{"id":"37b4c4a7-0f2e-4591-8afb-54d35df957d1","arxiv_id":"1908.08570","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Residential electricity demand in Pune rises about 1.5-2% per degree of apparent temperature, while industrial demand barely responds, according to regression and extreme-value models of feeder data.","lead":"A study of five years of Pune electricity data finds that residential demand rises about 1.5-2% for each extra degree of apparent temperature, while industrial demand stays nearly flat. The result gives planners a quantitative sectoral handle on how hot days stress the grid.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Likelihood-ratio test for non-stationary GEV is internally inconsistent: with the correct 2 df, only one division remains significant, undermining the claim that residential peak demand is significantly temperature-driven.","rationale":"The reader's concern about block maxima of daily sums versus true peak load is valid, but the more decisive problem is the likelihood-ratio test's degrees of freedom. Section 3.2.4 and Table 2 are irreconcilable: one says one parameter, the other says two. Applying the correct chi-square reference distribution changes the significance of Kothrud and Shivaji Nagar, which are the residential divisions the abstract highlights. Without a corrected test, the claim that non-stationary GEV models show significant temperature influence on peak residential demand does not follow from the reported numbers. The 1.5-2% figure in the abstract is also not derivable from the tables, and the daily-sum object differs from true peak load, but the LRT inconsistency alone is enough to require reanalysis.","tokens_in":15144,"tokens_out":8605,"duration_ms":87964,"concrete_test":"Re-evaluate the LRT values in Table 2 using chi-square with 2 degrees of freedom (critical value 5.99 at 95%). If the authors intended the df=1 test in Section 3.2.4, they must clarify which models were actually fitted; either way, the reported p-values and significance statements in Section 4.3 need to be recomputed. A simple spreadsheet check of the four LRT values against the two chi-square critical values settles the issue.","verdict_should_be":"REJECT","load_bearing_attack":"The strongest statistical evidence for the abstract's claim is the likelihood-ratio test (LRT) for the non-stationary GEV models. Section 3.2.4 states that the LRT compares Model 1 (μ only) with Model 2 (μ and σ) and uses a chi-square with 1 degree of freedom, critical value 3.84. However, the non-stationary model described in Section 4.3 varies both μ and σ, and Table 2's caption defines the LRT as 2*(nllh(non-stationary) - nllh(stationary)), which adds two parameters and should be compared to chi-square with 2 df (95% critical value 5.99). Under df=2, only Pimpri (LRT=7.30) is significant at 95%; Shivaji Nagar (5.01) and Kothrud (3.61) are not. The residential divisions Kothrud and Shivaji Nagar are the ones the abstract singles out, so the central claim that residential peak demand is significantly influenced by temperature is not supported at the stated significance level. This is not a cosmetic issue; the test's degrees of freedom change which of the four divisions count as temperature-sensitive, and the paper's own text and table cannot both be correct.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes feeder-level electricity demand data for four divisions of Pune (Bhosari, Kothrud, Pimpri, Shivaji Nagar) over 2008–2012, classifies divisions as industrial, residential, or mixed, and relates daily electricity demand to apparent temperature. It uses year-fixed-effects regressions and stationary and non-stationary GEV models with apparent temperature as a covariate. The authors conclude that industrial demand is largely temperature-insensitive, that residential demand shows roughly 1.5–2% change per degree of apparent temperature, and that non-stationary GEV models show peak electricity demand in residential areas is significantly influenced by temperature.","tokens_in":15390,"tokens_out":5016,"duration_ms":47756,"significance":"The disaggregated feeder-level dataset for an Indian city is a useful and relatively rare contribution, and the comparison of industrial versus residential temperature sensitivity is policy-relevant for electricity distribution planning. The fixed-effects regression is a sensible way to control for changing feeder composition, and the use of GEV models with temperature covariates is an appropriate framework for extreme-demand analysis. However, the headline claim about \"peak electricity demand\" is not directly supported by the analysis as written, and the likelihood-ratio test has an inconsistency in its degrees of freedom. If these issues are corrected, the paper could provide a credible estimate of temperature sensitivity of daily electricity demand extremes in Pune.","major_comments":[{"comment":"The block-maxima variable defined in Section 3.2.3 is a standardized daily total electricity demand per block, not an hourly or sub-daily peak load. The equation states that the block maximum is computed from ED_i, where ED_i is \"the daily sum of electricity demand for block i,\" standardized by block mean and standard deviation. The title and abstract claim \"peak electricity demand,\" but the GEV analysis estimates extremes of daily total demand. Please either obtain and use sub-daily peak-load data or revise the title, abstract, and conclusions to refer to extremes of daily total demand rather than peak load.","section":"Section 3.2.3"},{"comment":"The likelihood-ratio test is described in Section 3.2.4 as comparing Model 1 (only μ varying) with Model 2 (μ and σ varying), using a chi-squared distribution with one degree of freedom and a 95% critical value of 3.84. However, Table 2 defines the test statistic as 2*(nllh(non-stationary) - nllh(stationary)), which adds two parameters (μ1 and σ1) and should be compared with a chi-squared distribution with two degrees of freedom, whose 95% critical value is 5.99. Under the correct two-degree-of-freedom test, only Pimpri (7.30) is significant at 95%; Shivaji Nagar (5.01) and Kothrud (3.61) are not. This directly affects the abstract's claim that residential peak demand is significantly influenced by temperature. Please clarify which nested comparison is intended and correct the text and table accordingly.","section":"Section 3.2.4 and Table 2 (GEV table)"},{"comment":"The abstract states that residential activities show \"around 1.5-2% change in average electricity demand with 1 degree rise in AT,\" but no such percentage calculation appears in Section 4 or in Table 2, which reports only slopes in MW. The text after Table 2 says percentage changes will be small but does not provide the computed percentages or confidence intervals. Please add the percentage-change calculation with uncertainty bounds, or remove the quantitative claim from the abstract.","section":"Abstract and Section 4 regression results"},{"comment":"The conclusion states that the study \"predicts that there will be a significant rise of peak electricity load as a consequence of the increase in temperatures,\" but the paper does not present future temperature scenarios, climate projections, or a prediction exercise. The analysis estimates historical temperature sensitivity only. Please temper the prediction language or add an explicit scenario-based projection.","section":"Section 5 (Conclusion)"}],"minor_comments":[{"comment":"The section numbering is inconsistent: Section 3.2.1 is followed by Section 3.2.3, and Section 4.1 is followed by Section 4.3 with no Section 4.2. Please renumber the sections.","section":"Section numbering"},{"comment":"The paragraph beginning \"The maximum electricity demand of 15 days non-overlapping block is fitted to the GEV distribution...\" is repeated verbatim, and the second repetition contains an incomplete sentence. Please remove the duplication.","section":"Section 4.3"},{"comment":"Table 2 is used twice: once for the year-fixed-effects regression results and once for the GEV parameter estimates. Please renumber the tables and update all cross-references.","section":"Table numbering"},{"comment":"Figure 5 is captioned as \"Q–Q plots obtained by plotting values from the GEV stationary model against the empirical values,\" but the surrounding text in Section 4.3 refers to diagnostic plots for the non-stationary model. Please clarify which model the figure displays.","section":"Figure 5 caption"},{"comment":"The Steadman equation for apparent temperature is garbled in the text: \"T T0.33 0.7 A = a + * e − * w − 4\" is not readable as a formula. Please typeset the equation properly and define all variables.","section":"Section 3.2.1"},{"comment":"The statement that the non-stationary model \"would assume that present day temperature is dependent on the previous day's temperature\" is inaccurate; the non-stationary GEV model here uses apparent temperature as a covariate and does not impose an autoregressive structure. Please correct this description.","section":"Section 3.2.3"},{"comment":"Some references have inconsistent formatting and missing bibliographic details (e.g., Murari et al. 2015, Revadekar et al. 2012 are cited in the text but the reference list entries are incomplete or misnumbered). Please standardize the reference list.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript addresses a worthwhile question with an interesting disaggregated dataset, but the statistical presentation needs substantial revision. The two load-bearing issues—the mismatch between the block-maxima variable and the \"peak demand\" claim, and the degrees-of-freedom inconsistency in the likelihood-ratio test—are fixable in a revision, so I recommend major revision rather than rejection. I did not find evidence of questionable research practices, but the paper would benefit from careful checking of table numbering, duplicated text, and claim calibration."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, before you cite this paper, know two things. The sectoral dataset is genuinely new, but the GEV likelihood-ratio test that supports the headline abstract claim has an internal inconsistency that flips the conclusion.\n\nThe LRT as run compares a stationary GEV to a non-stationary GEV with both location and log-scale varying with temperature. That's two added parameters, so the 95% critical value is 5.99, not the 3.84 used in Section 3.2.4. With the correct df=2, only Pimpri (LRT=7.30) is significant at 95%. Shivaji Nagar (5.01) and Kothrud (3.61) are not. The paper's own text claims Shivaji Nagar significant at 95% and Kothrud at 90%; under df=2, neither holds. Since those are the residential divisions the abstract points to, the claim that residential peak demand is significantly temperature-driven is not supported by the reported test. This is a load-bearing flaw, though the regression analysis still stands.\n\nWhat's new and useful: feeder-level disaggregation for an Indian subtropical city, with a plausible division classification from sheddable/non-sheddable feeders. The year-fixed-effects regression shows residential slopes (40.8, 43.4 MW/°C) about 2.5x the industrial Bhosari slope (15.9), and the in-sample fit is good. That sectoral contrast is a real empirical contribution for utility planners.\n\nSoft spots aside from the LRT: the 'peak' is actually block maxima of standardized daily total demand, not hourly peak load, so the title overreaches. The 1.5-2% per degree figure does not appear anywhere in the tables—only MW slopes are given. The global-warming framing is an extrapolation without scenario runs or return-level estimates. No code or data are archived.\n\nOverall: this is useful applied science with a questionable central statistical test. The paper deserves peer review because the dataset is unique and the sectoral question matters, but a serious referee should demand a corrected LRT, a peak-load definition matching the claim, and a traceable percentage sensitivity.\n\nFor you: read if you work on India load forecasting or climate-demand; otherwise it's a cautionary example of an otherwise decent empirical study undermined by a degrees-of-freedom mistake.","headline":"The sectoral Pune load–temperature dataset is new and the regression results are believable, but a degrees-of-freedom error in the GEV likelihood-ratio test invalidates the paper's headline claim about residential peak demand.","tokens_in":15926,"tokens_out":4550,"would_cite":true,"duration_ms":40362,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62G32","62P12","62F03"],"pacs":[],"model":"deepseek-v4-flash","headline":"Using five years of feeder-level electricity data from Pune, this paper claims that residential electricity demand is temperature-sensitive—about 1.5-2% per 1°C of apparent temperature—while industrial demand is not, and that…","keywords":["electricity demand","peak load","apparent temperature","extreme value theory","GEV","Pune","global warming","sectoral analysis"],"falsifier":"Take the same five years of Pune feeder data, extract the true hourly or sub-daily peak demand for each block, and fit the same stationary and non-stationary GEV models with apparent temperature as covariate. If the likelihood-ratio statistics and temperature slopes for residential divisions weaken or vanish, the paper's peak-load claim would fail.","tokens_in":1705,"feed_emoji":"⚡","tokens_out":4302,"duration_ms":86532,"temperature":0.7,"pith_summary":"This paper asks whether rising temperatures from global warming push up peak electricity demand in a subtropical Indian city, and whether the effect differs between industrial and residential consumers. Using five years of feeder-level electricity data from Pune, the authors separate divisions by their mix of sheddable and non-sheddable feeders and compare electricity demand with apparent temperature. They find industrial electricity use is largely insensitive to temperature, while residential demand rises by roughly 1.5-2% per 1°C of apparent temperature on average. Applying stationary and non-stationary generalized extreme value models to block maxima of standardized daily demand, they find apparent temperature significantly explains peak demand in residential and mixed divisions but not in the industrial division. If correct, climate-driven warming would translate into rising peak loads mainly from homes, a signal for electricity infrastructure planning.","feed_headline":"Pune's residential power demand climbs 1.5-2% per extra degree","feed_subtitle":"Extreme-value analysis of feeder-level data links peak electricity load to apparent temperature in homes, not industry.","key_machinery":"The central mechanism is the generalized extreme value (GEV) distribution applied to block maxima: the distribution for the largest standardized demand in each week, fortnight, or month. The non-stationary version lets the location and log-scale be linear functions of apparent temperature, $\\mu(t) = \\mu_0 + \\mu_1 t$ and $\\log\\sigma(t) = \\sigma_0 + \\sigma_1 t$, and compares nested models using the likelihood-ratio statistic $2(\\mathrm{nllh}(M_1)-\\mathrm{nllh}(M_2))$ against a $\\chi^2_1$ reference with a 95% threshold of 3.84. Apparent temperature itself is built from daily average temperature, humidity, wind speed, and water vapor pressure using the Steadman formula, so the covariate carries both heat and moisture. This machinery decides whether temperature explains the peaks, not just the average, of electricity demand.","core_discovery":"The central claim is that in Pune over 2008-2012, industrial electricity demand is essentially decoupled from temperature while residential demand is temperature-sensitive, and that the temperature signal reaches the upper tail of demand. The evidence is a year-fixed-effects regression giving slopes of 15.9 MW/°C for the industrial Bhosari division versus 40.8, 43.4, and 45.7 MW/°C for residential and mixed divisions, summarized in the abstract as a 1.5-2% change in average residential demand per degree. In the extreme-value analysis, allowing the GEV location and log-scale to depend linearly on apparent temperature improves the fit for Kothrud at the 90% significance level and for Pimpri and Shivaji Nagar at the 95% level, but not for Bhosari. The authors conclude that non-stationary GEV models with apparent temperature as a covariate can capture climate-driven peak electricity load in residential areas.","pith_inferences":["The reported 1.5-2% average residential demand sensitivity probably understates the peak sensitivity, because cooling load concentrates in hot afternoon hours; fitting the same GEV models to hourly or sub-daily maxima would test this directly.","As income and air-conditioning penetration rise, the residential temperature slope may steepen, making the 2008-2012 estimates a plausible lower bound for future warming impacts in Pune.","If the true system peak is sub-daily, analyzing daily total demand could misdescribe peak behavior; comparing daily-sum block maxima with true hourly peaks would show whether the method needs adjustment.","Applying the same apparent-temperature GEV approach to other tropical and subtropical cities could reveal whether a common residential load-temperature response curve emerges across humid climates."],"forward_implications":["Residential peak load in Pune can be expected to rise as global warming increases apparent temperature, while industrial peak demand should remain nearly flat.","Utilities can use fitted non-stationary GEV models with temperature as a covariate to estimate return levels of peak demand under warmer future climates.","City-level aggregation would dilute or hide the residential temperature sensitivity, so sectorally disaggregated feeder data is needed for accurate climate-impact assessment.","The same workflow can be transferred to other subtropical and humid cities with feeder-level electricity data and station-level weather data.","Accounting for climate-driven peak load would support planning for distribution capacity, demand response, and cooling-related infrastructure investment."],"supporting_citations":[{"why":"Supplies the GEV and block-maxima framework, including the likelihood-ratio testing logic used to compare stationary and non-stationary models.","marker":"Coles et al., 2001"},{"why":"Provides the formula used to construct apparent temperature from dry-bulb temperature, humidity, wind speed, and water vapor pressure.","marker":"Steadman (1994)"},{"why":"Defines the feeder classification into sheddable and non-sheddable categories that the paper uses to identify industrial versus residential divisions.","marker":"MAHADISCOM, 2012"},{"why":"Gives the Thai estimate of roughly 4.6% peak-demand increase per degree, an international benchmark for the Pune peak-load results.","marker":"Parkpoom et al 2008"},{"why":"Supports the expectation that industrial electricity demand is largely insensitive to temperature, which the Pune industrial division confirms.","marker":"Moral-Carcedo et al. 2015"},{"why":"Provides the cross-country range of temperature sensitivity of electricity demand against which the 1.5-2% residential estimate is interpreted.","marker":"Cian et al. 2007"}],"fun_headline_variants":["Home power in Pune feels heat, industry doesn't","Temperature explains peak power in Pune homes, not factories","Residential power use in Pune jumps with temperature, industry flat","Extreme value stats show Pune homes react to heat, not industry","Pune homes: 1.5-2% power jump per degree, industry unaffected"],"cache_read_input_tokens":18048,"weakest_assumption_plain":"The load-bearing assumption is that the block-by-block maximum of each day's total electricity use stands in for peak electricity demand. If the true peak that stresses the grid is the highest hourly or sub-daily load, the extreme-value model targets a different quantity.","fun_headline_variants_meta":{"raw":{"variants":["Home power in Pune feels heat, industry doesn't","Temperature explains peak power in Pune homes, not factories","Residential power use in Pune jumps with temperature, industry flat","Extreme value stats show Pune homes react to heat, not industry","Pune homes: 1.5-2% power jump per degree, industry unaffected"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001209,"raw_usage":{"total_tokens":5012,"prompt_tokens":1010,"completion_tokens":4002,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":626,"completion_tokens_details":{"reasoning_tokens":3913}},"tokens_in":626,"tokens_out":4002,"duration_ms":26263,"temperature":1.0,"reasoning_tokens":3913,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:36:16.359084+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the same five years of Pune feeder data, extract the true hourly or sub-daily peak demand for each block, and fit the same stationary and non-stationary GEV models with apparent temperature as covariate. If the likelihood-ratio statistics and temperature slopes for residential divisions weaken or vanish, the paper's peak-load claim would fail.","supporting_citations":[],"review_version":1}