{"id":"37095490-6a2a-4682-a324-1466b90a2c5d","arxiv_id":"2506.14059","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":9,"one_line_summary":"A mean-reverting SDE with multiplicative fractional Brownian noise is proposed as a generative model for cycle-level queue length time series at signalized intersections, and it reproduces the observed queue length PDF and 1/f-type PSD.","lead":"This paper proposes a stochastic differential equation to describe how vehicle queues at traffic lights grow and shrink, with the queue pulled toward a daily pattern while random fluctuations with long memory add noise. The authors call it the first equation-based model of queue dynamics and say it reproduces the statistical fingerprints of real queue data, which could help embed traffic physics into machine-learning controllers.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 1/f PSD claim is untestable as reported: the Hurst exponent H that drives the fractional noise in Eq. 7 is never calibrated or reported, and the micro-to-macro bridge in Section IV-B3 is an assertion, so the PDF/PSD match does not validate the stated equation.","rationale":"I focus on the missing Hurst exponent rather than the reader's micro-to-macro objection because the latter affects the derivation narrative while the former blocks direct confirmation of the strongest empirical claim. Even if Eq. (7) is accepted as a proposed statistical model, the absence of H and the simulation scheme makes the 'replicates PSD' statement unfalsifiable. The micro-to-macro assertion in Section IV-B3 compounds this by giving no independent reason for the fBm driver to appear at the aggregate level. The calibration is also weakened by using the same data to construct φ_t and fit parameters before comparing PDFs and PSDs, but the H omission alone is enough to keep the reject verdict. The paper provides no parameter table including H, no explicit integration scheme, and no reproducibility protocol; these are exactly the places where the concern could be settled.","tokens_in":14065,"tokens_out":5546,"duration_ms":60125,"concrete_test":"On held-out data (e.g., the final week of each 90-day series, or a fifth corridor), estimate H from the empirical queue series using DFA or wavelet analysis, report the point estimate and confidence interval, and simulate Eq. (7) with that H using an explicit fBm approximation (e.g., Hosking or circulant embedding). If the simulated PSD no longer shows the reported 1/f slope or the PDF degrades, the central claim is falsified; if the simulation also reproduces the PSD under the correctly estimated H, the claim is supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central replication claim depends on Eq. (7), dY_t = μ(φ_t − Y_t) dt + γ_t Y_t dW_t^H, where the fractional Brownian driver W_t^H has a Hurst exponent H that controls the memory and the spectral slope of the process. H is never estimated, reported, or listed in Table I, and no discrete-time simulation scheme for the fBm integral is given. A reader cannot reproduce the simulated PSD in Fig. 9 or determine whether the 1/f-like match is a property of the model or a free choice of H. The only derivation step linking Eq. (5) to Eq. (7) is the sentence 'in the cumulative process, the noise will still remain multiplicative' (Section IV-B3); no aggregation theorem is supplied, and a sum of many multiplicative random variables need not remain multiplicative. In addition, the volatility process in Eq. (8), labeled a 'square-root diffusion', is actually an additive-noise Ornstein-Uhlenbeck process and does not enforce γ(t) ≥ 0. Thus the reported validation is not a test of the formally stated equation, and the central 'first equation-based model' claim is not supported as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a stochastic differential equation (SDE) model for cycle-to-cycle queue length dynamics at signalized intersections. The model is dY_t = μ(φ_t − Y_t) dt + γ_t Y_t dW_t^H, combining mean reversion toward a periodic seasonal mean φ_t, multiplicative noise with stochastic volatility γ_t, and fractional Brownian motion W_t^H. The authors motivate the model from a microscopic point-process model of vehicle arrivals whose interarrival times follow multiplicative noise, then assert that the cumulative queue process inherits multiplicative noise. They calibrate the model separately for four intersections on the Alafaya Trail corridor using Nelder–Mead optimization of RMSE, extract the seasonal component from the same time series, simulate the SDE, and compare simulated trajectories, Pearson correlations, power spectral densities, and probability density functions with the observed data. The central claims are that the model replicates the empirical PDF and 1/f-type PSD and that it is the first equation-based model for queue dynamics.","tokens_in":14429,"tokens_out":3145,"duration_ms":32615,"significance":"If the proposed SDE were properly derived, calibrated with a reported Hurst exponent, and validated on independent data, it could provide a compact generative description of queue length fluctuations that might be useful for simulation, prediction, and integration with physics-informed learning. The empirical observation of 1/f-like spectra in cycle-level queue lengths at multiple signalized intersections is a valuable data point in itself. However, as presented, the central derivation from Eqs. (2)–(5) to Eq. (7) is an unsupported assertion, the Hurst exponent H is never estimated or reported, the volatility process is mis-specified, and the validation is largely circular because the seasonal mean is extracted from the same series used for fitting and evaluation. These issues prevent the main claim of a physically grounded equation-based model from being substantiated. The paper also does not provide simulation code or a reproducible numerical scheme for the fractional Brownian motion, which is essential for independent verification.","major_comments":[{"comment":"The transition from the microscopic multiplicative-noise model of interarrival times to the macroscopic SDE for queue length is not derived. Eq. (5) is obtained from Eqs. (2)–(4) by Itô's lemma, but the move from Eq. (5) to Eq. (7) rests solely on the sentence \"in the cumulative process, the noise will still remain multiplicative.\" No aggregation theorem, limit argument, or scaling analysis is supplied. The cumulative sum of many multiplicative random variables need not remain multiplicative, and the mean-reversion term μ(φ_t − Y_t) and the fractional Brownian driver are introduced at this step without derivation. Consequently, Eq. (7) is not a derived equation of queue dynamics but an independently chosen phenomenological model, which undermines the paper's central claim of an equation-based model emerging from microscopic traffic behavior.","section":"Section IV-B3, Eqs. (2)–(7)"},{"comment":"The Hurst exponent H in Eq. (7) is never estimated, reported, or listed in Table I, and no discrete-time simulation scheme for the fractional Brownian motion integral is provided. The PSD of a fractional Gaussian noise is determined by H, so the 1/f-type match claimed in Fig. 9 cannot be reproduced by the reader and cannot be attributed to the model rather than to an unquantified choice of H. Since H is the parameter that controls the self-similarity and long-range dependence that the paper emphasizes, its omission makes the validation of Eq. (7) incomplete and the simulated PSD a genuinely unverifiable result.","section":"Section IV-C, Table I, and Fig. 9"},{"comment":"Eq. (8) is described as a \"mean-reverting Ornstein–Uhlenbeck-type square-root diffusion process,\" but the equation shown is dγ(t) = κ(γ̄ − γ(t)) dt + σγ dW_t, which is an additive-noise Ornstein–Uhlenbeck process. It is not a square-root diffusion (Cox–Ingersoll–Ross type), and it does not enforce γ(t) ≥ 0. The constraint \"γ(t) ≥ 0\" is stated but no mechanism (reflecting boundary, transformation, or parameter restriction) is given. If the simulated γ(t) crosses zero, the multiplicative noise term γ_t Y_t changes sign, which is not discussed and may alter the qualitative behavior of the model.","section":"Section IV-B3, Eq. (8)"},{"comment":"The model is not independently validated. The seasonal trend φ_t is extracted by spectral decomposition of the same zero-mean queue length time series that is later used for all comparisons, and the parameters θ are fitted by minimizing the RMSE between the simulated trajectory and that same empirical series. Under these conditions, the simulated process is constructed to follow the empirical periodic mean and is optimized to track the empirical path, so the reported Pearson correlations, PDF overlap, and PSD similarity in Figs. 7–10 are largely by construction. The paper does not report any out-of-sample or hold-out evaluation, and it does not compare the fitted model's performance against simpler benchmarks (e.g., a periodic mean plus noise of known distribution). The validation therefore does not support the claim that the specific SDE structure, rather than the fitted seasonal component, is responsible for the observed agreement.","section":"Section IV-C, validation and Figs. 7–10"}],"minor_comments":[{"comment":"There are numerous typos and grammatical errors, e.g., \"Alafya\" for \"Alafaya\", \"Nyquiest\" for \"Nyquist\", \"Instantenous\" for \"Instantaneous\", \"becuase\" for \"because\", and incomplete sentences such as \"Based on the finding we\" at the start of Section IV.","section":"General"},{"comment":"The reference \"Hinguich and Musha\" should be \"Higuchi and Musha\" (Musha and Higuchi, 1976); the citation given as [20] in the text appears with reversed author order.","section":"Section II-A"},{"comment":"The text mentions \"models like Hull-White and Viscek\" but does not define these models or cite the relevant references; presumably \"Viscek\" refers to the Vicsek model, but this is unclear in the current form.","section":"Section II-B"},{"comment":"Eq. (6) defines X_n as the accumulated volume in interval n, but the subsequent discussion identifies X_n with the maximum queue length Y_t in a cycle without addressing the distinction between total arrivals during the cycle and the maximum number of queued vehicles, which is a physical mismatch that the stochastic term is invoked to cover.","section":"Section IV-B2, Eq. (6)"},{"comment":"The figure caption states that a reference line with a 45-degree slope is included, but the axes of the scatter plot have different scales after smoothing; the claim that \"closer alignment of points along this line indicates better model performance\" is therefore not geometrically calibrated.","section":"Section IV-D, Fig. 8"}],"recommendation":"reject","confidential_remarks":"The paper addresses an interesting question and the empirical spectral analysis of cycle-level queue lengths is of some value. However, the core derivation is missing, a key model parameter (Hurst exponent) is left uncalibrated, the volatility submodel is mis-specified, and the validation is circular. These are not local presentation issues; they concern the central claims of the paper. Even under a permissive standard, the manuscript would require substantial reworking of the derivation, the calibration protocol, and the validation methodology to make the results reproducible and meaningful. The novelty claim of being \"the first equation-based model\" also seems overstated in the absence of a proper derivation, since the proposed equation is effectively a parametric statistical model."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper proposes dY_t = μ(φ_t − Y_t)dt + γ_t Y_t dW_t^H for queue length at signalized intersections, with φ_t a periodic mean and W_t^H fractional Brownian motion. That specific combination is new to the traffic-SDE literature, and the authors do a decent job motivating it from observed 1/f spectra and from a micro-model of headways with multiplicative noise. The empirical PSD analysis of four real intersections is also useful and reproduces the known 1/f pattern. Give credit: the paper is clearly written, the data are real, and the reported PDF/PSD matches are visually plausible.\n\nThe softer spots are mostly concentrated in the bridge between the micro model and the SDE, and in the validation. Equation (7) is not derived from the point-process model; it is asserted after Eq. (5) with the sentence that the noise 'will still remain multiplicative' in the cumulative process. That may be true in some scaling limit, but no aggregation argument is given, and the mean-reversion term and the fractional driver appear without derivation. This is the load-bearing step, and it is currently a modeling choice rather than a derivation. Second, the Hurst exponent H, which controls the spectral slope of the simulated PSD, is never estimated or reported. The PSD match in Fig. 9 therefore says as much about the chosen H as about the model. Third, the validation is circular in an ordinary way: φ_t is extracted from the same series being validated, and parameters are fit by RMSE to that series. So the 'replication' is an in-sample fit. Also, Eq. (8) is called a square-root diffusion but is an additive-noise OU process, so γ_t can go negative.\n\nNone of these are fatal to the idea, but they break the claim that the paper is the 'first equation-based model' that demonstrably replicates the statistics. The model could be useful as a compact generative tool for simulation and for physics-informed learning, if repositioned with out-of-sample validation and full parameter reporting.\n\nI would send this to review. A serious referee can push for the missing aggregation argument, the H estimate, and a proper train/test split. The paper has enough substance and enough real data to justify that effort.","headline":"A novel SDE for queue lengths that shows plausible fits but is held back by an asserted micro-to-macro derivation, an unreported Hurst exponent, and circular validation.","tokens_in":14913,"tokens_out":2375,"would_cite":false,"duration_ms":22971,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H10","60G22","90B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proposes a stochastic differential equation that reproduces both the distribution and the 1/f power spectrum of queue lengths at signalized intersections, and claims it is the first equation-based model for queue dynamics.","keywords":["stochastic differential equation","queue length","fractional Brownian motion","1/f noise","self-similarity","signalized intersection","power spectral density","multiplicative noise"],"falsifier":"Take a calibrated model for one intersection, simulate a 90-day trajectory, and compare the PSD and the empirical distribution against the observed time series on days excluded from calibration; the central claim fails if the simulated spectrum no longer shows the observed 1/f power-law decay or if a Kolmogorov–Smirnov test on held-out daily distributions rejects the match. A more direct check is to estimate the Hurst exponent from the data by rescaled-range or detrended fluctuation analysis and verify that the fitted H parameter reproduces it.","tokens_in":13834,"feed_emoji":"🚦","tokens_out":5086,"duration_ms":47548,"temperature":0.7,"pith_summary":"This paper proposes a single stochastic differential equation as a generative model for the cycle-to-cycle evolution of queue lengths at signalized intersections. The equation combines mean reversion toward a periodic daily-demand curve, multiplicative noise whose volatility itself follows a mean-reverting process, and a fractional Brownian motion driver with a Hurst exponent that controls long memory. The authors argue that this compact equation reproduces the two statistical signatures they measure in real data: the probability distribution of queue lengths and the 1/f-type power spectral density. They further claim that this is the first equation-based model of queue dynamics. If correct, the model offers an interpretable alternative to opaque deep-learning predictors and a physics-structured prior for hybrid learning systems.","feed_headline":"One SDE recreates intersection queue statistics","feed_subtitle":"A mean-reverting fractional Brownian model reproduces the 1/f spectra and distributions of real queues.","key_machinery":"The central object is the SDE itself, composed of three ingredients. Mean reversion toward the periodic mean $\\phi_t$ encodes the daily traffic demand and the negative feedback introduced by adaptive signal control. Multiplicative noise $\\gamma_t Y_t$ makes fluctuations scale with queue size and is motivated microscopically: if vehicle interarrival times follow a lognormal-type multiplicative process, the reciprocal instantaneous flow inherits multiplicative noise via Itô's lemma, and the authors assert that this survives aggregation to cycle level. Fractional Brownian motion $W_t^H$ supplies correlated increments, with $H > 0.5$ producing the long-memory, $1/f$-type spectral structure observed in the data. The volatility $\\gamma_t$ is itself an Ornstein–Uhlenbeck square-root process, adding stochastic volatility.","core_discovery":"The paper's central claim is that queue length dynamics at a signalized intersection can be described by $dY_t = \\mu(\\phi_t - Y_t)\\,dt + \\gamma_t Y_t\\,dW_t^H$, where $Y_t$ is the maximum queue length in a cycle, $\\phi_t$ is a periodic mean estimated from spectral decomposition, $\\gamma_t$ follows a mean-reverting square-root diffusion, and $W_t^H$ is fractional Brownian motion. The authors report that simulations calibrated per intersection reproduce the empirical PDF and preserve the roughly $1/f$ power-law decay seen in the data, with Pearson correlations above 0.74 between simulated and observed series over a week. The intended contribution is not a first-principles derivation but a statistically faithful, transparent equation that captures quasiperiodicity, multiplicative fluctuations, and long-range dependence.","pith_inferences":["A testable extension the paper leaves implicit is to use the SDE for short-horizon queue-length forecasting, conditioning on the current estimated state and $\\gamma_t$, and benchmark it against the deep-learning baselines the paper cites.","The paper does not derive the relationship between the fitted Hurst exponent and the measured 1/f slope; a natural check is whether the Hurst exponent estimated from the data by detrended fluctuation analysis matches the calibrated value across intersections.","One could stress the microscopic-to-macroscopic bridge by building the cumulative arrival process explicitly from a lognormal interarrival model and verifying numerically that the aggregated process's noise is indeed multiplicative before trusting the SDE as an equation of motion."],"forward_implications":["If the SDE is right, a handful of parameters per intersection—mean-reversion speed, volatility dynamics, Hurst exponent, and the periodic mean—encode the statistical behavior of queue length time series.","The model gives a generative simulator: one can draw new queue-length trajectories with realistic distributions and spectra without running a traffic microsimulation.","Its interpretable structure separates deterministic daily demand from stochastic fluctuation, allowing the two effects to be analyzed and controlled independently.","Because the SDE reproduces statistical signatures rather than exact trajectories, it is naturally suited as a physics-informed prior or regularizer for neural-network predictors.","The calibrated parameter set could serve as a compact descriptor for comparing the behavior of different signalized intersections."],"supporting_citations":[{"why":"Supplies the queue length dataset and cleaning procedure used for calibration and validation.","marker":"[10]"},{"why":"Documents long-range dependence in traffic flow time series, motivating the fractional-Brownian driver.","marker":"[17]"},{"why":"Documents self-similarity in highway traffic, part of the empirical motivation for the model.","marker":"[19]"},{"why":"Provides the original observation of 1/f fluctuation in traffic current, anchoring the PSD analysis.","marker":"[20]"},{"why":"Models 1/f noise generation from multiplicative point processes, supporting the noise structure of the SDE.","marker":"[54]"},{"why":"Shows that multiplicative noise in headway dynamics yields a lognormal distribution, the microscopic basis for multiplicative noise.","marker":"[55]"},{"why":"Gives simulation evidence that interarrival times at signalized intersections follow lognormal-type distributions.","marker":"[57]"}],"fun_headline_variants":["Mean-reverting SDE with fractional noise matches real queue data","First equation-based model of queue length dynamics","Stochastic model reproduces intersection queue statistics","SDE for queues with fractional Brownian motion fits data"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that multiplicative noise at the vehicle-headway level remains multiplicative after aggregation over a signal cycle, so that the macroscopic SDE inherits its noise structure from the microscopic point-process model; no aggregation theorem or derivation is supplied for that step.","fun_headline_variants_meta":{"raw":{"variants":["Mean-reverting SDE with fractional noise matches real queue data","First equation-based model of queue length dynamics","Stochastic model reproduces intersection queue statistics","SDE for queues with fractional Brownian motion fits data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000575,"raw_usage":{"total_tokens":2659,"prompt_tokens":833,"completion_tokens":1826,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":449,"completion_tokens_details":{"reasoning_tokens":1764}},"tokens_in":449,"tokens_out":1826,"duration_ms":13823,"temperature":1.0,"reasoning_tokens":1764,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:55:14.312962+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a calibrated model for one intersection, simulate a 90-day trajectory, and compare the PSD and the empirical distribution against the observed time series on days excluded from calibration; the central claim fails if the simulated spectrum no longer shows the observed 1/f power-law decay or if a Kolmogorov–Smirnov test on held-out daily distributions rejects the match. A more direct check is to estimate the Hurst exponent from the data by rescaled-range or detrended fluctuation analysis and verify that the fitted H parameter reproduces it.","supporting_citations":[{"cited_title":"Real-time signal queue length prediction using long short-term memory neural network,","cited_arxiv_id":null,"evidence_quote":"Supplies the queue length dataset and cleaning procedure used for calibration and validation."},{"cited_title":"A better understanding of long-range temporal dependence of traffic flow time series,","cited_arxiv_id":null,"evidence_quote":"Documents long-range dependence in traffic flow time series, motivating the fractional-Brownian driver."},{"cited_title":"Self-similarity in highway traffic,","cited_arxiv_id":null,"evidence_quote":"Documents self-similarity in highway traffic, part of the empirical motivation for the model."},{"cited_title":"The 1/f fluctuation of a traffic current on an expressway,","cited_arxiv_id":null,"evidence_quote":"Provides the original observation of 1/f fluctuation in traffic current, anchoring the PSD analysis."},{"cited_title":"Models for generation 1/f noise,","cited_arxiv_id":null,"evidence_quote":"Models 1/f noise generation from multiplicative point processes, supporting the noise structure of the SDE."},{"cited_title":"Modelling traffic flow fluctuations,","cited_arxiv_id":null,"evidence_quote":"Shows that multiplicative noise in headway dynamics yields a lognormal distribution, the microscopic basis for multiplicative noise."},{"cited_title":"Arrival processes at traffic intersections,","cited_arxiv_id":null,"evidence_quote":"Gives simulation evidence that interarrival times at signalized intersections follow lognormal-type distributions."}],"review_version":1}