{"id":"e1310326-3dbd-4845-8392-f709e3f52192","arxiv_id":"2508.01468","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A bounded fuzzy logic controller sets daily green-hydrogen delivery targets from day-ahead price and wind forecasts, reaching within 9% of perfect-foresight spot-market revenue in simulations.","lead":"This paper tests a fuzzy-logic controller that decides each day how much hydrogen a wind-powered electrolyser should deliver under a long-term purchase agreement, selling the rest into electricity and hydrogen markets. In simulations, the controller earns within 9% of a perfect-foresight optimum and outperforms a steady fixed-delivery schedule.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Revenues are computed under perfect intraday wind and price information; realistic day-ahead forecast error could push BFLC below the 92.8% floor.","rationale":"The reader's stated weakest assumption is the convex-hull generalisation from six benchmark years, which is a genuine extrapolation risk. However, the hull bounds are active by design in nearly every year and the out-of-sample years (2015, 2016, 2023) still meet the 92.8% floor, so this concern is partially mitigated by the reported results. The more load-bearing issue is that the evaluation is not actually carried out with forecasts: the input time series are realised day-ahead prices and reanalysis wind data, so the controller sees the day's outcome exactly. This affects every revenue figure and conflicts with the stated 'sequential decision making where only 24 h of forecast inputs are available.' It is not an extrapolation risk but a mismatch between the method as described and the method as implemented, and it is directly falsifiable by one rerun with forecast-error-perturbed wind. I agree with the reader's CONDITIONAL verdict; this concern reinforces it rather than changing it, so the verdict should remain UNCHANGED.","tokens_in":14103,"tokens_out":5450,"duration_ms":72977,"concrete_test":"Rerun the Danish 2015-2023 and UK case studies using true day-ahead wind forecasts (e.g., from an NWP reforecast dataset or renewables.ninja's forecast API) instead of MERRA-2 actuals, while keeping the day-ahead electricity prices and the same dispatch logic. Settle actual hydrogen production against realised wind, applying an imbalance penalty or a make-good purchase for any shortfall in the daily HPA target (or, if no make-good is allowed, record the violation rate). If the minimum normalised BFLC revenue drops below 92.8% or the HPA target is missed in any year, the central claim depends on perfect intraday information and the abstract's 'realistic' characterisation fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline number is a simulation result. Section V.A says the controller uses 'only 24 h of forecast inputs' and Figure 3 labels the inputs as day-ahead forecasts, but Appendix A reveals the actual implementation: electricity prices are historical ENTSO-E day-ahead prices and wind capacity factors are bias-corrected reanalysis (MERRA-2) actuals. No forecast error is introduced anywhere. The daily dispatch optimisation in Section V.A imposes \\bar{M}^d_2 as a minimum hydrogen production constraint and prohibits grid imports (Section II). If real wind falls short of the day-ahead value, the electrolyser cannot meet this target, and the year-end HPA obligation is jeopardised; the model has no recourse, imbalance settlement, or penalty. The benchmark (Section IV) also uses perfect foresight, so the comparison is internally consistent but is a comparison of two perfect-information schedulers, not a realistic sequential controller. The abstract's claim of 'realistic revenue quantification' and the Section VI revenue ratios therefore rest on the unvalidated assumption of zero intraday forecast error. Because the fuzzy controller's daily target is trained and bounded to mimic the perfect-foresight benchmark, any forecast error will degrade BFLC revenue more than it degrades the benchmark, and the 92.8% floor could be violated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Bounded Fuzzy Logic Control (BFLC) for scheduling hydrogen production under a long-term Hydrogen Purchase Agreement. The controller maps daily mean electricity price, hydrogen price, and wind capacity factor to a daily HPA delivery target; the target is capped by producibility and bounded by a convex hull of cumulative benchmark export paths, and is then imposed as a minimum production constraint in an hourly dispatch optimisation. The BFLC is trained with particle swarm optimisation against benchmark results obtained from a year-ahead perfect-foresight optimisation over 2017-2022, and revenue comparisons are reported for Denmark 2015-2023 and for a UK site. The central claim is that BFLC total revenue is within 9% of the perfect-foresight benchmark and consistently exceeds steady control, with the lowest normalised revenue of 92.8% in 2022.","tokens_in":14404,"tokens_out":6769,"duration_ms":82607,"significance":"If the claims held in realistic operation, the paper would make a useful practical contribution: an interpretable fuzzy controller running on daily information would capture most of the variable-market revenue available to a perfect-foresight scheduler while maintaining HPA deliverability. The use of out-of-sample years (2015, 2016, 2023) and a second country is a genuine strength, as is the explicit comparison against a steady controller and the use of established open-source tools (PyPSA, scikit-fuzzy). The main limitations are that the inputs are effectively perfect-information realisations rather than forecasts, the hydrogen price series is synthetic, and the in-sample years partly evaluate a controller fitted to the benchmark. These limitations do not invalidate the relative comparison, but they do mean the quantitative floor of 92.8% should be treated as a simulation result under idealised information rather than a realistic revenue guarantee.","major_comments":[{"comment":"The paper presents BFLC as using 'only 24 h of forecast inputs', but the implementation uses realised historical data: electricity prices are historical ENTSO-E day-ahead prices and wind capacity factors are bias-corrected reanalysis actuals (Appendix A). No forecast error is introduced anywhere, while the daily dispatch optimisation in Section V.A imposes the daily HPA target as a minimum production constraint and Section II prohibits grid imports. Under a realistic day-ahead wind forecast error, the electrolyser may be unable to meet the daily HPA target, and the model contains no imbalance settlement, storage recourse, or penalty mechanism. The benchmark in Section IV also uses perfect foresight, so the revenue comparison is internally consistent, but the abstract's claim that the BFLC enables 'realistic revenue quantification' is not supported by the simulation design; a sensitivity analysis with additive forecast error is needed before the 92.8% floor can be regarded as robust.","section":"Section V.A, Figure 3, Appendix A"},{"comment":"The membership functions and rule base are optimised by PSO to reproduce benchmark HPA supply over 2017-2022 (Section V.B), and Figure 8 reports the resulting in-sample BFLC revenues for the same years, including 2022, the year of the 92.8% floor. These in-sample results partly quantify fit quality rather than control performance. The out-of-sample years (2015, 2016, 2023) and the UK case in Appendix D provide genuine independent evidence and should be the primary support for the revenue claim, but the paper should state explicitly that the 92.8% floor occurs in a training year and should separate in-sample and out-of-sample results in the headline claims.","section":"Section V.B, Section VI, Figure 8"},{"comment":"The 'optimal space' is the convex hull of exactly six benchmark cumulative HPA-export trajectories (2017-2022; Figure 6). Whenever the cumulative export would leave this hull, the bounding logic overrides the trained fuzzy output, and Section V.C concedes that optimal paths for some unusual years may cross outside the hull. Since the 92.8% floor and the beats-steady-control claim depend on the hull containing the tested years, the paper should provide a concrete robustness test, such as leave-one-year-out construction of the hull, before generalising from six yearly paths.","section":"Section V.C, Figure 6"},{"comment":"Revenues from the hydrogen market are computed with a synthetic price series generated by scaling electricity prices to a 3 EUR/kg mean, adding uniform +/-25% random variation, and capping to 1-5 EUR/kg (Appendix A). Because hydrogen sales are one of the two revenue streams being optimised, every quantitative revenue ratio in Section VI is conditional on this constructed series. The relative comparison between controllers remains informative because all controllers face the same prices, but the absolute claim of 'realistic revenue quantification' is not supported; the authors should either use a published hydrogen price series or report sensitivity of the 92.8% figure to the hydrogen price model.","section":"Appendix A"}],"minor_comments":[{"comment":"In the paragraph after Figure 6, 'convex full' should be 'convex hull'.","section":"Section V.C"},{"comment":"The text contains the typo 'disptach optimisation'; it should read 'dispatch optimisation'.","section":"Section II"},{"comment":"The notation for the daily HPA target is inconsistent: Section II uses Mbar^d_2, Section V.A uses M^d_2, and Appendix B uses fM^d_2 and M^d_2; please unify the symbols and define the capped target before first use.","section":"Nomenclature and Appendix B"},{"comment":"The caption lists the colour coding for the electrical-energy panel but not for the hydrogen panel; make the caption self-contained.","section":"Figure 2 caption"},{"comment":"The statement that 40% of mean maximum production is 'assumed' to be a suitable contract volume is not justified; a sensitivity analysis over this fraction would clarify how the 92.8% result depends on the contract volume.","section":"Section III"},{"comment":"No code or data availability statement is included; given that the model is built on PyPSA and Gurobi, a reproducibility statement or repository link would materially strengthen the paper.","section":"Reproducibility"}],"recommendation":"major_revision","confidential_remarks":"The central comparison is internally coherent, but the paper's framing of 'realistic revenue quantification' overstates what the simulation evidence supports. If the authors can add a forecast-error sensitivity analysis, separate in-sample from out-of-sample claims, and either use real hydrogen price data or reframe the hydrogen price as a scenario, the result would be publishable. The synthetic hydrogen price and the perfect-forecast implementation should also be disclosed more prominently in the abstract or introduction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a solid, clearly written application of established fuzzy logic and PSO to a real operational problem—setting daily hydrogen delivery targets under an HPA—and the central simulation claim is plausible within the stated model. What is genuinely new: no one in the cited sequential-planning line ([21]–[23]) uses a fuzzy controller or benchmark-derived convex-hull bounds for daily HPA targets, and the bounding idea is a reasonable safeguard against the controller drifting too fast or too slow. The authors also did the right thing by testing on out-of-sample years (2015, 2016, 2023) and a UK site, and the BFLC consistently beats a steady controller in their simulations. I give them credit for that design.\n\nThe soft spots are real but not fatal. First, Appendix A shows that the \"day-ahead forecasts\" are actually historical day-ahead electricity prices and bias-corrected reanalysis wind actuals, with no forecast error introduced anywhere. The dispatch also has no recourse or imbalance penalty if wind falls short. So the comparison is internally consistent—benchmark and BFLC both see perfect information—but the abstract's phrase \"realistic revenue quantification\" oversells it. The 92.8% floor could erode under realistic day-ahead errors, exactly as the stress-test note says. This is the load-bearing weakness.\n\nSecond, hydrogen prices are synthetic: scaled electricity prices with ±25% noise, clipped to 1–5 €/kg. That is fine for a method demonstration but limits the economic realism of the revenue numbers. Third, the convex-hull bounds are built from only six yearly benchmark trajectories, and the paper itself concedes unusual years could leave the hull; the out-of-sample results help, but the generalization is borrowed confidence. Fourth, 2017–2022 results partly evaluate a controller fitted to the benchmark, so the in-sample revenue ratios measure fit quality as much as control quality. No code or data are released, so independent reruns are impossible.\n\nWho is this for? People working on operational hydrogen scheduling and HPA contract structures, not control theorists. The paper deserves a serious referee: the method is clearly explained, the comparison is coherent, and the flaws are addressable rather than structural. A good referee should ask for a forecast-error stress test, a clearer distinction between actuals and forecasts in the text, and ideally code/data release.","headline":"A competent, clearly written fuzzy-control application for daily HPA targets that plausibly beats steady control inside its model, but the 9% claim should be read as perfect-information, not realistic-forecast, performance.","tokens_in":14926,"tokens_out":1719,"would_cite":true,"duration_ms":23987,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A bounded fuzzy logic controller using day-ahead forecasts recovers over 92% of perfect-foresight green hydrogen revenue.","keywords":["green hydrogen","fuzzy logic control","hydrogen purchase agreement","renewable energy scheduling","revenue maximisation","day-ahead forecasting","wind-powered electrolysis","bounded control"],"falsifier":"Take a year, real or synthetic, whose optimal cumulative HPA-export path leaves the convex hull built from the 2017-2022 benchmark trajectories, run the BFLC on day-ahead data alone, and compare its normalised revenue against the perfect-foresight benchmark; if it falls below 92% or below steady control, the central claim is falsified. A cheaper check is to test whether any year-optimal path from a different site or market regime crosses the hull boundary.","tokens_in":13860,"feed_emoji":"⚡","tokens_out":7177,"duration_ms":85643,"temperature":0.7,"pith_summary":"This paper argues that long-term hydrogen purchase agreement obligations can be scheduled almost as well with a simple interpretable controller as with an all-knowing year-ahead optimizer. It introduces a Bounded Fuzzy Logic Control (BFLC) that reads day-ahead forecasts of electricity price, hydrogen price, and wind capacity, sets a daily HPA delivery target, and lets an hourly dispatch optimization allocate energy and hydrogen. Across several years of Danish wind and price data, the BFLC recovers 92.8% or more of the benchmark revenue in every year, including the volatile 2022 market, and always beats a steady daily delivery policy. The practical point is that realistic, bankable revenue estimates for green hydrogen projects need not assume perfect foresight; a rule-based controller using only day-ahead information is enough to stay within striking distance of the theoretical optimum.","feed_headline":"Fuzzy control captures 92% of perfect-foresight hydrogen revenue","feed_subtitle":"A bounded fuzzy logic controller using only day-ahead data stays within 9% of ideal revenue and beats steady delivery.","key_machinery":"The central object is the Bounded Fuzzy Logic Controller, a three-input, one-output fuzzy system with triangular membership functions and centroid defuzzification, optimised against perfect-foresight benchmark trajectories. The load-bearing addition is the optimal-space bound: the convex hull of six benchmark cumulative-HPA-export curves, which clips the fuzzy output whenever cumulative deliveries would leave the hull. The daily target then enters a dispatch optimisation that maximises revenue from hourly electricity and hydrogen sales subject to meeting that target, with the fixed HPA target $M_2^*$ set at 48 tonnes.","core_discovery":"If the paper's results are right, the optimal sequencing of hydrogen deliveries under an HPA can be approximated by a transparent three-input fuzzy rule base plus a safety envelope. The envelope is the convex hull of the cumulative HPA-delivery paths produced by perfect-foresight optimization over six historical years; whenever the fuzzy controller's cumulative exports try to leave that hull, the target is clipped back inside. With that clipping, total market revenue (electricity plus hydrogen spot sales, excluding the fixed HPA value) stays within 9% of the perfect-foresight benchmark every year tested, with the worst case 92.8% in 2022, and consistently exceeds steady control. The largest gains over steady control occur exactly when electricity prices are high and volatile.","pith_inferences":["A natural extension is to re-derive the convex-hull bounds for other plant sizes, storage additions, or hybrid wind-solar sites; the paper only demonstrates the method on the six-year Danish and UK cases, so the transfer is plausible but unproven.","Because the hydrogen price series are synthetic, derived by scaling electricity prices and adding noise, part of the claimed performance may depend on the correlation between the two price series; an independent real hydrogen-price series would test that.","The same controller logic could be applied at shorter time scales, such as weekly or intraday targets, if the benchmark paths were recomputed at that resolution; the paper does not test this.","The worst-case year 2022 is inside the training hull because 2022 is a training year, so out-of-sample performance in a similarly extreme but structurally different year would be the stricter test."],"forward_implications":["A hydrogen project can be operated in real time with only day-ahead market data while keeping annual market revenue within 9% of the perfect-foresight optimum.","The controller's guaranteed delivery bounds make it possible to quote a bankable HPA volume without either over-committing or stranding market upside.","The advantage over steady delivery grows when electricity prices are high and volatile, so the method is most valuable in the market regimes that are hardest to plan.","Because the HPA contract value is fixed, the same framework can separate scheduling skill from contract terms when evaluating project economics."],"supporting_citations":[{"why":"Supplies the graphical sequential daily-planning baseline that the paper contrasts with the bounded fuzzy approach.","marker":"[21]"},{"why":"Supplies the co-optimisation sequential planning method that the BFLC extends to wind-only operation with a hydrogen market.","marker":"[22]"},{"why":"Provides the open-source modelling environment used to simulate the hydrogen plant and its energy and mass balances.","marker":"[31]"},{"why":"Provides the optimisation solver that computes both the perfect-foresight benchmark and the daily dispatch decisions.","marker":"[32]"},{"why":"Provides the particle swarm optimiser that tunes the fuzzy membership parameters and rule base.","marker":"[36]"},{"why":"Supplies the rule-activation method used to select the most consistent fuzzy rules from benchmark trajectories.","marker":"[37]"},{"why":"Supplies the day-ahead electricity price time series used to train and test the controller.","marker":"[38]"},{"why":"Supplies the bias-corrected reanalysis data from which the wind capacity factor time series are constructed.","marker":"[39]"},{"why":"Supplies the wind generation time series for the sites used in the Danish and United Kingdom case studies.","marker":"[40]"}],"fun_headline_variants":["Bounded fuzzy logic stays within 9% of optimal hydrogen revenue","Fuzzy controller hits 92.8% of perfect hydrogen revenue","Fuzzy scheduling nears ideal hydrogen revenue with safety envelope","92.8% of optimal revenue via bounded fuzzy logic","Fuzzy logic clipped to perfect-foresight envelope hits 92.8%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The assumption that the convex hull of six historical optimal delivery paths covers the optimal paths of future years is load-bearing; if an unusual wind-price year makes the true optimal path leave that hull, the controller's bounds force it off the trained behavior and the 9% claim could fail.","fun_headline_variants_meta":{"raw":{"variants":["Bounded fuzzy logic stays within 9% of optimal hydrogen revenue","Fuzzy controller hits 92.8% of perfect hydrogen revenue","Fuzzy scheduling nears ideal hydrogen revenue with safety envelope","92.8% of optimal revenue via bounded fuzzy logic","Fuzzy logic clipped to perfect-foresight envelope hits 92.8%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000565,"raw_usage":{"total_tokens":2660,"prompt_tokens":909,"completion_tokens":1751,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":525,"completion_tokens_details":{"reasoning_tokens":1661}},"tokens_in":525,"tokens_out":1751,"duration_ms":14625,"temperature":1.0,"reasoning_tokens":1661,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T05:36:31.455897+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a year, real or synthetic, whose optimal cumulative HPA-export path leaves the convex hull built from the 2017-2022 benchmark trajectories, run the BFLC on day-ahead data alone, and compare its normalised revenue against the perfect-foresight benchmark; if it falls below 92% or below steady control, the central claim is falsified. A cheaper check is to test whether any year-optimal path from a different site or market regime crosses the hull boundary.","supporting_citations":[{"cited_title":"Beyond short-duration energy storage","cited_arxiv_id":null,"evidence_quote":"Supplies the graphical sequential daily-planning baseline that the paper contrasts with the bounded fuzzy approach."},{"cited_title":"Power-to-gas and power-to-x—the history and results of developing a new storage concept","cited_arxiv_id":null,"evidence_quote":"Supplies the co-optimisation sequential planning method that the BFLC extends to wind-only operation with a hydrogen market."},{"cited_title":"Brown, J","cited_arxiv_id":null,"evidence_quote":"Provides the open-source modelling environment used to simulate the hydrogen plant and its energy and mass balances."},{"cited_title":"H2global – idea, instrument & in- tentions","cited_arxiv_id":null,"evidence_quote":"Provides the optimisation solver that computes both the perfect-foresight benchmark and the daily dispatch decisions."},{"cited_title":"Soliman, Hany M","cited_arxiv_id":null,"evidence_quote":"Provides the particle swarm optimiser that tunes the fuzzy membership parameters and rule base."},{"cited_title":"A review on applications of fuzzy logic control for refrigeration systems","cited_arxiv_id":null,"evidence_quote":"Supplies the rule-activation method used to select the most consistent fuzzy rules from benchmark trajectories."},{"cited_title":"Steering control in electric power steering autonomous vehicle using type-2 fuzzy logic control and pi control","cited_arxiv_id":null,"evidence_quote":"Supplies the day-ahead electricity price time series used to train and test the controller."},{"cited_title":"Beshir, and Zhongxia Zhang","cited_arxiv_id":null,"evidence_quote":"Supplies the bias-corrected reanalysis data from which the wind capacity factor time series are constructed."},{"cited_title":"Delegated regulation on Union methodology for RFNBOs","cited_arxiv_id":null,"evidence_quote":"Supplies the wind generation time series for the sites used in the Danish and United Kingdom case studies."}],"review_version":1}