{"id":"a3029136-bc4c-4e22-a67b-f989c80de700","arxiv_id":"2509.08201","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"An LQR-tuned MIMO PI current controller for inverters improves dynamic response and weak-grid synchronization stability compared with a conventional SISO PI controller in simulation.","lead":"This paper designs a multivariable current controller for grid-connected inverters using optimal control theory, and shows in simulation that it responds faster and stays synchronized in weak grids than a conventional PI controller. A generalist reader might care because the controller keeps the familiar PI structure, so existing inverters could adopt it with near-software-only changes.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"MIMO performance claim may reflect loop-bandwidth mismatch, not multivariable structure; baseline SISO gains are ~2x slower.","rationale":"The central claim is comparative: MIMO outperforms conventional SISO. The evidence for this is a set of time-domain simulations and eigenvalue plots. The most load-bearing weakness is that the SISO baseline is not shown to be fairly tuned. A simple loop-gain estimate from the published gains indicates the MIMO loop is roughly twice as fast, so the observed improvements could be a bandwidth effect rather than a structural benefit of the multivariable optimal PI. This is directly testable by retuning the SISO to the same crossover/phase margin. The reader's stated weakest assumption was robustness to unmodeled dynamics, which is also a real issue, but the fairness of the baseline is more immediately falsifiable from the numbers given. Because the missing test does not disprove the claim, the conditional verdict remains appropriate; no adjustment is needed, but the authors should supply this controlled comparison before the claim is accepted.","tokens_in":12488,"tokens_out":11318,"duration_ms":138598,"concrete_test":"Re-run the SCR=1/X/R=1 power-limit scenario with a conventional SISO PI retuned so that its d-axis current loop has the same crossover frequency and phase margin as the MIMO controller (e.g., kp≈0.269, ki≈7.0, keeping ideal decoupling and vo feedforward, with the same PLL gains and control delay). Record the maximum P/Q before loss of synchronism. If it rises to ~1.6/0.66 p.u., the MIMO advantage is not structural; if it remains near 0.94/0.49 p.u., the multivariable cross-coupling is doing real work.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The quantitative central claim (1.66 vs 0.94 p.u. at SCR=1, X/R=1) is presented as evidence of a multivariable-structure advantage, but the comparison baseline is not controlled for loop speed. With Lf=600 μH and Rf=20 mΩ, the listed SISO gains kp=0.13, ki=11.25 give a d-axis current-loop crossover of roughly 216 rad/s (zero at ~86 rad/s), while the MIMO diagonal gains KP=0.269, KI=7.0 give roughly 448 rad/s (zero at ~26 rad/s). The MIMO loop is therefore about twice as fast before considering cross-coupling. Faster bandwidth alone can raise the achievable power-transfer/synchronization limit of a grid-following converter, so the comparison does not isolate the multivariable structure. The paper does not show that the SISO tuning is a standard, well-tuned, or fair baseline; calling it 'conventional' is not sufficient. The reported '36%' and '24%' improvements are also inconsistent with the quoted numbers (0.72/0.94 ≈ 77%, 0.17/0.49 ≈ 35%), suggesting the quantitative comparison was not carefully normalized. A fair test must hold loop speed, or closed-loop bandwidth, equal between the two structures before attributing the gains to the MIMO design.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript proposes a multivariable PI current controller for grid-connected VSCs. The controller is obtained by applying LQR synthesis to the dq-frame RL filter model augmented with integral states. The resulting control law has a two-by-two PI structure with cross-coupling terms, so it resembles conventional SISO-PI vector control. The authors compare dynamic performance and grid-synchronization stability against a conventional SISO-PI controller using Simulink time-domain simulations and eigenvalue analysis, reporting faster tracking, smaller overshoot, better axis decoupling, and higher achievable power transfer in weak grids (e.g., 1.66 p.u. vs 0.94 p.u. active power at SCR=1 with X/R=1).","tokens_in":12828,"tokens_out":5747,"duration_ms":68851,"significance":"The LQR formulation is standard and correctly executed, and the controller structure is attractive because it preserves the conventional PI architecture while adding only two off-diagonal terms. The paper also includes a control delay in the simulation model and robustness checks against filter parameter variation. If the performance advantage survives a fair comparison, the proposed tuning method could be a useful systematic alternative to existing current-control tuning rules. However, the current evidence does not isolate the contribution of the multivariable structure because the SISO baseline is not bandwidth-matched, and some numerical claims are internally inconsistent.","major_comments":[{"comment":"The comparison baseline is not controlled for loop bandwidth. With Lf=600 μH and Rf=20 mΩ, the conventional PI gains kp=0.13, ki=11.25 give a d-axis current-loop crossover of roughly 216 rad/s (zero at ki/kp≈86.5 rad/s), whereas the MIMO diagonal gains KP=0.269, KI,11=7.0076 give a crossover of roughly 448 rad/s (zero at ≈26 rad/s). The MIMO loop is therefore about twice as fast before considering the off-diagonal terms. Because faster current-loop bandwidth is itself known to improve weak-grid synchronization of grid-following converters, the reported improvements in Figs. 5-9 and the stability margins in Figs. 10-12 may reflect loop-speed difference rather than the multivariable structure. The central claim of the paper requires a matched-bandwidth comparison (e.g., tune the SISO PI to the same crossover frequency as the MIMO design, or sweep SISO bandwidth).","section":"Section IV.C, Table I"},{"comment":"The reported percentage improvements are inconsistent with the stated numerical limits. The active-power improvement is 1.66 p.u. vs 0.94 p.u., i.e., an increase of 0.72/0.94≈76.6%, not 36%. The reactive-power improvement is 0.66 p.u. vs 0.49 p.u., i.e., 0.17/0.49≈34.7%, not 24%. Please correct these values and state explicitly the base used for the percentage calculation; as written, the quantitative comparison appears not to have been carefully normalized.","section":"Section IV.C"},{"comment":"The eigenvalue comparison suffers from the same confound as the time-domain comparison. The SISO and MIMO systems have different current-loop bandwidths, so the observation that the MIMO eigenvalues remain in the left half-plane for lower SCR values does not by itself demonstrate a structural advantage of the multivariable controller. A matched-bandwidth eigenvalue study, or a sensitivity analysis over SISO PI gains, is needed before claiming enhanced synchronization stability.","section":"Section IV.E"},{"comment":"The methodology for determining the maximum injectable power before loss of synchronization is not described. The reader is not told what disturbance is applied, how the power setpoint is ramped or stepped, or what criterion is used to declare loss of synchronization. Without this information, the reported limits (0.94 p.u. and 1.66 p.u. active power, 0.49 p.u. and 0.66 p.u. reactive) are not reproducible, and the quantitative claim is difficult to verify.","section":"Section IV.C"}],"minor_comments":[{"comment":"In the text after Eq. (1), 'vid and vid' should presumably be 'vid and viq'.; likewise check 'viq' usage in the controller schematic in Fig. 4, where 'k24+k24/s' appears to be a typo for 'k24+k24/s' or the intended k22+k24/s.","section":"Section I, Nomenclature"},{"comment":"The control delay entry is garbled: 'td Control delay 32fsw' should read '3/(2 fsw)' as stated in the text. Please correct the table formatting.","section":"Table I"},{"comment":"The conclusion states that 'the LQR naturally selects only the integral action,' but the controller in Table I has a nonzero proportional gain matrix KP=diag(0.269,0.269). This statement is contradicted by the reported parameters and is not demonstrated in the body; please revise or clarify.","section":"Conclusion"},{"comment":"The eigenvalue plots have no labeled axes. Please label the real and imaginary axes and indicate the stability boundary so the reader can interpret the eigenvalue trajectories.","section":"Figures 11 and 12"},{"comment":"The Introduction contains duplicated sentences (e.g., the text starting 'where the performance is enhanced by incorporating the PLL dynamics...' appears twice). This appears to be a text-merge artifact; it should be cleaned up.","section":"Introduction"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern is valid and should be addressed head-on. The paper's core idea is reasonable, but the central performance claim is currently supported by an unfair bandwidth comparison and incorrect percentages. I would ask the authors to re-run the comparison with matched current-loop bandwidth, correct the arithmetic, and provide the detailed simulation protocol for the maximum-power limits. If they do so, the paper may become suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The math is fine and the controller structure is genuinely practical, but the evidence for the central claim does not hold up. The paper builds a MIMO PI current controller by applying LQR on an augmented integral state, and the resulting structure closely matches the standard VSC PI control loop. That is a real convenience: manufacturers could adopt it with minimal software changes, and the tuning is systematic. The derivation from the RL filter model to the PI gains is correct and easy to follow. The step-response simulations also show clean transient improvements, and the eigenvalue plots suggest better stability margins in weak grids, though the full state-space model is only cited, not given.\n\nThe soft spot is the comparison. With Lf=600 µH and Rf=20 mΩ, the listed SISO gains give a d-axis current-loop crossover around 216 rad/s, while the diagonal MIMO gains give roughly 448 rad/s—about twice the bandwidth. A faster loop alone can raise synchronization limits, so the claimed advantage is not isolated to the multivariable structure. The paper does not show its SISO tuning is standard or well-tuned; calling it conventional is not enough. Also, the arithmetic is off: 1.66 vs 0.94 p.u. is a 77% increase, not 36%, and 0.66 vs 0.49 is 35%, not 24%. Those inconsistencies suggest the numbers were not checked carefully. The analytical eigenvalue analysis is described as derived according to [29], but no matrices or details are provided, so a referee cannot verify it. One more mismatch: the conclusion says LQR naturally selects only the integral action, but Table I shows nonzero proportional gains (KP diagonal 0.269), so that statement is wrong as written.\n\nNone of this kills the idea. The LQR-based MIMO PI controller is a legitimate contribution, and the design’s robustness to neglected PLL/delay dynamics is plausible even if the LQR margins only apply to the nominal plant. But the paper overclaims by not controlling for loop speed and by misreporting its own numbers. A fair test with a matched-bandwidth SISO baseline, corrected percentages, and a fully specified eigen-analysis would be necessary before I believe the synchronization-stability advantage.\n\nThis paper deserves a serious referee: the topic is relevant, the derivation is sound, and the practical potential is real. I would send it to review with the expectation of major revision—mainly to redo the baseline comparison and fix the numerical errors.","headline":"A clean LQR-based MIMO PI derivation with a practical structure, but the headline performance gains rest on an unfair SISO baseline and miscomputed percentages.","tokens_in":13286,"tokens_out":3389,"would_cite":false,"duration_ms":41494,"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":"A MIMO current controller tuned by optimal control keeps weak-grid inverters synchronized at 1.66 p.u. of power, where the standard PI loop fails at 0.94 p.u.","keywords":["current controller","MIMO-PI","optimal control theory","IBR","weak grid","grid synchronization stability","LQR","vector current control"],"falsifier":"Repeat the Section IV-D line-outage scenario (SCR 4 → 2, 0.66 p.u. active and reactive injection) in a detailed electromagnetic-transient simulation that includes the phase-locked loop, measurement filters, and the 0.3 ms PWM delay; if the MIMO-controlled inverter loses synchronism, or if a sweep of the active-power setpoint at SCR = 1 and X/R = 1 loses sync below 1.66 p.u., the central claim fails. A complementary check: compute the small-signal eigenvalues of the full system (PLL plus grid impedance included) with the paper's fixed gains; any right-half-plane crossing for SCR ≥ 1 at the clai","tokens_in":12419,"feed_emoji":"⚡","tokens_out":24822,"duration_ms":210721,"temperature":0.7,"pith_summary":"This paper argues that the standard way to control the current of an inverter-based resource — decoupling the d and q axes with measured feedforward terms, then running two independent PI controllers — is what makes grid-following inverters oscillate and lose synchronism in weak grids. The authors replace that scheme with a two-by-two PI controller whose gains are computed in one shot by solving a linear-quadratic (LQR) optimal control problem on the filter model, with the integral of the tracking error added as a state. Because the resulting controller has the same input-output structure as the conventional one, the claim is that it can be adopted with only minor changes to existing vector-control schemes. The payoff, if true, is operational: time-domain simulations and eigenvalue analysis show faster settling, better axis decoupling, and, at SCR = 1 with X/R = 1, a synchronization limit of 1.66 p.u. active power instead of 0.94 p.u. A careful reader would care because weak-grid synchronization is a live problem for solar and battery plants, and this is a drop-in fix rather than a new control architecture.","feed_headline":"Lifts weak-grid inverter limit to 1.66 p.u. with MIMO current loop","feed_subtitle":"The multivariable PI, tuned by LQR, holds sync where the standard dq-axis PI stalls: 1.66 vs 0.94 p.u. active power.","key_machinery":"The central machinery is an augmented state-space optimal-control design. The RL filter model (states iid, iiq; inputs vid, viq) is extended by the integral of the tracking error, z(t) = ∫(x − x*)dt, and the control input is the deviation eu = u − u* from the feedforward u* = −B−1Ax*. Solving an LQR cost on this augmented system gives a 2×4 gain K that splits into proportional (KP) and integral (KI) 2×2 blocks, yielding u = KP(x* − x) + KI∫(x* − x)dτ + u*. With the paper's penalties, KP is diagonal and KI has equal diagonal entries with opposite-sign off-diagonal entries, so the rotating-field, cross-axis action that conventional controllers try to create with measured feedforward terms come","core_discovery":"Central claim: a MIMO current controller tuned as an optimal PI regulator beats the usual SISO-PI controller in speed and grid-synchronization stability. The design augments the filter model with the error integral, solves the Riccati equation, and splits the gain into proportional and integral blocks. The proportional block is diagonal and the integral block is cross-coupled, so integral action supplies the decoupling that feedforward terms cannot once PLL delay and weak-grid impedance act. Simulations, eigenvalue sweeps, and an SCR 4-to-2 outage confirm the advantage. At SCR = 1, X/R = 1, the claimed limit is 1.66 p.u. active and 0.66 p.u. reactive versus 0.94 and 0.49 p.u.","pith_inferences":["The antisymmetric off-diagonal entries of the integral gain matrix (KI12 = −4.571, KI21 = +4.571 with equal diagonals 7.008) are structurally a complex-vector, rotating-field integrator; an obvious test is whether the LQR penalty choices can be mapped onto the complex-vector gains of the earlier approaches the paper lists, which would unify the two design traditions.","Because the design model ignores PLL and grid impedance, the claimed robustness should degrade gracefully rather than suddenly; a concrete extension is to re-solve the LQR on an augmented model that includes the PLL angle as a state and check whether the diagonal-proportional/cross-coupled-integral structure survives, which would let the design extend to plants with explicitly modeled synchronizat","The quantitative results are for a 100 kW, 500 V system; before utility-scale deployment, the 1.66 p.u. limit should be re-checked with realistic measurement-filter and PWM delays, since those are exactly the unmodeled dynamics the gain robustness must absorb.","The construction is generic for any two-axis RL-type plant, so the same augmented-LQR procedure could be applied to other coupled dq loops such as voltage or power-synchronization control, where cross-axis coupling is also the limiting factor."],"forward_implications":["A line outage that drops grid strength from SCR = 4 to SCR = 2 no longer costs synchronization: with the MIMO controller the inverter keeps running at its pre-fault setpoints, where the conventional SISO controller pulls out of synchronism.","Active-power capability before loss of synchronism at SCR = 1 and X/R = 1 rises from 0.94 p.u. to 1.66 p.u.; reactive capability rises from 0.49 p.u. to 0.66 p.u., with the same controller gains across all tested operating points, requiring no re-tuning as grid strength changes.","Because the proportional block stays diagonal and only the integral block carries cross-coupling, the controller keeps the conventional PI structure and needs no state observers or grid Thevenin-impedance estimates, so retrofitting existing vector-control schemes is a small change.","Eigenvalue sweeps indicate the stability margin degrades more slowly under filter-parameter variation (up to 2× resistance, down to 0.5× inductance) than for the conventional controller, supporting the robustness story.","LQR synthesis gives the controller the classical gain and phase margins of optimal control, so tuning is a systematic penalty selection rather than heuristic gain iteration."],"supporting_citations":[{"why":"Supplies the optimal proportional-plus-integral regulator method the controller design is built on.","marker":"[21]"},{"why":"Provides the MIMO PI formulation with integral state on which the augmented LQR construction rests.","marker":"[22]"},{"why":"Standard linear optimal control reference used for the error-system formulation and Riccati solution.","marker":"[23]"},{"why":"Justifies the one-and-a-half-sample control delay included in the simulated system.","marker":"[24]"},{"why":"Establishes that PLL dynamics plus weak-grid impedance couple the dq current loops, the failure mode the MIMO controller addresses.","marker":"[25]"},{"why":"Supplies the static power transfer limit equations and SCR definition used to frame the power-capability comparison.","marker":"[26]"},{"why":"The Arizona solar-plant weak-grid incident that motivates the line-outage synchronization scenario.","marker":"[28]"},{"why":"Provides the dq state-space model used for the eigenvalue-based synchronization stability analysis.","marker":"[29]"},{"why":"Prior multivariable-PI dq current control with axis decoupling that this LQR-based design extends and simplifies.","marker":"[12]"},{"why":"The bandwidth-reduction tuning approach for extremely weak grids that the paper claims its controller makes unnecessary.","marker":"[18]"}],"fun_headline_variants":["MIMO current loop: 1.66 p.u. weak-grid sync hold","LQR-designed PI for IBRs: 1.66 p.u. in weakest grid","MIMO PI extends inverter sync limit to 1.66 p.u.","Weak-grid stability at 1.66 p.u. via MIMO current control","Optimal MIMO current loop beats SISO at 1.66 p.u. weak-grid"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The entire advantage rests on LQR gains computed from the bare RL filter model staying effective once the phase-locked loop, grid impedance, and control delay are added; if that robustness assumption fails, the synchronization improvements do not carry over.","fun_headline_variants_meta":{"raw":{"variants":["MIMO current loop: 1.66 p.u. weak-grid sync hold","LQR-designed PI for IBRs: 1.66 p.u. in weakest grid","MIMO PI extends inverter sync limit to 1.66 p.u.","Weak-grid stability at 1.66 p.u. via MIMO current control","Optimal MIMO current loop beats SISO at 1.66 p.u. weak-grid"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000822,"raw_usage":{"total_tokens":3393,"prompt_tokens":667,"completion_tokens":2726,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":411,"completion_tokens_details":{"reasoning_tokens":2616}},"tokens_in":411,"tokens_out":2726,"duration_ms":24475,"temperature":1.0,"reasoning_tokens":2616,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T21:03:10.001673+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the Section IV-D line-outage scenario (SCR 4 → 2, 0.66 p.u. active and reactive injection) in a detailed electromagnetic-transient simulation that includes the phase-locked loop, measurement filters, and the 0.3 ms PWM delay; if the MIMO-controlled inverter loses synchronism, or if a sweep of the active-power setpoint at SCR = 1 and X/R = 1 loses sync below 1.66 p.u., the central claim fails. A complementary check: compute the small-signal eigenvalues of the full system (PLL plus grid impedance included) with the paper's fixed gains; any right-half-plane crossing for SCR ≥ 1 at the clai","supporting_citations":[],"review_version":1}