{"id":"625f0b85-6617-409a-a19b-836e147e9bc1","arxiv_id":"2501.11633","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"PSO-optimized sliding mode current control reduces tracking error by about 11.6% and converges about 86% faster than GA on the tested grid-forming inverter scenarios.","lead":"This paper tunes the parameters of a sliding mode current controller for a grid-forming inverter using particle swarm optimization (PSO). It reports lower tracking error and faster optimization convergence compared with genetic algorithm, simulated annealing, and the expert-tuned baseline.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 11.61% tracking-error improvement is computed on the same simulation scenarios used to train the PSO; without an out-of-sample test, the central claim may be circular rather than a real performance gain.","rationale":"The reader's weakest assumption correctly identifies the lack of out-of-sample testing. My stress-test sharpens this into a concrete circularity: the PSO cost function is the IAE over the very same simulation scenarios that are later used for the reported comparison. This makes the 11.61% improvement an expected consequence of fitting the test set, not evidence of a better controller. The absence of standard deviations and the arbitrary convergence threshold further weaken the quantitative claims. The paper is not internally inconsistent or nonsensical; the control design and optimization procedure are plausible. The concern is evidentiary: the current manuscript cannot distinguish overfitting from genuine improvement. A hold-out experiment, as specified, would settle the issue. Since the reader already assigned a conditional verdict based on this concern, no adjustment is needed; the verdict remains CONDITIONAL pending additional validation.","tokens_in":6962,"tokens_out":2827,"duration_ms":30544,"concrete_test":"Hold out scenario 3: run PSO using only the first two scenarios (linear load change and nonlinear load connection) as the training cost, then evaluate the optimized ksat, kcd, kcq on scenario 3 (40% parameter uncertainty) and on a previously unseen load profile (e.g., a 50% load step or different R_L and L_L values). Compare the out-of-sample IAE of PSO-DAM-SMC with the heuristic DAM-SMC. If the gain is substantially smaller than 11.61% or reverses, the in-sample claim is not a general performance result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on comparing the IAE of the PSO-tuned DAM-SMC against the heuristic DAM-SMC of [7] using the same three-scenario 0.7 s simulation. Algorithm 1 minimizes IAE exactly over that simulation (including the 0.1 s load change, 0.2 s nonlinear load connection, and 0.5 s +40% parameter variation), and Table II reports IAE for the same run. Thus the optimization has access to the test scenarios, so the 11.61% improvement is partly a measure of how well PSO fits the chosen scenarios, not a validated advantage under new operating conditions. The paper also reports only the mean IAE of 10 runs, with no variance or significance test, and the convergence comparison uses an arbitrary threshold (IAE = 0.037). Together these features leave the headline improvement unverifiable as a general property.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Particle Swarm Optimization (PSO)-based tuning method for the three control parameters (ksat, kcd, kcq) of a Decoupled Average Model-based Sliding Mode Current Controller (DAM-SMC) for a grid-forming inverter with an LC filter. The control laws are taken from prior work [7], and the parameters are optimized offline by minimizing the integral absolute error (IAE) of the dq-axis inductor currents over a 0.7 s simulation that includes a linear load change, connection of a nonlinear load, and a 40% parameter variation. The optimized controller is compared with the heuristic DAM-SMC of [7], a GA-based DAM-SMC, and an SA-based DAM-SMC. The paper reports an 11.61% IAE reduction compared to the heuristic controller and a much faster convergence of the PSO (3 iterations) versus GA (22) and SA (27).","tokens_in":7211,"tokens_out":4237,"duration_ms":43420,"significance":"If the reported improvements are generalizable, the paper offers a simple, practical way to replace expert-knowledge tuning of an existing sliding-mode current controller with a standard metaheuristic, while preserving fixed switching frequency and low computational cost. The manuscript is clearly organized, provides a complete plant and controller model, and makes a fair algorithmic comparison in the sense that all three optimizers minimize the same IAE over the same simulation scenarios with the same population size and number of iterations. The inclusion of a 40% parameter-uncertainty scenario is a useful stress test. However, the central claim is validated only on the same scenarios used for optimization, and the reported statistics are too thin to establish that the PSO result is reliably better than GA or SA.","major_comments":[{"comment":"The IAE defined in Eq. (9) over the 0.7 s simulation with the three fixed scenarios is both the objective minimized by PSO and the performance metric reported in Table II. Therefore the reported 11.61% improvement over the heuristic DAM-SMC of [7] measures in-sample fit to those specific scenarios, not a validated advantage under new operating conditions. I recommend an out-of-sample evaluation, for example by re-tuning on a subset of scenarios and testing on the remaining scenario, or by applying the optimized parameters to a different load profile and parameter variation.","section":"Section III (Algorithm 1, Eq. (9)) and Section IV-B (Table II)"},{"comment":"The text states that each algorithm is run 10 times and the mean IAE is obtained, but Table II reports only means. No standard deviation, confidence interval, or significance test is provided. The differences among PSO (0.08108), GA (0.08324), and SA (0.08320) are small relative to typical run-to-run variability of stochastic optimizers; without dispersion measures or a statistical test, the ranking of the algorithms is not supported.","section":"Section IV-A and Table II"},{"comment":"The convergence-speed comparison uses an arbitrary threshold IAE = 0.037, with no justification for this value. The number of iterations needed to reach a threshold depends on the threshold choice and on the scale of the IAE; a more principled comparison would report full convergence curves or define a stopping rule based on stagnation of the best fitness.","section":"Section IV-B and Figure 5"},{"comment":"The optimized parameter values kopt_sat, kopt_cd, and kopt_cq are never reported. Without these values the experiment is not reproducible, and it is impossible to verify that the optimizer actually moved the parameters away from the expert-knowledge values used in [7] in a meaningful way. The manuscript should list the optimized parameters for all three algorithms.","section":"Section III (Algorithm 1) and Table I"}],"minor_comments":[{"comment":"The abstract reports an 11.61% improvement, while the introduction states approximately 11.67%; these numbers should be reconciled.","section":"Abstract and Section I"},{"comment":"Scenario-specific improvements (31.27% at the linear load change, 14.375% at the nonlinear load connection) are presented alongside the overall IAE improvement of 11.61%; the paper should clarify that the scenario-specific numbers are local IAE reductions, not the overall metric in Table II, and should define how each is computed.","section":"Section IV-B"},{"comment":"The labels for the optimization bounds appear reversed: the row labeled \"Upper limit\" lists [1, 1, 0.001] and the row labeled \"Lower limit\" lists [2000, 2000, 15], which contradicts the text in Section III that the boundaries are [1,2000] for kcd and kcq and [0.001,15] for ksat. Please correct the labels or the values.","section":"Table I"},{"comment":"The figure includes a text label \"IAE = 0.0016\" that is not defined or referenced in the body of the paper; please add an explanation or remove the label.","section":"Figure 3"},{"comment":"The PSO hyperparameters (inertia range, c1, c2) and the GA/SA settings are listed without justification or sensitivity analysis; a brief explanation or a reference for these choices would strengthen the comparison.","section":"Section III and Table I"}],"recommendation":"major_revision","confidential_remarks":"The paper applies PSO to tune three parameters of an existing DAM-SMC; this is incremental but potentially useful. The main issue is the in-sample validation: the IAE used as the optimization objective is exactly the IAE reported in Table II, and the convergence comparison relies on an arbitrary threshold. These concerns are fixable within the scope of the manuscript by adding an out-of-sample test, reporting variance across the 10 runs, and providing the optimized parameter values. I therefore recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Routine tuning paper, and the abstract nearly says so. The new bit is taking the Decoupled Average Model-based Sliding Mode Current Controller (DAM-SMC) from [7], keeping its control law, and optimizing its three parameters (ksat, kcd, kcq) offline with PSO. That specific application — PSO-tuned DAM-SMC for a grid-forming inverter in rotating frame, benchmarked against GA and SA — is not in the cited literature, so the novelty is narrow but real. There is no new control law, no new model, and no new stability result.\n\nWhat it does well: the system setup is clearly described, the three test scenarios (load change at 0.1 s, nonlinear load at 0.2 s, 40% parameter mismatch at 0.5 s) are reasonable stress tests, and benchmarking against two other metaheuristics is a sensible thing to do. The 3-iteration convergence of PSO versus 22-27 for GA/SA is a concrete practical observation, though the convergence comparison uses an arbitrarily chosen IAE threshold of 0.037.\n\nThe soft spot is the one the stress-test note flags, and it lands. The PSO minimizes IAE over the exact 0.7 s simulation that Table II uses to compute the 11.61% improvement. That makes the improvement a measure of how well PSO fits those scenarios, not evidence of a general gain. The robustness scenario (the 40% parameter variation) is also inside the same simulation the optimizer sees, so it is not an out-of-sample test. The authors report no variance or confidence intervals for the 10 runs, and they never give the optimized parameter values — a strange omission for a tuning study. The improvement numbers are also inconsistent across the paper (11.61% in the abstract, 11.67% in the introduction, 31.27% and 14.375% in scenario-specific discussions). These issues are fixable, not disqualifying.\n\nThe central argument — that PSO can tune this controller at least as well as GA or SA — holds up for the scenarios tested. What does not hold up is the stronger claim that the improvement will carry over to other operating conditions. There is no experiment beyond the in-sample simulation.\n\nThis paper is for practitioners in inverter control who want a quick way to tune an SMC. It is not for someone seeking a new control concept. I would not cite it in my own work, but it is coherent and honest in its method. I would send it to review rather than desk-reject: the topic is relevant and the issues are addressable, but a reviewer should demand the missing parameter values, a validation on a scenario not used for fitting, and run-to-run statistics.","headline":"A plausible but in-sample PSO-tuning study whose headline claim is measured on the same scenarios used to fit the parameters.","tokens_in":7661,"tokens_out":3488,"would_cite":false,"duration_ms":34244,"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":"Offline particle-swarm tuning of sliding-mode current control cuts tracking error by 11.61% and converges in three iterations.","keywords":["grid-forming inverter","sliding mode current control","particle swarm optimization","decoupled average model","integral absolute error","chattering reduction","inverter-based resources","rotating dq frame"],"falsifier":"Run the PSO-optimized controller on a scenario not used in tuning, such as a different load step size, a new nonlinear load impedance, or a d-q reference step, and compare its IAE with the conventional DAM-SMC under identical conditions. If the improvement disappears or reverses, the reported 11.61% gain is an artifact of the tuning scenarios rather than a general property of the controller.","tokens_in":6808,"feed_emoji":"⚡","tokens_out":5991,"duration_ms":60118,"temperature":0.7,"pith_summary":"Grid-forming inverters need current controllers that respond quickly, hold a fixed switching frequency, and avoid overshoot; the decoupled average-model sliding mode current controller (DAM-SMC) meets these needs, but its three tuning parameters are usually chosen by expert judgment. This paper claims those parameters can instead be tuned offline with particle swarm optimization (PSO), using the integral absolute tracking error as the cost. In simulations, the PSO-tuned controller reduces the mean tracking error by 11.61% compared with the expert-tuned DAM-SMC and reaches the target fitness in 3 iterations, versus 22 for genetic algorithm tuning and 27 for simulated annealing. The tuned controller also keeps the current close to its reference when the plant inductance and resistance are varied by 40%, which the paper reads as robustness to model uncertainty.","feed_headline":"PSO-tuned sliding mode current control cuts error 11.61%","feed_subtitle":"Three parameters tuned in 3 iterations, versus 22 for GA and 27 for SA, with fixed switching frequency.","key_machinery":"The key machinery is the combination of the DAM-SMC law in the rotating dq frame, whose three tunable parameters $k_{\\mathrm{sat}}$, $k_{\\mathrm{cd}}$, and $k_{\\mathrm{cq}}$ shape the saturation boundary and switching gains, with particle swarm optimization as an offline search over those parameters. PSO treats each candidate parameter set as a particle, updates particle velocity from inertia and from attraction to the best personal and global solutions, and evaluates every candidate by running the Simulink model and computing the integral absolute error. The result is a tuned parameter set that the existing controller law can use without structural change.","core_discovery":"On the paper's own terms, the discovery is that the DAM-SMC's control parameters $k_{\\mathrm{sat}}$, $k_{\\mathrm{cd}}$, and $k_{\\mathrm{cq}}$ form a low-dimensional search space that PSO can optimize more efficiently than genetic algorithms or simulated annealing, and the resulting controller beats the expert-chosen baseline. The paper reports a mean IAE of 0.08108 with PSO, an 11.61% improvement over the conventional DAM-SMC's 0.09173, and convergence to IAE = 0.037 in 3 iterations compared with 22 for GA and 27 for SA. These numbers come from ten runs of each optimizer on a MATLAB/Simulink model covering linear load change, nonlinear load connection, and a 40% increase in inductance and resistance.","pith_inferences":["A direct extension the paper does not test is out-of-sample performance: evaluating the PSO-tuned parameters on load profiles or operating points outside the three training scenarios would show whether the 11.61% gain generalizes or reflects overfitting to the chosen events.","The optimization could be reformulated to include chattering amplitude or switching effort in the cost function alongside IAE, which would make the chattering reduction an explicit objective rather than an indirect outcome.","The same PSO wrapper could be applied to the outer PI voltage loop, since the cascade structure means the inner-loop optimization's benefit depends partly on the fixed outer-loop gains."],"forward_implications":["The optimized parameter set can be embedded into existing DAM-SMC current controllers with no hardware change, turning a manual tuning step into a one-time offline computation.","Because chattering is reduced while fixed switching frequency is retained, the approach may extend the usable life of switching devices and simplify filter design.","PSO's 3-iteration convergence makes offline re-tuning practical whenever the plant parameters drift or the operating envelope changes.","The 40% parameter-mismatch test suggests the optimized controller tolerates aging or estimation errors in the LC filter and load, rather than relying on exact model knowledge."],"supporting_citations":[{"why":"Supplies the DAM-SMC control laws, the plant model, and the expert-tuned baseline that the proposed method inherits and improves upon.","marker":"[7]"},{"why":"Provides the particle swarm optimization update rules used to search the three control parameters.","marker":"[15]"},{"why":"Defines the predictive current control alternative that motivates the need for a current controller not requiring a highly accurate model.","marker":"[6]"},{"why":"Documents the chattering phenomenon in sliding mode control that the optimization aims to reduce.","marker":"[13]"},{"why":"Supports the robustness property of sliding mode control that the paper relies on under parameter variations.","marker":"[12]"}],"fun_headline_variants":["PSO-tuned sliding mode cuts inverter error 11.6%","PSO optimizes sliding mode controller, cuts error 11.6%","PSO tunes sliding mode 86% faster than GA/SA, error down 11.6%","PSO-optimized controls: 11.6% lower error, 86% faster convergence"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes that the average absolute tracking error computed over its three fixed simulation scenarios is a valid measure of control quality, and that parameters optimized against those scenarios will also perform better in normal operation.","fun_headline_variants_meta":{"raw":{"variants":["PSO-tuned sliding mode cuts inverter error 11.6%","PSO optimizes sliding mode controller, cuts error 11.6%","PSO tunes sliding mode 86% faster than GA/SA, error down 11.6%","PSO-optimized controls: 11.6% lower error, 86% faster convergence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000944,"raw_usage":{"total_tokens":4056,"prompt_tokens":994,"completion_tokens":3062,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":610,"completion_tokens_details":{"reasoning_tokens":2969}},"tokens_in":610,"tokens_out":3062,"duration_ms":25910,"temperature":1.0,"reasoning_tokens":2969,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:00:46.945557+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the PSO-optimized controller on a scenario not used in tuning, such as a different load step size, a new nonlinear load impedance, or a d-q reference step, and compare its IAE with the conventional DAM-SMC under identical conditions. If the improvement disappears or reverses, the reported 11.61% gain is an artifact of the tuning scenarios rather than a general property of the controller.","supporting_citations":[{"cited_title":"Decoupled average model-based sliding mode current control of lc-filtered inverters in rotating frame,","cited_arxiv_id":null,"evidence_quote":"Supplies the DAM-SMC control laws, the plant model, and the expert-tuned baseline that the proposed method inherits and improves upon."},{"cited_title":"Particle swarm optimization,","cited_arxiv_id":null,"evidence_quote":"Provides the particle swarm optimization update rules used to search the three control parameters."},{"cited_title":"Predictive current control of a new three-phase voltage source inverter with phase shift compensation,","cited_arxiv_id":null,"evidence_quote":"Defines the predictive current control alternative that motivates the need for a current controller not requiring a highly accurate model."},{"cited_title":"Higher-order sliding modes, differentiation and output- feedback control,","cited_arxiv_id":null,"evidence_quote":"Documents the chattering phenomenon in sliding mode control that the optimization aims to reduce."},{"cited_title":"Improved robustness and perfor- mance of discrete time sliding mode control systems,","cited_arxiv_id":null,"evidence_quote":"Supports the robustness property of sliding mode control that the paper relies on under parameter variations."}],"review_version":1}