{"id":"33729242-6ebe-43ed-b023-66b5b1451382","arxiv_id":"2507.14385","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A bi-level model predictive control framework jointly sets product prices and hourly production schedules to shift demand toward periods of high solar generation, reducing grid costs and increasing profit in a five-day simulation.","lead":"This paper proposes a two-layer control framework that sets product prices daily and production schedules hourly, using solar forecasts and real-time electricity prices. In a simulated battery-pack factory, the method cut grid electricity costs by about half while raising total profit by 3.7 percent.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm 1's approximate gradient is the load-bearing step: it is not the derivative of the upper objective and has an apparent sign inconsistency in Eq. (24), so the reported price path and profit gains are not yet connected to any optimum.","rationale":"The reader's weakest assumption matches the most load-bearing concern. The proposed framework's novelty is the approximate gradient for a bi-level MIQP; if that update is not a valid search direction, the 'optimizes' claim and the case-study improvements rest on an arbitrary heuristic. The concern is testable because the upper variable is one-dimensional, making exhaustive search cheap. The paper has real strengths: a concrete network model, realistic data, and reproducible MIQP solves, but the missing benchmark is necessary. Baseline specification is a secondary issue; the main risk remains Algorithm 1. If the grid-search test confirms Algorithm 1's prices, the concern is resolved and the existing CONDITIONAL verdict could be upgraded; if not, the central claim would need substantial revision.","tokens_in":932,"tokens_out":1779,"duration_ms":657735,"concrete_test":"Because the upper-level decision is a single scalar daily price, replace Algorithm 1 with exhaustive grid search for each of the five days: for p in {70,70.1,...,120}, solve the L-MPC MIQP (17), evaluate the exact upper objective (18), and take the best p. Compare the resulting price path, grid cost, profit, and renewable fraction with Table 2. If Algorithm 1's price is not within a small tolerance (e.g., 1%) of the grid-search-optimal price or its objective value is materially worse, the approximate gradient is not a valid search direction. As a second check, re-run Algorithm 1 with the sign of G_u flipped to determine whether the implementation follows Eq. (24)/(26) or the text's stated behavior.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the bi-level MPC jointly optimizes price and schedule and that this yields a 49.4% grid-cost reduction and a 3.7% profit gain. The load-bearing step is Algorithm 1's update rule p_{k+1} = projection[p_k - kappa_u G_u(p_k) - kappa_d G_d(p_k)] (Eq. 26). Three problems make the resulting price path unsupported as an optimum. First, G_u in Eq. (24) is a heuristic that drives the renewable fraction toward iota*, not a derivative of the renewable term -E_r in the upper objective (18), so there is no reason its zero should coincide with the upper objective's stationary point. Second, G_d in Eq. (25) is the derivative of p(a-bp), i.e., revenue from potential demand, whereas the upper objective uses realized lower-level outflows d*(p); these need not agree because the lower-level MIQP may not deliver the full target or may deliver more due to scheduling constraints. Third, the paper's own description contradicts the update: for iota < iota*, Eq. (24) gives G_u > 0, so Eq. (26) lowers the price, while the text says the rule raises price when renewable utilization is below the desired level. If the implemented sign matches the text instead, the case-study results are produced by a different algorithm than the one stated. No convergence proof, stationarity check, or comparison against an exact solution is provided. The 49.4% and 3.7% figures are therefore not yet attributable to the claimed joint optimization.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a bi-level model predictive control framework for a make-to-stock manufacturing system with onsite solar generation. At the higher level, a daily product price is chosen to maximize revenue and renewable-energy usage; at the lower level, an hourly MIQP scheduling problem is solved to meet the resulting demand at minimum grid-energy, startup, holding, and slack cost. The higher-level price is updated by a component-wise approximate gradient rule (Algorithm 1, Eq. (26)). In a five-day simulation of a lithium-ion battery pack production line, the authors report a 49.4% reduction in grid energy cost and a 3.7% increase in profit relative to a grid-only baseline with constant pricing. The central claim is that the proposed bi-level scheme jointly and profitably coordinates dynamic pricing with energy-aware scheduling.","tokens_in":8022,"tokens_out":3497,"duration_ms":38842,"significance":"If the heuristic gradient were proven to provide a reliable search direction for the nonconvex bi-level problem, the framework would be a useful step toward integrating strategic pricing with operational production control. The paper has clear strengths: the lower-level MIQP is carefully specified with realistic on/off, minimum-run, buffer, and production-tightening constraints; the case study is detailed and uses publicly available price and solar data; and the authors honestly label the gradient approximation as a heuristic. However, the load-bearing step—the component-wise approximate gradient—is not validated against any optimum, and the baseline comparison is not fully specified. The empirical claims therefore are not yet attributable to the proposed joint optimization in the way the paper states.","major_comments":[{"comment":"The approximate gradient is the load-bearing component of the paper, but no evidence is given that it points toward a stationary point of the upper-level objective (18). G_u in Eq. (24) is a heuristic function of the renewable fraction, not the derivative of -E_r with respect to price, and G_d in Eq. (25) is the derivative of p(a-bp), using potential demand rather than the realized lower-level outflow d*(p) that appears in the objective. The algorithm provides no convergence proof, no residual or stationarity check, and no comparison against an exact solution. I recommend adding a numerical validation on small instances (for example, comparing Algorithm 1 with a dense price grid or an exact reformulation) or a formal convergence analysis for a simplified version of the problem.","section":"Section 4.2, Eqs. (22)-(26), Algorithm 1"},{"comment":"The baseline scenario is not an apples-to-apples comparison with the solar-integrated scenario. The baseline keeps price constant at 82.50 USD while the solar case uses dynamic pricing, but the paper does not explain how the baseline price 82.50 was chosen or whether it was optimized under the grid-only scenario. As a result, the reported 49.4% grid-cost reduction and 3.7% profit increase conflate the effect of solar integration with the effect of switching from a constant to a dynamic pricing policy. The paper should specify the baseline pricing policy and, ideally, decompose the outcomes into (i) the effect of dynamic pricing alone and (ii) the effect of solar integration.","section":"Section 5.2, Table 2 and Fig. 3"},{"comment":"The quantitative claims rest on a single five-day simulation with no sensitivity analysis or uncertainty quantification. Several free parameters in Table 1—particularly the gradient weights kappa_u=0.4 and kappa_d=0.6, the desired renewable fraction iota*=0.5, and the L-MPC weights—are likely to influence the resulting price path and the numbers in Table 2. A sensitivity study over these parameters, or at least a small set of alternative settings, is needed to support the claim that the 49.4% and 3.7% figures are robust properties of the framework rather than artifacts of a particular parameter choice.","section":"Section 5, Tables 1-2"}],"minor_comments":[{"comment":"Line 6 of Algorithm 1 reads pk+1 = projection[pk - kappa_u G_d(pk) - kappa_d G_d(pk)], with G_d appearing twice; one of these should be G_u. The surrounding text and Eq. (26) use G_u and G_d, so this is a typo, but it makes the algorithm as printed ambiguous.","section":"Algorithm 1, line 6"},{"comment":"The upper-level objective uses p(k) and d*(k) as time-indexed quantities, while the text states that the higher level determines a daily price. The notation should be clarified to distinguish the daily price decision from hourly time indices, and to define how the daily price enters the H-step summation.","section":"Section 4.1, Eq. (18)"},{"comment":"The table entry \"1,5937\" appears to be a typo for \"15,937\" (and similarly \"1,6530\" for \"16,530\"). The formatting should be corrected.","section":"Section 5.2, Table 2"},{"comment":"G_u does not explicitly depend on p; for a fixed renewable fraction it is a constant sign, so calling it a gradient direction is a loose use of the term. Clarifying that it is a heuristic surrogate would help readers calibrate the strength of the claim.","section":"Section 4.2, Eq. (24)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal, but the central methodological assumption—validity of the approximate gradient as a search direction—is currently unsupported. The missing baseline specification and lack of sensitivity analysis reinforce the need for revision before the empirical claims can be accepted. I would not reject the paper outright because the lower-level formulation and case-study setup are well presented, and the heuristic could be salvageable with additional numerical validation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper is a clearly written proposal for a bi-level MPC that couples daily product pricing with hourly production scheduling under solar availability. The joint treatment of pricing, scheduling, and renewables in one framework appears to be new relative to the pairwise studies they cite, and the lower-level MIQP with machine on/off constraints is well specified. The case study is plausible, and the reported 49.4% grid-cost reduction and 3.7% profit gain are concrete numbers.\n\nThe big soft spot is the higher-level optimizer. Algorithm 1 uses a component-wise approximate gradient, but there is no proof that it descends on the upper objective or that its stationary point matches the true bi-level optimum. More troubling, the sign of the renewable gradient in Eq. (24) is inconsistent with the text: for a renewable fraction below the desired level, G_u is positive, so the update in Eq. (26) lowers the price, while the text says the rule should raise the price in that situation. Either the equation or the narrative is wrong, and the reported price paths could come from a different algorithm than the one stated. Algorithm 1's line 6 also writes both weights on G_d, which looks like a typo and adds to the confusion.\n\nThe revenue gradient G_d approximates revenue using the potential demand a-bp rather than the realized lower-level output d*(p). The paper acknowledges this is an approximation, but without a benchmark against an exact solution you cannot tell how far it is from optimal. The validation is a single five-day deterministic run with no sensitivity analysis, and the baseline keeps a constant price while the solar case uses dynamic pricing, so the comparison conflates pricing strategy with solar integration. The qualitative finding that prices drop during high solar is largely built into the gradient component, so it is not a surprising discovery.\n\nWhat the paper does well is lay out a sensible model and a concrete integration idea. The flaws are addressable: fix the sign error, add a benchmark or stationarity check, run more scenarios, and specify the baseline pricing rule. I would not desk reject this; the problem is relevant and the framework has merit, but the core algorithm needs serious scrutiny before any efficiency claim can be trusted.\n\nRecommendation: send to peer review with major revisions expected.","headline":"A plausible integration of pricing and scheduling MPC, but the core approximate-gradient algorithm has a sign inconsistency and no validation, so the headline gains are unsupported.","tokens_in":8530,"tokens_out":4007,"would_cite":false,"duration_ms":48485,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A bi-level model predictive control framework that sets daily prices and hourly production schedules together can cut grid electricity costs by 49.4% and raise profit by 3.7% for a make-to-stock manufacturer with onsite solar.","keywords":["bi-level model predictive control","production scheduling","dynamic pricing","renewable energy integration","price elasticity","make-to-stock manufacturing","mixed-integer quadratic program","energy-aware manufacturing"],"falsifier":"Solve the same five-day case study by an exact or exhaustive method, for example by evaluating the bi-level objective on a fine grid of admissible daily prices or by using a global MIQP solver with KKT conditions for the lower level, and compare with Algorithm 1's price path. If the exact optimum does not reproduce the 49.4% grid-cost saving and 3.7% profit gain, or if Algorithm 1's result depends on the initial price, the central claim fails.","tokens_in":7389,"feed_emoji":"☀️","tokens_out":6971,"duration_ms":71490,"temperature":0.7,"pith_summary":"This paper argues that product pricing and hourly production scheduling should be optimized together, not separately, when a manufacturer has onsite solar generation. It builds a bi-level model predictive control framework: a slower loop chooses the daily price to shape demand through price elasticity, and a faster loop schedules machines hour by hour against real-time electricity prices and solar availability. The higher level rewards renewable usage, so the controller lowers prices during sunny hours, pulling demand toward cheap, clean energy. In a lithium-ion battery pack case study the framework cuts grid energy costs by 49.4% and increases profit by 3.7% compared with a grid-only baseline.","feed_headline":"Bi-level control cuts grid costs 49% and lifts profit 3.7%","feed_subtitle":"Solar-aware pricing shifts demand to sunny hours, cutting grid use in a battery plant case.","key_machinery":"The load-bearing object is Algorithm 1, the component-wise approximate gradient method for bi-level MPC, combined with the affine price elasticity $\\gamma = a - bp$ that couples the two levels. The algorithm replaces the intractable hypergradient $\\nabla_p F = \\partial F/\\partial p + (\\partial F/\\partial u)(\\partial u/\\partial p) + (\\partial F/\\partial d)(\\partial d/\\partial p)$ by $\\kappa_u G_u(p_k) + \\kappa_d G_d(p_k)$, where $G_u$ measures renewable utilization and $G_d$ is the derivative of the revenue curve. The lower-level MIQP supplies the production schedule and energy mix for each candidate price, and the time-varying slack constraint lets the controller defer production early and converge to the target as the deadline approaches.","core_discovery":"The central claim is that a tractable component-wise approximate gradient method can solve the bi-level pricing-and-scheduling problem, and that the resulting policy — lower prices when solar is abundant — reconciles sustainability with profitability. On the paper's own terms, the discovery is that the bi-level MPC structure with an MIQP lower level need not be solved by KKT reformulation or exact hypergradients; an approximated gradient combining a revenue component $a - 2bp_k$ and a renewable-utilization component $(1 - \\iota^*/\\iota(p_k))(p_{\\max}-p_{\\min})$ is enough to drive the daily price updates. In the five-day case study this produces a 9.4% average price reduction, 11.5% more production, 49.4% lower grid energy cost, 25.9% lower startup cost, and 3.7% higher profit while all demand targets are met.","pith_inferences":["The 49.4% grid-cost saving is only as reliable as the approximate gradient: were the update direction systematically biased, the price path would be an artifact rather than the optimum, which is why the result should be stress-tested against a price grid search.","The same two-timescale bi-level template could be applied beyond make-to-stock settings, for example to demand response with storage, where the lower level would schedule charging instead of production and the price gradient would target load shifting.","The rise in holding costs suggests inventory is the buffer that absorbs solar variability; optimizing buffer capacity alongside price could improve the trade-off further.","Because the case study omits solar capital costs, the reported profit gain is operational only; including levelized cost of energy would move the break-even point and may change the optimal price level."],"forward_implications":["A make-to-stock manufacturer with onsite solar can use dynamic pricing to shift demand into high-solar hours without sacrificing profit, because the lower price is more than offset by reduced grid costs.","The bi-level scheme is computable in practice: the MIQP lower level is solved with an off-the-shelf optimizer and the price update uses only two scalar gradient components, so no costly KKT reformulation is needed.","Grid energy costs fall roughly by half (49.4%) and startup costs by a quarter (25.9%) in the studied system, at the cost of a modest 5.6% rise in buffer holding costs.","Lowering the average price by 9.4% stimulates 11.5% more production, and all daily demand targets remain satisfied, so the policy does not trade away service.","The paper notes that comparable solar installations have approximately 24% return on investment and may qualify for U.S. tax credits, so the long-term economic case is stronger than the operational-only gain."],"supporting_citations":[{"why":"Supplies the hyper-gradient formula and the alternating-gradient perspective that the paper approximates in Algorithm 1.","marker":"Chen et al., 2021"},{"why":"Demonstrates bi-level optimization for real-time control and is the direct inspiration for the component-wise approximate gradient approach.","marker":"Olkin and Ames, 2024"},{"why":"Provides the make-to-stock production system model that motivates the demand-fulfillment scheduling problem.","marker":"Ha, 1997"},{"why":"Establishes the price elasticity dynamics used to argue that lowering price raises demand in make-to-stock systems.","marker":"Fibich et al., 2005"},{"why":"Underpins the affine price elasticity function as the theoretical demand model.","marker":"Mankiw, 2021"},{"why":"Earlier energy-aware scheduling framework for batch manufacturing that this paper extends by adding the pricing level.","marker":"Li et al., 2023"},{"why":"Source of the lithium-ion battery pack manufacturing parameters used to build the case study.","marker":"Heimes et al., 2018"},{"why":"Provides the real-time electricity price data used in the five-day simulation.","marker":"PJM Interconnection, LLC, 2023"},{"why":"Supplies the solar radiation data and calculator used to estimate onsite PV availability.","marker":"National Renewable Energy Laboratory (NREL), 2023"},{"why":"Gives the 24% average ROI figure for comparable solar installations, used to argue the long-term economic case.","marker":"Formica and Pecht, 2017"}],"fun_headline_variants":["Solar-aware pricing cuts grid spend 49%, lifts profit 3.7%","Bi-level MPC: 49% cheaper grid energy, 3.7% higher profit","Price drops when sun shines: 49% grid savings, profit up","Renewable-responsive scheduling trims grid costs 49%","Smart pricing shifts load to solar, saving 49% on grid"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire result depends on the assumption that the simplified price-update rule used in Algorithm 1 points toward the true best price; the paper does not prove this, so the reported savings could be an artifact of the update rule if that assumption fails.","fun_headline_variants_meta":{"raw":{"variants":["Solar-aware pricing cuts grid spend 49%, lifts profit 3.7%","Bi-level MPC: 49% cheaper grid energy, 3.7% higher profit","Price drops when sun shines: 49% grid savings, profit up","Renewable-responsive scheduling trims grid costs 49%","Smart pricing shifts load to solar, saving 49% on grid"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000255,"raw_usage":{"total_tokens":1529,"prompt_tokens":862,"completion_tokens":667,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":478,"completion_tokens_details":{"reasoning_tokens":568}},"tokens_in":478,"tokens_out":667,"duration_ms":7379,"temperature":1.0,"reasoning_tokens":568,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:58:03.102049+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve the same five-day case study by an exact or exhaustive method, for example by evaluating the bi-level objective on a fine grid of admissible daily prices or by using a global MIQP solver with KKT conditions for the lower level, and compare with Algorithm 1's price path. If the exact optimum does not reproduce the 49.4% grid-cost saving and 3.7% profit gain, or if Algorithm 1's result depends on the initial price, the central claim fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the hyper-gradient formula and the alternating-gradient perspective that the paper approximates in Algorithm 1."},{"cited_title":"and Ames, A.D","cited_arxiv_id":null,"evidence_quote":"Demonstrates bi-level optimization for real-time control and is the direct inspiration for the component-wise approximate gradient approach."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the price elasticity dynamics used to argue that lowering price raises demand in make-to-stock systems."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Underpins the affine price elasticity function as the theoretical demand model."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the lithium-ion battery pack manufacturing parameters used to build the case study."},{"cited_title":"PJM-Your Power, Our Grid","cited_arxiv_id":null,"evidence_quote":"Provides the real-time electricity price data used in the five-day simulation."},{"cited_title":"Open energy information","cited_arxiv_id":null,"evidence_quote":"Supplies the solar radiation data and calculator used to estimate onsite PV availability."},{"cited_title":"and Pecht, M","cited_arxiv_id":null,"evidence_quote":"Gives the 24% average ROI figure for comparable solar installations, used to argue the long-term economic case."}],"review_version":1}