{"id":"4b63f529-9cda-42bb-9d63-d2702a9f2514","arxiv_id":"2506.12062","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"In lossless CEED tests on the IEEE 30-bus system, particle swarm optimization is reported to converge in about a quarter to a third of the genetic algorithm's time, with slightly lower total cost.","lead":"This paper reviews ten nature-inspired optimization methods for combined economic and emission dispatch, then reports a MATLAB comparison of particle swarm optimization and a genetic algorithm on the IEEE 30-bus test system. It finds the swarm version runs about three to four times faster and gives a slightly lower combined fuel and emission cost in the two load cases tested.","discovery_kind":"incremental","skeptic_critique":{"model":"deepseek-v4-flash","headline":"PSO's reported 1500-MW solution violates the stated load-balance constraint (sum of P_i = 1499.4 MW), so the headline speed/optimality comparison rests on an infeasible dispatch.","rationale":"The reader's weakest assumption focused on tuning fairness and absence of statistical spread. While those issues are real, the more decisive problem is internal: the PSO solution reported for PD=1500 MW does not satisfy the paper's own equality constraint. This is not a benchmarking-philosophy disagreement; it is a concrete arithmetic inconsistency that invalidates one of the two experimental comparisons supporting the central claim. Because the reader's verdict is already CONDITIONAL, and this concern strengthens the need for corrections rather than introducing a wholly new verdict category, the verdict remains CONDITIONAL (UNCHANGED). The concrete test would settle whether the claim survives after fixing the infeasible dispatch and adding an optimum baseline.","tokens_in":8894,"tokens_out":5234,"duration_ms":53601,"concrete_test":"Recompute the sum of the reported PSO generation values at PD=1500 MW in the Table; if it is 1499.40 rather than 1500.00, re-run the PSO experiment with an exact equality-constraint repair (e.g., slack reallocation or penalty) and compare the repaired cost against GA. Independently solve the same lossless 6-generator CEED problem with lambda-iteration or quadratic programming to obtain the true optimum. If PSO's repaired solution is not below GA's and/or not at the true optimum, the 'global optimal' and faster-convergence claims fail.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that PSO converges faster to the global optimal solution than GA. The table in Section 5 lists PSO generation for PD=1500 MW as P1..P6 = 195.79, 256.55, 381.25, 81.69, 381.85, 202.27, which sum to 1499.40 MW, not 1500 MW. Section 5 states transmission losses are not accounted for, and Section 2 Eq. (5) requires sum(P_i) = PD, so this solution is infeasible by 0.6 MW—far larger than any 2-decimal rounding artifact. An under-generating solution can trivially lower fuel and emission costs, so the PSO-vs-GA comparison at 1500 MW is invalid. The other three rows (GA 1500, PSO 2000, GA 2000) sum exactly to the required load, making this a targeted infeasibility rather than a general rounding issue. Furthermore, no independent optimum baseline is provided, so 'global optimal' is unsupported even for the feasible 2000-MW case. The reported total-cost difference (33948.83 vs 34005.52, ~0.17%) is small and could be explained by the constraint violation. The conclusion's strongest claim therefore does not follow from the reported data.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a combined review and implementation study of artificial intelligence techniques for the combined economic emission dispatch (CEED) problem. The review portion surveys ten metaheuristic methods (PSO, GA, ABC, ACSA, FA, SFLA, EP, SA, GSA, BBO) and summarizes prior CEED applications. The implementation portion formulates CEED with a weighted penalty-factor objective, applies PSO and GA to a six-generator IEEE 30-bus system for load demands of 1500 MW and 2000 MW, and compares the two algorithms in terms of total cost (fuel plus emission penalty) and wall-clock time. The central claim, stated in Section 6, is that PSO converges faster to the global optimal solution than GA, based on the reported table and convergence figures.","tokens_in":9149,"tokens_out":3113,"duration_ms":30256,"significance":"If the numerical comparison were sound, the paper would provide a straightforward empirical benchmark of two canonical metaheuristics on a classic CEED test case, along with a useful literature digest. The review portion is a reasonable, if introductory, catalog of ten AI methods, and the paper honestly states that transmission losses are neglected. However, the empirical contribution is weakened by the absence of any optimality baseline, statistical spread, or reproducibility details, and, most seriously, by an infeasible reported dispatch. The significance of the headline claim is therefore limited: a corrected, properly benchmarked comparison would be a modest but useful data point, but in its current form the paper does not establish the claimed PSO superiority.","major_comments":[{"comment":"The PSO solution for PD=1500 MW violates the load balance constraint stated in Eq. (5). The reported P1..P6 values (195.79, 256.55, 381.25, 81.69, 381.85, 202.27) sum to 1499.40 MW, not 1500 MW, and since transmission losses are explicitly not accounted for, Eq. (5) requires exact equality. An under-generating dispatch of 0.6 MW can trivially reduce fuel and emission costs, so the PSO-vs-GA comparison at 1500 MW is invalid. The other three rows in the table sum exactly to their required loads, indicating a targeted infeasibility rather than simple rounding. This directly undermines the conclusion's strongest claim.","section":"Section 5, Table"},{"comment":"The claim that PSO 'converges faster to global optimal solution' is unsupported because no independent optimum is provided. There is no comparison with an established optimal solution for this test system (e.g., lambda-iteration, dynamic programming, or best-known values from the literature), and no convergence criterion is defined; the fixed 500-iteration budget is a stopping rule, not evidence of convergence. The term 'global optimal' should either be replaced by 'best solution found' or justified by a reference baseline.","section":"Section 6, Conclusion"},{"comment":"The results from 50 stochastic trials are summarized only by a single value for each quantity, with no standard deviation, best/worst bounds, or statistical significance test. The reported total-cost difference between PSO and GA at 1500 MW (33948.83 vs 34005.52, about 0.17%) may be smaller than the run-to-run variability of either algorithm. Report mean ± standard deviation over the 50 trials, and ideally a paired test, before claiming that PSO outperforms GA.","section":"Section 5, Simulation Results"},{"comment":"The description of the price-penalty-factor calculation is unclear: the sentence 'These obtained his are listed in increasing order and added with Pi,max of every generator one by one starting from the first hi in the list until Σ Pi,max ≥ PD' does not define a reproducible algorithm. Please rewrite this step-by-step, and show explicitly how the penalty factors hNOX, hCOX, and hSOX for PD=1500 MW and 2000 MW are computed from Eq. (8) and the coefficients taken from [3].","section":"Section 2, Eq. (8) and following procedure"}],"minor_comments":[{"comment":"The sentence 'The solutions with average operating time (t) have been selected out of 50 trials' is grammatically unclear; it should state how the reported generation values were selected (e.g., from the trial with median cost or median time).","section":"Section 5, Table"},{"comment":"There are several typographical errors: 'Biogeoraphy' should be 'Biogeography', 'offsrings' should be 'offsprings', and 'dictates' in Section 6 should be 'indicate' or 'show'.","section":"Section 3.x"},{"comment":"The convergence figures are not described in the text; specify whether the vertical axis is total cost or fitness value, and describe the convergence behavior that the figures are meant to illustrate.","section":"Figures 1 and 2"},{"comment":"The claim that 'classical optimization methods lack the ability' to solve the multi-objective problem is made without a citation or a direct comparison; please support or soften this statement.","section":"Section 1, Introduction"}],"recommendation":"major_revision","confidential_remarks":"The paper appears to be a re-posted journal article (Sci.Int.(Lahore), 2016) and its numerical section is brief. The infeasible PSO dispatch at 1500 MW is a concrete, checkable error that the authors must correct; if the corrected results change the ranking, the conclusion should be revised accordingly. I do not see evidence of circularity or data fabrication, but the lack of statistical reporting and the absence of an optimum baseline are real weaknesses. A major revision with rerun simulations, variance reporting, and a reference optimum would bring the empirical claim to a supportable level."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a competent restatement of a 2016 review-and-demo paper, and its own table undercuts its headline claim. The PSO result for PD = 1500 MW violates the load-balance constraint stated in Eq. (5): the six P_i values sum to 1499.40 MW, not 1500 MW. The other three rows sum correctly. So the reported PSO-vs-GA comparison at 1500 MW pits a slightly under-generating dispatch against a feasible one; the lower cost and faster convergence at that load do not follow.\n\nWhat the paper does well: the survey of AI techniques (PSO, GA, ABC, ACSA, FA, SFLA, EP, SA, GSA, BBO) is organized and readable, and the implementation follows the penalty-factor procedure from the literature with the coefficient source stated. The 2000 MW rows are feasible, and the PSO/GA ranking there is plausible and consistent with earlier work. The paper also reports its hyperparameters and trial count, which is more than many demo papers do.\n\nSoft spots, in order of severity. First, the infeasible row above: 0.6 MW is not a rounding artifact, and an under-generating solution can trivially lower fuel and emission costs, so that row cannot be used as evidence. Second, \"global optimal\" appears in the conclusion with no baseline; no conventional-method reference solution or known optimum is given, and with only 10 particles or individuals over 500 iterations, global optimality is not established. Third, results are summarized without variance or error bars, so the 0.17% cost difference at 1500 MW and the timing comparison are hard to evaluate. Fourth, there is no code or raw data, so the numbers cannot be independently reproduced from the paper. The citation practice is also sloppy: originators like Kennedy/Eberhart and Goldberg/Holland are cited through application papers [3] and [5], which makes the review harder to use as a map of the field.\n\nI agree with the reader's conditional verdict but would push it further: the central comparison is invalid at one of the two test loads, so acceptance requires fixing the constraint violation, re-running the 1500 MW case, adding dispersion and a baseline, and removing the global-optimality language. The 2000 MW comparison might survive.\n\nWho this is for: someone teaching a first course in metaheuristics for dispatch could use the review sections as a short reading, and the table is a useful classroom example of why you verify constraint satisfaction in stochastic optimization. As a research contribution, it adds nothing over the cited literature.\n\nRecommendation: desk reject or, at most, send to a referee only if the venue has a strong interest in reproducible demo or negative results. I would not recommend sending this to a serious archival venue as-is.","headline":"The review half is a clear, if dated, survey, but the headline PSO-vs-GA comparison is undermined by a load-balance violation in the paper's own 1500 MW table.","tokens_in":9694,"tokens_out":2580,"would_cite":false,"duration_ms":25038,"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":"PSO beats GA on green economic dispatch in about a third of the time.","keywords":["economic load dispatch","combined economic emission dispatch","particle swarm optimization","genetic algorithm","IEEE 30-bus system","penalty factor","emission dispatch","MATLAB simulation"],"falsifier":"Re-run the same six-generator IEEE 30-bus CEED case at 1500 MW with a genetic algorithm that uses a larger population (for example, 100 individuals) or a convergence-based stopping rule instead of a fixed 500 iterations; if GA then matches or beats PSO's $33,948.83/h total cost in comparable time, the reported speed and cost ranking would not be a stable property of the algorithms.","tokens_in":8629,"feed_emoji":"⚡","tokens_out":7240,"duration_ms":64510,"temperature":0.7,"pith_summary":"The paper argues that combined economic emission dispatch (CEED)—deciding how much each thermal generator should produce to minimize both fuel cost and pollutant emissions while meeting demand—can be solved well by population-based AI methods, and that particle swarm optimization (PSO) outperforms a genetic algorithm (GA) on a standard test case. On the lossless IEEE 30-bus six-generator system at 1500 MW and 2000 MW, PSO reports lower total cost (fuel plus emission-weighted cost) than GA and runs in roughly a quarter to a third of the wall-clock time. The author credits PSO's simple two-equation velocity-and-position update for faster convergence to the claimed global optimum, compared with GA's selection, crossover, and mutation layers. The paper also surveys ten AI techniques for CEED as context for the implementation.","feed_headline":"PSO beats GA on green dispatch in one-third the time","feed_subtitle":"In tests, PSO finds a cheaper combined fuel-and-emission dispatch than GA in about a quarter of the time.","key_machinery":"The load-bearing machinery is the price-penalty-factor transformation that collapses the two conflicting objectives—fuel cost and pollutant emission—into one scalar objective, together with the two competing search mechanisms. The penalty factor $h$ is computed per gas (for example, $h_{\\mathrm{NOX}}=3.1669$, $h_{\\mathrm{COX}}=0.1221$, and $h_{\\mathrm{SOX}}=0.9182$ at 1500 MW) so that the combined cost is $k_1 F_T + k_2 h E_T$; both algorithms then minimize that single number. PSO carries the search with velocity and position update equations using a constriction factor (CF $=0.7298$, $c_1=c_2=2.05$), while GA carries it with binary-string selection, crossover, and mutation. The paper credits the simplicity of PSO's two equations for its faster convergence.","core_discovery":"The paper's central discovery, on its own terms, is that on a lossless IEEE 30-bus system with six thermal generators, particle swarm optimization solves the combined economic emission dispatch problem better than a genetic algorithm. With all three pollutant classes (NOX, COX, SOX) folded into the fuel cost through price penalty factors, PSO reports a total cost of $33,948.83/h at 1500 MW and $56,988.30/h at 2000 MW, against GA's $34,005.52/h and $57,097.16/h, and it does so in 0.35 s and 0.36 s average operating time versus GA's 1.30 s and 1.54 s. The author attributes the gap to PSO's simpler two-equation velocity-position update compared with GA's three-layer selection, crossover, and mutation machinery, which lets PSO converge to the claimed global optimum in fewer iterations.","pith_inferences":["Beyond the paper: the absence of a convergence criterion means the measured speed gap partly reflects the fixed 500-iteration budget; re-running with function-evaluation or convergence-based stopping would separate algorithmic speed from iteration-count artifacts.","Beyond the paper: the penalty-factor values depend on the load demand and the gas-specific formula, so the 'green' weighting is not a fixed constant; different penalty schedules could dispatch different generators and change the emission profile.","Beyond the paper: the lossless six-generator test is a smooth, small problem; on larger systems with line losses and valve-point effects, GA's population diversity may outperform PSO's faster local convergence, so the ranking should not be assumed to scale."],"forward_implications":["The fixed penalty factors computed for the three gases define a reproducible single-objective dispatch problem that any optimizer can be benchmarked on.","For small benchmark systems, PSO's lower total cost and much shorter running time make it a practical first-choice optimizer for CEED when dispatch solutions must be recomputed as load changes.","GA still returns close total costs (about $57/h$ and $109/h$ more at the two loads), so both algorithms are near each other in solution quality; the clear difference is computational time.","Because both methods are parents of the other surveyed techniques, the paper's framing implies that hybrid algorithms borrowing PSO's velocity update and GA's recombination are a natural next step for better CEED performance."],"supporting_citations":[{"why":"Supplies the quadratic fuel-cost and emission formulations that define the two dispatch objectives.","marker":"[1]"},{"why":"Provides the price-penalty-factor procedure used to turn the multi-objective problem into a single objective.","marker":"[2]"},{"why":"Source of the IEEE 30-bus six-generator cost and emission coefficient data and the PSO constraint-handling approach being implemented.","marker":"[3]"},{"why":"Supplies the constriction-factor variant of PSO that the paper uses to update particle velocities.","marker":"[4]"},{"why":"Supplies the genetic-algorithm background (selection, crossover, mutation) that the GA implementation follows.","marker":"[5]"},{"why":"Reports that GA gives quality CEED solutions but with slower computation, the comparison point for the paper's PSO-versus-GA speed result.","marker":"[14]"}],"fun_headline_variants":["PSO beats GA on green dispatch, 4x faster","PSO finds cheaper green dispatch in a quarter of GA's time","Green dispatch: PSO trumps GA on cost and speed","PSO outperforms GA for combined economic emission dispatch"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The ranking depends on treating 10 particles versus 10 individuals with the listed parameter settings over 500 iterations as a fair and representative tuning of both algorithms, and on taking the best-of-50-trials results as typical performance.","fun_headline_variants_meta":{"raw":{"variants":["PSO beats GA on green dispatch, 4x faster","PSO finds cheaper green dispatch in a quarter of GA's time","Green dispatch: PSO trumps GA on cost and speed","PSO outperforms GA for combined economic emission dispatch"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000701,"raw_usage":{"total_tokens":3160,"prompt_tokens":933,"completion_tokens":2227,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":549,"completion_tokens_details":{"reasoning_tokens":2157}},"tokens_in":549,"tokens_out":2227,"duration_ms":17919,"temperature":1.0,"reasoning_tokens":2157,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:00:08.332864+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the same six-generator IEEE 30-bus CEED case at 1500 MW with a genetic algorithm that uses a larger population (for example, 100 individuals) or a convergence-based stopping rule instead of a fixed 500 iterations; if GA then matches or beats PSO's $33,948.83/h total cost in comparable time, the reported speed and cost ranking would not be a stable property of the algorithms.","supporting_citations":[{"cited_title":"The total generation should meet the load demand of consumers and also the line losses if they are considered","cited_arxiv_id":null,"evidence_quote":"Supplies the quadratic fuel-cost and emission formulations that define the two dispatch objectives."},{"cited_title":"Both fuel cost and pollutants emission can be equated as quadratic functions of generator real power","cited_arxiv_id":null,"evidence_quote":"Provides the price-penalty-factor procedure used to turn the multi-objective problem into a single objective."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the IEEE 30-bus six-generator cost and emission coefficient data and the PSO constraint-handling approach being implemented."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the constriction-factor variant of PSO that the paper uses to update particle velocities."},{"cited_title":"Particle Swarm Optimization In 1995 , particle swarm optimization was invented by two scientists Kennedy and Eberhart","cited_arxiv_id":null,"evidence_quote":"Supplies the genetic-algorithm background (selection, crossover, mutation) that the GA implementation follows."},{"cited_title":"Emission - economic dispatch using a nove l constraint handling particle swarm optimization strategy,","cited_arxiv_id":null,"evidence_quote":"Reports that GA gives quality CEED solutions but with slower computation, the comparison point for the paper's PSO-versus-GA speed result."}],"review_version":1}