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
C6Layer 5 / 10medium~4 min read
What explainable AI techniques are available in the framework?
SHAP (42% solar/wind CAPEX variance), counterfactuals (gradient-based perturbation), sensitivity heatmaps (∂Renewable Share/∂Parameter).
Key Takeaway
The framework uses SHAP, counterfactuals, and sensitivity heatmaps for explainability.
Detailed Explanation
Three complementary approaches to explainability help us interpret the learnt policy. First, we can perform a SHAP (SHapley Additive exPlanations) analysis, which decomposes RL policy decisions into technology-specific, region-specific and time-specific contributions. For example, we find that solar/wind CAPEX contributes 42% of the variance in renewable share outcomes. Similarly, carbon price contributes 18% and transmission capacity 12%. Secondly, we can use Counterfactuals to automatically generate policy counterfactuals. For example, we can ask "what investment pathway will get us a 95% renewable share (as opposed to 97% which we have made an obligation) at least cost?" Our implementation does so using gradient-based perturbation of the RL action space with constraint satisfaction. Finally, we use Sensitivity Heatmaps, which provide visualisations of ∂ (Renewable Share)/∂ (Parameter) gradients across all our input parameters, highlighting which areas offer the instruments of maximum impact. For example, we find that accelerating solar deployment offers 3× more impact per dollar invested than wind in the 2025–2030 timeframe due to comparative positions in the learning curve of different technologies. Together, these techniques transform the opaque optimization of AI into transparent decision-support tools that accessible for a non-specialist policymaker.
Why It Matters
Research significance: Addresses explainability gap from literature.
Energy policy relevance: Enables stakeholder communication for Turkey policy.
Methodological contribution: SHAP + counterfactuals + sensitivity heatmaps.
Research Graph — Related Questions
C6 What explainable AI techniques are avail…
Sub-questions (10)
C6.1Layer 5 / 10medium~2 min read
What is SHAP in the framework?
SHapley Additive exPlanations—decomposes policy.
Key Takeaway
Technology/region/time.
Detailed Explanation
SHAP decomposes RL decisions into technology, region, and time contributions.
Why It Matters
Research significance: Explainability.
Energy policy relevance: Transparency.
Methodological contribution: C6.
C6.2Layer 5 / 10medium~2 min read
What is 42% solar/wind CAPEX variance?
SHAP contribution to renewable share.
Key Takeaway
CAPEX most important.
Detailed Explanation
Solar/wind CAPEX contributes 42% of variance in renewable share outcomes.
Why It Matters
Research significance: Driver identification.
Energy policy relevance: Cost focus.
Methodological contribution: C6.
C6.3Layer 5 / 10medium~2 min read
What are counterfactuals in C6?
Gradient-based perturbation with constraints.
Key Takeaway
95% renewable at least cost.
Detailed Explanation
Counterfactuals: e.g. 'what pathway gives 95% renewable at least cost?' via gradient perturbation.
Why It Matters
Research significance: Policy exploration.
Energy policy relevance: Scenario analysis.
Methodological contribution: C6.
C6.4Layer 5 / 10medium~2 min read
What are sensitivity heatmaps?
∂(Renewable Share)/∂(Parameter) visualizations.
Key Takeaway
Solar 3× wind 2025-2030.
Detailed Explanation
Heatmaps show parameter impact; solar 3× more impact per dollar than wind in 2025-2030.
Why It Matters
Research significance: Visual explainability.
Energy policy relevance: Instrument selection.
Methodological contribution: C6.
C6.5Layer 5 / 10medium~2 min read
What is carbon price 18% contribution?
SHAP: carbon price explains 18% variance.
Key Takeaway
Policy lever.
Detailed Explanation
Carbon price contributes 18% to renewable share variance per SHAP.
Why It Matters
Research significance: Driver ranking.
Energy policy relevance: Carbon policy.
Methodological contribution: C6.
C6.6Layer 5 / 10medium~2 min read
What is transmission 12% contribution?
SHAP: transmission capacity 12%.
Key Takeaway
Grid investment.
Detailed Explanation
Transmission capacity contributes 12% to renewable share variance.
Why It Matters
Research significance: Driver.
Energy policy relevance: Grid investment.
Methodological contribution: C6.
C6.7Layer 5 / 10medium~2 min read
What is constraint satisfaction in counterfactuals?
Perturbation respects feasibility.
Key Takeaway
Gradient + constraints.
Detailed Explanation
Counterfactual perturbation maintains constraint satisfaction for valid alternatives.
Why It Matters
Research significance: Method detail.
Energy policy relevance: Feasibility.
Methodological contribution: C6.
C6.8Layer 5 / 10medium~2 min read
How are heatmaps useful for policymakers?
Show which parameters offer maximum impact.
Key Takeaway
Accessible visualization.
Detailed Explanation
Heatmaps highlight which parameters offer maximum impact—accessible for non-specialists.
Why It Matters
Research significance: Usability.
Energy policy relevance: Policy design.
Methodological contribution: C6.
C6.9Layer 5 / 10medium~2 min read
What is the 95% vs 97% renewable counterfactual?
Example: pathway for 95% at least cost.
Key Takeaway
Relaxation analysis.
Detailed Explanation
Counterfactual: 'what if 95% renewable (vs 97% obligation) at least cost?'
Why It Matters
Research significance: Example.
Energy policy relevance: Target flexibility.
Methodological contribution: C6.
C6.10Layer 5 / 10medium~2 min read
How do three approaches complement?
SHAP + counterfactuals + heatmaps = full explainability.
Key Takeaway
Transparent decision-support.
Detailed Explanation
SHAP (decomposition), counterfactuals (what-if), heatmaps (impact)—together enable transparent policy.
Why It Matters
Research significance: Integrated XAI.
Energy policy relevance: Stakeholder communication.
Methodological contribution: C6.