⁶ for 2 uncertainties×3 levels×26 years).
Why It Matters
Research significance: Uncertainty challenge.
Energy policy relevance: Scenario planning.
Methodological contribution: Monte Carlo alternative.
B3.3Layer 5 / 10hard~2 min read
What are non-convex objectives?
Learning-by-doing, network effects violate convexity.
Key Takeaway
MILP assumes convexity.
Detailed Explanation
Real-world phenomena (learning-by-doing, network effects) are non-convex; MILP requires convexity for global optimum.
Why It Matters
Research significance: Methodology limitation.
Energy policy relevance: Optimality.
Methodological contribution: RL advantage.
B3.4Layer 5 / 10hard~2 min read
What are the 10-15% approximation errors?
MILP linearizes S-curves, hydropower variability.
Key Takeaway
Linearization loses accuracy.
Detailed Explanation
MILP linear approximations of S-curve diffusion and climate-driven variability introduce 10-15% errors.
Why It Matters
Research significance: Accuracy trade-off.
Energy policy relevance: Pathway quality.
Methodological contribution: Neural alternative.
B3.5Layer 5 / 10hard~2 min read
What is the 446-dimensional state space?
Current capacity, generation, demand, costs, emissions.
Key Takeaway
High-dim state.
Detailed Explanation
MAPPO state space: 182 (capacity) + 91 (generation/demand) + 84 (economic) + 42 (emissions) + 47 (storage/transmission) = 446 dim.
Why It Matters
Research significance: Problem scale.
Energy policy relevance: State representation.
Methodological contribution: C3.
B3.6Layer 5 / 10hard~2 min read
What is the 119-dimensional action space?
Investment allocation, technology portfolio, installation rates.
Key Takeaway
MAPPO action design.
Detailed Explanation
MAPPO action space: 7 (allocation) + 26 (portfolio) + 26 (installation) + 28 (transmission) + 14 (storage) + 18 (retirement) = 119 dim.
Why It Matters
Research significance: Action design.
Energy policy relevance: Policy granularity.
Methodological contribution: C3.
B3.7Layer 5 / 10hard~1 min read
Why does MILP have prohibitive solve times?
Weeks to months for realistic problems.
Key Takeaway
MILP too slow.
Detailed Explanation
MILP formulations for realistic energy transition problems require weeks to months to solve.
Why It Matters
Research significance: Computational motivation.
Energy policy relevance: Timeline.
Methodological contribution: RL speed.
B3.8Layer 5 / 10hard~2 min read
What is stochastic programming limitation?
Scenario tree explodes; discretization loses information.
Key Takeaway
Stochastic programming scales poorly.
Detailed Explanation
Stochastic programming requires scenario discretization; tree grows exponentially with uncertainty dimensions.
Why It Matters
Research significance: Uncertainty method.
Energy policy relevance: Scenario planning.
Methodological contribution: Monte Carlo.
B3.9Layer 5 / 10hard~2 min read
How does MAPPO handle high dimensions?
Continuous action space, gradient-based, sample efficient.
Key Takeaway
MAPPO 4× faster than DQN/GA.
Detailed Explanation
MAPPO uses continuous action space and gradient-based optimization; converges 4× faster than DQN/GA.
Why It Matters
Research significance: RL advantage.
Energy policy relevance: Feasibility.
Methodological contribution: C3.
B3.10Layer 5 / 10hard~1 min read
What is the 7-day MILP timeout?
MILP rank 10 could not solve within 7 days.
Key Takeaway
Timeout demonstrates intractability.
Detailed Explanation
MILP ranked 10th among 13 algorithms—unable to solve within 7-day timeout.
Why It Matters
Research significance: Empirical validation.
Energy policy relevance: Tool limitation.
Methodological contribution: F2 benchmark.