{"id":"295dfcf1-592f-41c1-a9d6-a15393d99af0","arxiv_id":"2505.09235","paper_version":2,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A genetic algorithm chooses customer phase reassignments to reduce load imbalance and end-of-line voltage drop while keeping the number of changed connections small, tested on a simulated single-circuit low-voltage network.","lead":"In low-voltage power grids, uneven loads across the three phases cause losses and poor voltage at the end of the line. This paper uses a genetic algorithm to choose which customers to move to a different phase, minimizing the number of moves, and tests it on a simulated 60-customer network modeled after Tucumán, Argentina.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Static 200 W load assumption is the load-bearing weakness: the abstract promises 'optimal' re-phasing under realistic conditions, but the algorithm is only evaluated at a single fixed load point, and the paper's own validation text acknowledges this is a 'caso teo´rico de prueba.'","rationale":"The paper's central claim is essentially a promise about operational benefit: an 'optimal' set of re-phasing changes that improves balance and voltage while minimizing changes. The reader's weakest-assumption identification—constant customer loads—is exactly the point where the claim is least secure. The full text contains no quantitative results, no baseline comparison, and no sensitivity analysis; the only evaluation described is a single static load case, which the authors themselves label as a theoretical test case. A phase plan derived from one operating point can be arbitrarily bad at another operating point, so the missing robustness check is not a stylistic gap but a logical one for the claim as stated. I agree with the reader's judgment that the approach is plausible but unverified, and CONDITIONAL is the appropriate verdict: the authors should supply the missing evaluation, including at least one additional load scenario or a time-series test, plus the actual results and baseline comparisons, before the abstract's 'optimal' language can be accepted. The concern is not that the algorithm is wrong internally; it is that the evidence presented is insufficient to support the generality of the claim. The proposed concrete test would settle whether the static-load assumption is indeed load-bearing or harmless, because it directly measures how a plan optimized at 200 W performs when the load changes.","tokens_in":5776,"tokens_out":1822,"duration_ms":16623,"concrete_test":"Run the same genetic algorithm on a simple two-step load profile with the same 60-customer, 250 kVA, 15 mm² cable network: first evaluate with all loads at 200 W/20 var; then evaluate the same optimized phase assignment at a second operating point, e.g., 600 W/60 var for all customers or a 30% random per-customer variation. If the objective (weighted combination of imbalance B, maximum end-of-line voltage drop PV, and number of changes N) at the second point is worse than the initial 112233 baseline, the constant-load assumption is load-bearing and the abstract's 'optimal' claim needs qualification. A simpler analytical check: compute PV at the end node for a fixed phase assignment as load scales; if a plan optimized at 200 W yields a voltage drop larger than the unoptimized baseline at 600 W, the plan is not robust.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that the genetic algorithm can 'optimally identify which loads should be reassigned' to improve phase balance and end-of-line voltage while minimizing changes. For that claim to hold in any operational sense, a plan computed from the optimization must remain beneficial under real, time-varying load patterns. The paper's only evaluation uses a constant 200 W active and 20 var reactive load per customer throughout the period (page with Ecuación 4-6, 'Validación del algoritmo'), and the text itself flags this as below typical residential values and only 'adecuados como caso teórico de prueba.' No time-series, no load profiles, no sensitivity analysis, and no robustness check appear in the available text. Moreover, the fitness function's B, PV, and N objectives are normalized by hand-chosen constants (Bmax=100, PVmax=10, Nmax=50) and combined with weights r, t, v that are also set a priori, with no reported results showing how the choice affects the solution. The strongest claim is not supported by the presented evidence; the constant-load assumption is the most load-bearing weakness because if real loads fluctuate, a re-phasing plan optimized at one operating point can worsen both imbalance and voltage at other times (e.g., evening peaks vs. midday). This is internally flagged: the paper calls the setup 'caso teórico de prueba.' The missing piece is not a consensus disagreement but a quantitative robustness gap between the claim and the test.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a genetic-algorithm-based method for reassigning single-phase customers across the three phases of a low-voltage distribution transformer. Three objectives are defined: phase imbalance (Eq. 2), maximum voltage drop over the analysis period (Eq. 4), and the number of customer phase changes (Eq. 5), combined into a scalar fitness function (Eq. 6). The algorithm is tested in a PandaPower model of a simple radial feeder with 60 residential customers in Tucumán, initially with constant 200 W active and 20 var reactive loads per customer. The available text stops before presenting any post-optimization results or comparisons with the initial network configuration.","tokens_in":6068,"tokens_out":4305,"duration_ms":41759,"significance":"If fully validated, the approach would be practically useful for distribution utilities with smart meters, because the change-count term directly minimizes field work and the formulation is easy to adapt. The objective definitions are coherent and the use of an open-source power-flow tool is a strength. However, the current manuscript provides no quantitative evidence of improvement, and the single-point constant-load test is acknowledged by the authors as only a theoretical test case. The practical significance of the method therefore cannot be assessed from the presented results.","major_comments":[{"comment":"The manuscript does not contain a results section: after defining the test network and the 112233 connection scheme, the text stops before reporting the optimized values of B, PV, and N or the pre-optimization baseline. Consequently, the abstract's claim that the genetic algorithm 'optimally identifies' which loads should be reassigned is not supported by any quantitative outcome in the submitted version.","section":"Validación del algoritmo"},{"comment":"Equations 2 through 6 are evaluated only at a single operating point: every customer draws a constant 200 W with 10% reactive power, a value the authors themselves describe as 'adecuados como caso teórico de prueba.' Real residential loads fluctuate over time, and an optimum at one loading point can worsen imbalance or voltage at other times. The paper should include at least a time-series or multi-scenario test (for example, peak versus off-peak loading) to demonstrate robustness; without this, the operational claim is not established.","section":"Validación del algoritmo"},{"comment":"The fitness function depends on hand-chosen normalization constants Bmax=100, PVmax=10, and Nmax=50, and on weights r, t, v that are set a priori. The manuscript provides no sensitivity analysis or empirical rationale tying these values to the Tucumán network data, so any reported solution would be conditional on arbitrary parameter choices. A sensitivity study, or a comparison with a Pareto-front multi-objective formulation, is needed to support the optimality claim.","section":"Combinación de objetivos, Eq. (6)"},{"comment":"The paper does not compare the genetic algorithm's solution against the original connection scheme or against a simpler baseline such as a greedy phase-balancing heuristic. Because the fitness function is maximized by construction, an improvement in f(S) alone is not evidence of practical benefit; a before/after table reporting B, PV, N and, ideally, neutral current or losses is required.","section":"Validación del algoritmo"}],"minor_comments":[{"comment":"The manuscript contains corrupted accents and replacement glyphs in several places, including the title ('L´ÕQHD') and some equations; a careful proofreading pass is needed.","section":"General"},{"comment":"Equation 4 uses the maximum over t but does not explicitly define the index range n or the meaning of the subscript in 'PV = max{PVt}_n'; the sample interval and period should be stated.","section":"Ecuación 4"},{"comment":"The test network parameters list transformer rating, cable type, and geometry, but the conductor impedance values used in the PandaPower model are not provided; these should be included for reproducibility.","section":"Validación del algoritmo"},{"comment":"Figure 5 is referenced in the text but does not appear in the submitted extract; the authors should ensure that all figures are included and correctly numbered in the final version.","section":"Figuras"},{"comment":"Several references are incomplete (for example, [4] and [15]) and [15] is a TechRxiv preprint; the authors should complete the bibliographic data and, where possible, cite peer-reviewed versions.","section":"Referencias"}],"recommendation":"major_revision","confidential_remarks":"The version under review appears to be an incomplete extract from the conference proceedings: the results section is missing and Figure 5 is absent. If the full paper exists, the editor should request it before making a decision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a short conference paper, not a full study. It proposes a genetic algorithm for phase rebalancing that minimizes a weighted sum of phase imbalance, end-of-line voltage drop, and number of customer reconnections. The objective is coherent, and the authors are honest that the test case is a toy. But the text stops before any results, so the abstract's claim that the algorithm \"optimally\" identifies rephasing is unsupported.\n\nWhat is actually new is modest. The core idea—a discrete GA to re-phase customers for load balance while penalizing changes—is already in Homaee et al. (ref [9]) and in the authors' own earlier work (ref [11]). The contribution here is adding end-of-line voltage drop as a third objective and folding the three terms into a convex combination with hand-set weights. That is a reasonable engineering extension, not a conceptual breakthrough. The definitions of imbalance, voltage drop, and number of changes are clean, and the paper correctly cites prior art rather than ignoring it.\n\nThe soft spots are significant. Most importantly, there is no evaluation. The validation section describes a 60-customer, single-feeder network with every customer at a constant 200 W plus 20 var, and the authors themselves call it \"caso teórico de prueba.\" Real loads fluctuate. A rephasing plan computed at one flat operating point can worsen imbalance and voltage at other times. Without time-series loads or at least sensitivity analysis, the \"optimal\" language is not justified. Also, the weights r, t, v and the normalization constants Bmax, PVmax, Nmax are fixed a priori, with no showing that results are robust to those choices. That is a moderate problem, not a fatal one; any field implementation would tune them. The missing results section is the bigger issue, and it may be an artifact of the truncated proceedings text. If the full paper contains the results, this criticism partly evaporates, but based on what is in front of us, the central claim hangs on a single static test case.\n\nWho is this for? Practitioners in distribution utilities, particularly in Tucumán, who want a quick, low-cost rephasing plan supported by smart meter data. It is not ready for a rigorous journal. I would not send this to peer review in its current form, but I would encourage the authors to add the missing evaluation, a time-varying load scenario, and robustness checks, then resubmit as a full paper.","headline":"A sensible but incomplete conference short-paper: the GA objective is coherent and the authors are candid about the toy test case, but the missing results section leaves the central claim unsupported.","tokens_in":6623,"tokens_out":2865,"would_cite":false,"duration_ms":29171,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A genetic algorithm can identify the fewest customer phase changes that improve both phase balance and end-of-line voltage.","keywords":["phase balancing","load re-phasing","genetic algorithm","low-voltage distribution","voltage drop","service quality","smart meters","distribution transformer"],"falsifier":"Run the algorithm on the same 60-customer network and 112233 connection scheme, but replace the fixed 200 W loads with a measured 24-hour smart-meter profile from a real low-voltage circuit; then apply the recommended phase reassignment. If any hour of the day shows the end-of-line voltage drop or the imbalance index worse than the original connection, the claim that the method improves both balance and voltage under real conditions is false.","tokens_in":5548,"feed_emoji":"⚡","tokens_out":11897,"duration_ms":104573,"temperature":0.7,"pith_summary":"Unbalanced loads on a three-phase distribution circuit waste energy, stress equipment, and pull down voltage at the far end of the line. The paper proposes a genetic algorithm whose search space is the assignment of each customer to one of the three phases; each candidate plan is scored by how much it improves phase balance, how much it reduces the end-of-line voltage drop, and how few customers have to be reconnected. The authors say this yields the optimal set of re-phasing moves, so a utility with smart-meter data can turn the result into a concrete low-cost field action. The method is validated on a simplified 60-customer, single-circuit network with realistic transformer and cable parameters, under the assumption that every customer draws the same constant load.","feed_headline":"Genetic algorithm balances load and voltage with fewest phase changes","feed_subtitle":"A weighted genetic search tells a utility which customers to switch to improve end-of-line service quality.","key_machinery":"The carrying mechanism is the genetic algorithm plus its fitness function: $f(S) = 1 - \\frac{r B/B_{\\max} + t PV/PV_{\\max} + v N/N_{\\max}}{r+t+v}$. Here $B$ is the phase-imbalance index over the analysis period, $PV$ is the maximum voltage drop at the final node, $N$ is the number of customers whose phase changes, and $B_{\\max}, PV_{\\max}, N_{\\max}$ are expected maxima that normalize the three objectives. The coefficients $r,t,v$ make the fitness a convex combination, so the operator can decide whether balance, voltage, or few changes matters more. The genetic algorithm's chromosome is the phase label (1, 2, or 3) for each customer, and the search returns the assignment with the highest fitness.","core_discovery":"The central claim is that a single weighted fitness function can guide a genetic algorithm to phase-reassignment plans that improve both load balance and voltage quality at the end nodes while minimizing the number of changes. Each solution $S$ is an assignment of every customer to a phase; the fitness $f(S)$ combines the period imbalance index $B$, the worst end-of-line voltage drop $PV$, and the number of phase changes $N$, each normalized by expected maxima, with weights the operator chooses. Maximizing this fitness selects plans that trade the three objectives off in a controllable way. Evaluated with a power-flow simulation on a 60-customer circuit built from real low-voltage network characteristics, the algorithm identifies which loads to move, to which phase, and how many visits the work requires.","pith_inferences":["A rolling implementation is a natural extension: re-run the search on each day's smart-meter profiles and re-phase only when the optimized plan changes enough to justify the visits.","The normalized weighted objective makes it easy to add other operational goals, such as limiting neutral current or cutting resistive losses, by inserting one more normalized term into the fitness function.","Robustness to mislabeled phases is untested; adding random noise to the assumed phase of a fraction of customers would show how much error the planning step tolerates.","The static-load evaluation covers one network shape; testing the same fitness function on longer feeders, multiple circuits on one transformer, or networks with distributed generation would show whether the method scales."],"forward_implications":["A utility can turn the algorithm's output directly into a work order: the list of customers whose phase connection changes, and the target phase for each.","Operators can tune the three weights to favor service quality (balance and voltage) or cost (fewer field changes) before running the search.","The same weighted objective can be applied to any transformer circuit once its $B_{\\max}$, $PV_{\\max}$, and $N_{\\max}$ constants are set to match that circuit's expected ranges.","Balancing is achieved by reconnecting existing service drops rather than adding capacitors or filters, so the main cost is the number of physical changes, which is what the algorithm minimizes."],"supporting_citations":[{"why":"Prior phase-reassignment formulation for low-voltage networks that the genetic algorithm here adapts and optimizes.","marker":"[11]"},{"why":"Supplies the data-driven method for identifying which customer is on which phase, the input the re-phasing search needs.","marker":"[12]"},{"why":"Supplies the open-source power-flow simulation tool used to evaluate candidate phase-reassignment plans.","marker":"[19]"},{"why":"Defines the transformer, conductor, and service-cable parameters used to build the test network.","marker":"[4]"},{"why":"Provides an existing discrete-genetic-algorithm approach to customer re-phasing that the proposed weighted-objective GA builds on.","marker":"[9]"}],"fun_headline_variants":["AI finds minimal phase swaps to balance grid loads","Genetic search optimizes phase reassignment for voltage and balance","Fewest phase changes to balance loads and boost end-of-line voltage","Genetic algorithm trims phase changes while improving grid balance and voltage"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"All 60 customers are modeled as drawing exactly 200 W active power and 10% of that value in reactive power for the whole period, so the recommended re-phasing is tuned to a constant-load world; with fluctuating real loads it could fail to improve balance or voltage, or even make them worse.","fun_headline_variants_meta":{"raw":{"variants":["AI finds minimal phase swaps to balance grid loads","Genetic search optimizes phase reassignment for voltage and balance","Fewest phase changes to balance loads and boost end-of-line voltage","Genetic algorithm trims phase changes while improving grid balance and voltage"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000506,"raw_usage":{"total_tokens":2416,"prompt_tokens":839,"completion_tokens":1577,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":455,"completion_tokens_details":{"reasoning_tokens":1509}},"tokens_in":455,"tokens_out":1577,"duration_ms":9192,"temperature":1.0,"reasoning_tokens":1509,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:36:11.189126+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the algorithm on the same 60-customer network and 112233 connection scheme, but replace the fixed 200 W loads with a measured 24-hour smart-meter profile from a real low-voltage circuit; then apply the recommended phase reassignment. If any hour of the day shows the end-of-line voltage drop or the imbalance index worse than the original connection, the claim that the method improves both balance and voltage under real conditions is false.","supporting_citations":[{"cited_title":"Phase Reassignment for Load Balance in Low- Voltage Dis- tribution Networks","cited_arxiv_id":null,"evidence_quote":"Prior phase-reassignment formulation for low-voltage networks that the genetic algorithm here adapts and optimizes."},{"cited_title":"A new data- driven method based on Niching Genetic Algorithms for phase and substation identification","cited_arxiv_id":null,"evidence_quote":"Supplies the data-driven method for identifying which customer is on which phase, the input the re-phasing search needs."},{"cited_title":"‹•‡¿‘• †‡ ‹•–”‹„— ‹‘ † ‘ ƒ„Ž‡ Preensamblado, March 2015","cited_arxiv_id":null,"evidence_quote":"Defines the transformer, conductor, and service-cable parameters used to build the test network."},{"cited_title":"A practical approach for distribution network load balancing by optimal re-phasing of single phase customers using discrete genetic algorithm","cited_arxiv_id":null,"evidence_quote":"Provides an existing discrete-genetic-algorithm approach to customer re-phasing that the proposed weighted-objective GA builds on."}],"review_version":1}