Confidence and robustness in fuel cycle simulations - PowerPoint Presentation

1. Confidence and robustness in fuel cycle simulations| PAGE 1CEA – SPRC | Guillaume Krivtchik3rd Technical Workshop on Fuel Cycle Simulation3rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Inspired by the uncertainty breakout session talks from TW FCS 2 – Columbia

2. | PAGE 23rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Confidence in nuclear fuel cycle simulation Sociology insight: scenarios disconnected from the political world [1]Originates from a lack of trust?Scenarios appear as written in stone while even short-term future is unpredictableInstrumentalized to defend positions motivated by other parameters?Up to the scenario community to step up and provide better decision-making materialFirst stepsEvaluate confidence in scenario results, including uncertaintyEvaluate and expose flexibility and adaptability of scenariosProvide scenarios especially designed around their capability to deal with uncertainty[1] S. Tillement, “Between heterogeneity and cooperation, the (electronuclear) scenario as a “boundary object” for decision-making”, 2nd annual Technical Workshop on Fuel Cycle Simulation, Columbia, USA, 2017.

3. Uncertainty / Bias / Variability sources – 1/2 The uncertainty associated with existing thingsModel BiasSimulation backbone: time model (steps vs event), flow vs batchPhysical models: depletion, cooling, fresh fuel equivalenceFacilities models: formalism of plants and associated mass flows“Explicit hypotheses”: transfer functions“Implicit hypotheses”: push vs pull, transitions between campaignsBenchmarks (micro / macro) and validation give the magnitude of biasNuclear data (cross-sections, fission yields, etc.)Well defined framework coming from safety (covariance matrices etc.) Not so easy to take into account adequately (collapsed data + simplified models)Usually relatively low impact (at least on global, integrated outputs)Convenient to study for physicists – possibly too much emphasis on nuclear data?Past and current industrial data (current burnup uncertainty, yields, rates, etc.)Difficult to obtain data Can be tricky to take into account (may require more detailed models or data framework than usually present in scenario codes)Impact may be moderate to strong, but there is possibility of validation3rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018| PAGE 3

4. Uncertainty / Bias / Variability sources – 2/2The uncertainty associated with non-existing thingsProspective industrial data (prospective burnup, yields, rates, etc.)Hypotheses on prospective industrial parametersNo measure, no certaintyImpact may be strongScenario hypotheses ~ strategy (reactor fleet, fuel recycling strategy, etc.)Hypotheses on prospective industrial strategies Choices made to provide a scenario meeting economic & policy goalsCannot be predicted because the future context is unknownImpact is strongPolicy changes and future economy (installed power, scenario objectives, etc.)Wide range of expected future economic trendsExtremely limited knowledge of possible policy changes – even at short-termImpact is game-changing (e.g. French Act on Energy Transition)3rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018| PAGE 4

5. | PAGE 53rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Uncertainty in the scenario as a physical systemThe scenario is a physical systemIt is a trajectory between a beginning and endA scenario is a physical system, with its input and its outputThe input bears uncertainty so does the outputUncertainty propagation methodsDeterministic (mostly for simplified models)Sensitivity (derivatives or deltas) + sandwich formula, problem with thresholdsMonte-Carlo sampling / parameters space explorationBrute-force (+surrogate models / sub-models)Global sensitivity, variance decomposition and rankingMorris, Sobol etc.Which uncertainty sources?Nuclear dataIndustrial data?Scenario hypotheses and policy changes??The scenario was defined to reach a given goalIt is possible for a scenario to become obsolete after a preference changeDoes it mean that scenarios are irrelevant because of “deep”[2] uncertainty ? [2] W.E. Walker et. al., “Deep Uncertainty”, Encyclopedia of Operations Research and Management Science

6. | PAGE 63rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Scenario simulation and scenario searchScenario simulation (linear)Scenario search (iterative)scenario codescenario inputscenario outputscenario codeadjustable scenario inputobjective scenario output12123scenario searchadjustable scenario inputfixed scenario input21fixed scenario input(fixed) goalnon-objective scenario outputnon-objective scenario outputWhile objective output ≠ goal(fixed) goal|||

7. | PAGE 73rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Uncertainty in the scenario search as a decision-making toolThere are major differences between scenarios and other physical systemsThe impact of the “environment” is strong [3]Unforeseen eventsChanging preferencesActions of other playersThe system is not closed between the beginning and the endAdditional data can be extracted during the simulated / real timeDecisions can be made during the simulated / real timeA scenario search does not provide a prediction but an example, usually associated with answering the question “which action available today are likely to best serve the future”Does not imply that all decisions must be made todayPossible to adapt an ongoing scenario to any changes or biasDoes not imply that goals remain identical over timeConclusion: important to take adaptability into account[3] W.E. Walker et. al., “Adaptive policies, policy analysis, and policy-making”, European Journal of Operational Research 128 (2001) 282-289

8. | PAGE 83rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Adaptability of strategies under uncertaintyResistant strategy: performs almost the same in any situation“If I close my eyes for the next 50 years…” policyRaw uncertainty propagation checks if scenarios are resistantPossibly oversized and too expensive, may not existResilient strategy: scenario can be adjusted back into expected behaviorNeed to determine leversPotentially expensive computation (sampling + optimization…)Robust strategy: finds another way to reach goals, or at least “do well” in any situationSame philosophy as resilient scenario, but more flexible frameworkPossibility to change goals?Implementation may be difficultResistantResilientRobustleversleversinin’outout’ ≈ out such that Goal OK out’ ≈ out such that Goal OK outinin’inin’outout’ ≠ out such that Goal’ OK

9. | PAGE 93rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Remarks and conclusionsScenario studies: from spreadsheets-based trends to roadmaps for deploymentMore uncertainty than certaintyScenarios are not predictions, they are rather used to demonstrate feasibility of a strategy in a context regardless the plausibility of the contextGoal of scenario studies is not to make a predictors, but having confidence in the strategiesTraditional uncertainty propagation methods can be applied to scenarios to assess resistanceNeed of innovative methods can be used to test the adaptability of strategies in a range of contexts

10. | PAGE 103rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Confidence and robustness in fuel cycle simulationBefore LunchG. Krivtchik et. al., Nuclear Scenarios: an exercise of robustness analysisA.V. Skarbeli et. al., Strategies of the uncertainty quantification of fuel cycle scenariosAfter lunchA.A. Zakari-Issoufou et. al., Impact of macro reactor approximation on scenario modelization in CLASSN. Thiollière et. al., Functionality Isolation TestDiscussion

11. | PAGE 113rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Thank you for your attentionOne does not predict the future, but prepares for it. Maurice BlondelPrediction is very difficult, especially about the future. Niels Bohr

12. | PAGE 123rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Backup

13. | PAGE 133rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018QuestionsWhat information is provided by a nuclear scenario? → trend / example / roadmap / prediction…2. What is the question that a scenario is supposed to answer?3. Is the scenario (vs strategy) the right object for uncertainty propagation?4. What is a strategy? → set of rules / policies to reach a goal?

14. | PAGE 143rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018General context of nuclear scenario studiesWhy – Provide decision-making elementsAssess the performance of a nuclear system (reactors, reprocessing, etc.)Identify strengths and weaknessesAnalyze strategies and help choosing a strategyFor who – Decision-makers? It is not always clear…For the industry, who owns the fleetFor the government, who defends the interest of the civil societyFor the scientists, in order to guide researchWhat – Characterize fleets and fuel cyclesFeasibility and impact of decisions & technologies on facilitiesPerformanceCostHow – Assess the material flows and inventoriesDynamic modeling of the nuclear fuel cyclePhysical models for depletion, cooling, equivalence etc.

15. | PAGE 153rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Appendix – vocabularyreal timesimulated time ≠ CPU timeFirst operationinitializationobjectivescenarioscenario familyobservablesimulatedobservable

16. | PAGE 163rd Technical Workshop on Fuel Cycle Simulation | 9-11 JULY 2018Policy changes, future economyScenario hypotheses (strategy guidelines)Prospective industrial data(strategy implementation)