Pith. sign in

REVIEW 4 major objections 5 minor 55 references

BayesSeg: A Bayesian Optimization Framework for State Segmentation of Electricity Consumption Time Series

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Bayesian optimization finds near-optimal segmentation parameters in roughly 100 evaluations, matching a million-point grid search within 0.35% on real household electricity data.

desk verdict The BO-vs-gridsearch story is internally consistent, but the event_F1 rule in §2.2 rewards placing a steady segment inside a transition, so the headline event and composite scores do not measure switching-event quality. read the letter →

arxiv 2608.00513 v1 pith:MD4STFOK submitted 2026-08-01 cs.AI physics.data-an

Zhenya Zhang , Wendi Zhu , Ping Wang , Hongmei Cheng , Shuguang Zhang This is my paper · ORCID
classification cs.AIphysics.data-an
keywords non-intrusiveloadmonitoringtime-seriessegmentationsteady-statedetectionevent-levelF1normalizedmutualinformationBayesianoptimizationTPEsurrogatemodelelectricityconsumptionstatetimeseries
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

BayesSeg claims that electricity-load segmentation can be made fully automatic: instead of hand-tuning thresholds, a Bayesian optimizer chooses them. The paper's pipeline labels each point as steady or transition using a local rule based on the tail and mean of the running steady segment, maps the result to a binary sequence, scores it with an equal-weighted combination of event-level F1 and normalized mutual information, then searches the two thresholds with a tree-structured Parzen estimator. On the SustDataED2 Philips TV record, roughly 100 evaluations find parameters whose composite score is within 0.35% of the best of 1,002,001 grid-search combinations, while cutting search time from about 5,300 seconds to under one second. The paper argues this removes the manual calibration bottleneck in non-intrusive load monitoring and transfers to any time series with stable and transient regimes.

What carries the argument

The load-bearing object is the pair (ε, Δ) in the dual steady-state criterion: a new point s joins the current steady segment if |s − tail(es)| < ε, or if that fails but |s − mean(es)| < Δ. This local rule, applied sequentially and then complemented, fixes which segments exist. The evaluation objective is y = 0.5·event_F1 + 0.5·NMI, where event_F1 uses tolerance τ to match predicted steady segments to ground-truth switching events, and NMI compares the binary state label sequences. The optimizer is the tree-structured Parzen estimator: it splits past trials into good and bad sets by y, estimates densities l(x) and g(x), and proposes the next point maximizing l(x)/g(x). The argument depends o

What would settle it

On the SustDataED2 Philips TV labels, implement the Section 2.2 matching condition exactly and inspect which predicted segments are counted as hits. If predicted steady segments lying inside ground-truth transition intervals are matched as events, the event_F1 objective is inverted and the composite scores in Tables 1–2 must be recomputed under the intended transition-overlap rule; if no such inversion appears, the reported event_F1 stands. A second, independent check: run BO with τ=2 and 100 evaluations over 50 seeds and verify the best composite stays within 0.35% of 0.717455.

Watch

Extended reading notes

Core claim

The paper's central claim is that a three-layer segmentation–evaluation–optimization loop can replace manual parameter tuning for electricity consumption state series. The segmentation layer defines steady-state segments by a dual local-statistics criterion—a point is absorbed into a steady candidate if it is close to the candidate's last value, or if that fails but is close to the candidate's mean—and extracts transition segments as the complement between consecutive steady segments. The evaluation layer projects both prediction and ground truth to binary steady/transient sequences, computes event_F1 by tolerance matching with slack parameter τ and NMI for global structure, and combines the

Load-bearing premise

The framework's headline numbers rest on treating a steady segment that lies inside a true transition interval as a successful detection of that transition; read literally, that rule rewards the opposite of what event detection should reward, so the high scores do not necessarily measure switching-event quality.

Editorial extensions

If this is right

  • Segmentation thresholds can be re-fit automatically for each new household, meter, or appliance in under a second, removing the manual calibration step.
  • Because the plateau of near-optimal composite scores is broad, precise threshold values matter less than entering the right region; a cheap optimizer suffices.
  • Combining event-level F1 with NMI gives a metric that can distinguish parameter quality where pointwise F1 saturates or collapses.
  • The framework's independence from domain features means the same segmentation–evaluation–optimization loop can be dropped into other steady/transient time-series tasks, such as industrial process or biomedical monitoring.
  • In deployment, the budget can be cut from 100 to roughly 30 evaluations with only about 1% loss, making the optimizer suitable for online adaptation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The published event-matching condition counts a steady segment inside a ground-truth transition interval as a hit for that transition; if implemented literally, reported event_F1 values would need recomputation under the inverse transition-overlap semantics before interpreting them as event-detection quality.
  • A natural stress test is to shrink the evaluation budget to 20–30 iterations and measure the spread of composite scores across many seeds; the broad plateau predicts little degradation, while a spike in variance would suggest the claim depends on the 100-iteration budget.
  • Since α=0.5 is fixed, the reported optimal parameters are only optimal for that weighting; treating event_F1 and NMI as separate objectives and tracing the Pareto frontier would show how the optimal (ε, Δ) region changes when detection rate matters more than structural consistency.
  • The single-appliance, single-household validation leaves the domain-agnostic claim open; the immediate next test is multi-appliance overlapping operation and coarser sampling rates, where the plateau structure may not survive.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes BayesSeg, a pipeline for segmenting electricity consumption time series into steady and transition states using two thresholds (ε, Δ), evaluating with a composite score of event-level F1 and NMI, and tuning the thresholds via TPE-based Bayesian optimization. On a single Philips TV series from SustDataED2, it reports that BO with 100 evaluations reaches a composite score of 0.7149 and event_F1 of 0.9340, within 0.35% of an exhaustive 1,002,001-point grid search, at a claimed speedup over 5700×. It also compares BO with PSO, GA, and random search under paired seeds.

Significance. The BO-versus-grid-search comparison is a meaningful engineering result if the objective being optimized is legitimate, and the paper usefully documents the plateau structure of its composite score and performs a paired multi-optimizer study. The central evaluation layer, however, appears to define event matching in a way that rewards the inverse of event detection, so the reported event_F1 and composite scores do not currently measure the paper's stated contribution. The single-appliance, in-sample nature of the experiments further limits the generality claims. The significance is therefore conditional on correcting the matching definition and re-running the evaluation and optimization experiments.

major comments (4)
  1. [§2.2, Eq. (9), Table 2] The event_F1 matching rule is semantically inverted. The paper defines real events as switching events (transition intervals), but declares a successful match when a predicted steady-state segment s_j lies inside the true event interval: start(g_i)+τ ≤ start(s_j) and end(s_j) ≤ end(g_i)−τ. A correct event detector should output a transition segment overlapping the true event; a predicted steady segment inside the event is a false negative. Implemented literally, this rule rewards exactly the failure mode the abstract says the method avoids. Because the composite objective uses event_F1, the parameter selection and the headline numbers (event_F1=0.9340, composite=0.7149) measure the tendency to place steady segments inside transitions, not switching-event detection quality. The matching should be redefined in terms of predicted transition segments (or the complement of steady segments) an
  2. [§4.1–4.2, Table 2] All reported optimal parameters are in-sample maxima computed on the same Philips TV series used to select them. There is no train/test split, cross-validation, or evaluation on other appliances/households. The conclusion acknowledges the single-appliance limitation, but this is not a minor caveat: the claimed 'automated configuration' and domain-agnostic generalization (Contribution 4) require evidence that optimized parameters transfer. As it stands, the 0.35% deviation claim is about matching the grid-search optimum of the same training objective, not about predictive performance on unseen data.
  3. [§4.3–4.4, Tables 4 and 5] The convergence reporting is inconsistent. Table 4 states that for τ=2–6 the 99%-of-optimum threshold requires 28 iterations, while Table 5 reports for τ=4 a mean first_hit of 16.6±7.6 (50/50 success) for the same threshold. Table 4's first_hit values also appear to come from a single run rather than a distribution over seeds. Since convergence speed is part of the speedup claim, please clarify whether these are single-run or averaged values and reconcile the discrepancy.
  4. [§4.4, Table 5] The statistical comparison shows that PSO is not significantly worse than BO for τ≥2 after Bonferroni correction (p>0.0167; Cohen's d≈0.27–0.30), and PSO is faster in wall-clock time (≈300 ms vs ≈530 ms). The text nevertheless concludes that BO 'leads in both solution quality and convergence speed.' The data support only a non-significant quality advantage and a runtime disadvantage. Please temper this claim or provide a stronger argument—beyond statistical significance—for preferring TPE in this application.
minor comments (5)
  1. [Definition 1 / Eq. (1) and Algorithm 1] Eq. (1) is typeset in a corrupted/malformed way; the steady-state condition should be written explicitly so that it matches the JudgeState pseudocode, which uses T=|s−es[|es|]| and M=|s−mean(es)|.
  2. [§2.2] Please define precisely whether each real event g_i is a point index or an interval, and how start(g_i) and end(g_i) are derived from the ground-truth transition labels. This ambiguity is critical given the tolerance-based matching.
  3. [Tables 2–4] Report the random seed(s) and number of repetitions used for Tables 2–4. Table 5 uses seeds 42–91 and reports means and standard deviations, but Tables 2–4 appear to be single-run results, which makes the headline 0.35% deviation fragile.
  4. [Abstract and Table 2] The abstract reports event_F1=0.9340, but Table 1 shows that the maximum event_F1 at τ=2 is 0.978723 for a different parameter set. Please clarify that 0.9340 is the event_F1 at the BO-selected parameters, not the global maximum.
  5. [References] Several references are dated 2026 (Refs. 3, 19, 24). If these are in-press or preprints, please add appropriate arXiv/DOI identifiers so readers can access them.

Circularity Check

1 steps flagged · score 6.0 of 10

event_F1 is defined by an inverted matching rule, so the headline event and composite scores are definitional outputs rather than measurements of switching-event quality.

  1. self definitional [Section 2.2, event_F1 matching rule; Abstract; Eq. (9)]
    "if there exists a steady-state segment s j ∊ S such that start( gi) + τ ≤ start( sj) and end( sj) ≤ end( gi) – τ ... then s j is considered a successful match for g i."

    The event_F1 matching rule defines a successful event detection as a predicted steady-state segment contained inside the true transition interval. Thus event_F1 (and the composite score in Eq. 9, which is a weighted combination with NMI) is by construction a count of how often the algorithm places steady segments inside transitions—not a count of how well it detects the transitions themselves. The abstract's claim that 'event_F1 quantifies switching-event precision and recall via tolerance matching' is therefore a self-referential relabeling: the reported event_F1 = 0.9340 and the parameter choices selected by maximizing the composite score are generated by this inverted definition, not by an independent measure of event-detection quality. The result reduces to the metric's own matching ru

full rationale

The core BO-vs-grid-search comparison is not circular: both methods optimize the same composite objective, and reporting that BO reaches within 0.35% of the grid-search optimum on that objective is a legitimate empirical optimization result. There is also no load-bearing self-citation chain. However, the evaluation layer that defines the objective contains a self-definitional flaw. The event_F1 matching rule explicitly treats a predicted steady segment lying inside a ground-truth transition interval as a successful event match. This is the inverse of event-detection semantics, and it makes event_F1 and the composite score measure a quantity defined by the matching rule itself. The headline 'event_F1 of 0.9340' and the selected optimal parameters are therefore outputs of the metric's own definition rather than evidence of switching-event detection quality. Because the paper's central evaluation claim reduces to this definitional rule, a score of 6 is appropriate: partial circularity in the evaluation/performance claim, while the optimization efficiency claim remains internally consistent but inherits the metric's definitional problem.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on two tuned thresholds (epsilon, delta), a hand-chosen tolerance tau and weight alpha in the composite metric, and labeled data for the objective. The dual-threshold segmentation heuristic is a domain assumption, not a derived law. No new physical or latent entities are introduced.

free parameters (5)
  • epsilon (steady-state tail threshold) = 3.97 (BO), 4.20 (GS) at tau>=1
    Key segmentation threshold controlling whether a point joins the current steady segment based on tail deviation; optimized by BO/GS.
  • delta (steady-state mean threshold) = 3.80 (BO), 2.07 (GS) at tau>=1
    Threshold on deviation from the current segment mean; optimized jointly with epsilon.
  • tau (tolerance in event matching) = 0 to 6; headline at tau=2
    Tolerance window for boundary deviations in event_F1; swept across seven values and the abstract's numbers correspond to tau=2.
  • alpha (composite weight) = 0.5
    Weight between event_F1 and NMI in Eq. (9); fixed without sensitivity analysis.
  • TPE good/bad quantile = 10%
    Threshold y* that splits observed parameters into good/bad sets in Eq. (10); standard but chosen.
assumptions (4)
  • domain assumption Ground-truth state labels for the time series are available and correct for computing the composite objective
    The optimization and all evaluation metrics rely on labeled binary state sequences from SustDataED2; without labels the automatic parameter selection cannot run.
  • domain assumption The dual threshold criterion (tail value and mean) characterizes steady-state segments in electricity data
    Algorithm 1 and Definition 1 assume that local stability can be detected from these two statistics; this is a heuristic modeling choice not derived from physics.
  • standard math NMI on binary state sequences is a valid global structural consistency measure for segmentation
    NMI is a standard information-theoretic measure, but its use here assumes the binary steady/transition mapping preserves segmentation semantics.
  • domain assumption The TPE surrogate model accurately extrapolates the composite score landscape from limited evaluations
    Bayesian optimization's convergence guarantee is heuristic; the claim that 100 evaluations suffice is empirical, not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BayesSeg: A Bayesian Optimization Framework for State Segmentation of Electricity Consumption Time Series." pith.science (2026). https://pith.science/paper/MD4STFOK

@misc{pith2026260800513,
  author       = {Pith},
  title        = {Pith review of: BayesSeg: A Bayesian Optimization Framework for State Segmentation of Electricity Consumption Time Series},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MD4STFOK}},
  note         = {Machine review of arXiv:2608.00513}
}
read the original abstract

In Non-Intrusive Load Monitoring (NILM), adaptive segmentation of electricity consumption time series is critical for appliance recognition. However, prevailing methods face challenges including heuristic parameter tuning, boundary sensitivity, and metric saturation. This paper proposes BayesSeg, a unified framework integrating time-series segmentation, multidimensional evaluation, and automatic parameter optimization. The segmentation layer employs a dual steady-state criterion based on the tail value and mean of preceding subsequences, combined with a sequential extraction and complement-set parsing strategy, to achieve precise unsupervised partitioning of steady-state and transition-state segments. The evaluation layer maps segmentation results to binary state sequences and formulates a composite metric integrating an event-level F1 score (event_F1) with Normalized Mutual Information (NMI). The event_F1 quantifies switching-event precision and recall via tolerance matching, while NMI captures global structural consistency, jointly overcoming the boundary sensitivity and limited discriminability of point-wise metrics. In the optimization layer, the composite score serves as the objective function for Bayesian optimization, which constructs a TPE surrogate model for efficient global parameter-space exploration. Experiments on the SustDataED2 dataset demonstrate that Bayesian optimization requires only ~100 objective evaluations to locate a parameter region within 0.35% deviation of the exhaustive grid-search optimum. The framework achieves a weighted composite score of 0.7149 and an event_F1 of 0.9340 while reducing optimization latency from ~5300 seconds to under 1 second, a speedup exceeding 5700x. BayesSeg automates segmentation configuration and provides a scalable, efficient solution for time-series analysis in NILM and related domains.

Figures

Figures reproduced from arXiv: 2608.00513 by the authors.

Figure 2
Figure 2. BayesSeg Segmentation Framework ε and Δ are two key parameters in the steady-state segment identification method, which is based on the tail value of the preceding subsequence and its mean, as defined in Eq. (1). Variations in their values will significantly affect the accuracy of the user's electricity consumption time-series segmentation. To quickly obtain optimal ε and Δ, [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Framework of Segmentation Parameter Optimization To segment an electrical state time series, a candidate parameter pair (ε, Δ) is a 2-dimensional vector x = (x1, x2) T ∊ R2 . A single trial is the process of applying a candidate parameter vector to time-series segmentation, then evaluating the resulting segments using a target function. Let 1,1 1,2 2,1 2,2 ,1 ,2 ... n nn xx xx X xx =        be the compl… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages

  1. [1]

    Introduction In building energy management and demand- side response, modern electrical monitoring systems generate high-resolution electricity consumption time series from sensors such as smart meters, smart circuit breakers, and smart sockets [1-3, 16-20]. These time series capture dynamic variations in electrical parameters, such as current and power, ...

    work page

  2. [2]

    A comprehensive evaluation metric is constructed by integrating the event_F 1 score and NMI. The segmentation result is mapped to a binary state sequence of the same length as the original time series, enabling joint assessment along two complementary dimensions: event -detection accuracy and global structural consistency. The event _F1 measures detection...

    work page

  3. [3]

    BayesSeg, an automatic parameter-optimization framework for segmentation, is built on Bayesian optimization. Using the proposed composite score as the objective function, BayesSeg employs a sequential sampling strategy with surrogate model updates to efficiently search the parameter space for optimal values of the key parameters ε and Δ. In contrast to an...

    work page

  4. [4]

    A dual -criteria steady -state discrimination method and an automatic segmentation algorithm featuring sequential extraction and complement -set parsing are proposed. At the implementation level, the dual steady-state discrimination criteria, based on the tail value and mean of the preceding subsequence, are combined with the sequential extraction and com...

    work page

  5. [5]

    Although BayesSeg is developed and validated in the context of NILM, the proposed framework is inherently domain-agnostic. The segmentation layer relies solely on local statistical properties of the time series —specifically, the tail value and the mean of the preceding subsequence —without requiring domain -specific features or prior knowledge of the num...

    work page

  6. [6]

    When an appliance remains in a steady state, the electricity consumption profile on the supply circuit remains relatively stable

    Model and Methods During operation, electrical appliances routinely switch between operating states. When an appliance remains in a steady state, the electricity consumption profile on the supply circuit remains relatively stable. Conversely, as the appliance transitions to another state, the power trend exhibits pronounced fluctuations. Subsequent segmen...

    work page

  7. [7]

    if (T < ε) or (T ≥ ε and M < ∆)

    work page

  8. [8]

    The thresholds Δ and ε are defined by inequality (1)

    return flag In Algorithm 2, the input current_series is the full electricity consumption time series. The thresholds Δ and ε are defined by inequality (1). The output of Algorithm 2 is a steady-state segment. Algorithm 2: Sequential Extraction Algorithm for Steady-State Segments Function ExtractSteadySegments(current_series, ∆, ε) Input: current_series, ∆...

    work page

Show all 55 references
  1. [9]

    es =φ; steady_segments=φ

  2. [10]

    for each s in current_series

  3. [11]

    flag = JudgeState(es, s, ∆, ε)

  4. [12]

    steady_segments = steady_segments ∪{ es }

  5. [13]

    A novel segmentation approach for work mode boundary detection in MFR pulse sequence[J]

    CHI K., SHEN J., LI Y., et al. A novel segmentation approach for work mode boundary detection in MFR pulse sequence[J]. Digital Signal Processing, 2022, 126: 103462

    2022

  6. [14]

    seg = φ// The steady segment currently being traversed

  7. [15]

    start = 0; end =0; // Start and end indices of the current steady segment

  8. [16]

    last_end = 0; // The index immediately after the tail of the previous steady segment

  9. [17]

    for each seg in steady_segments do

  10. [18]

    (start, end) ←(headLoc(seg(head)) tailLoc(seg(tail)))

  11. [19]

    steady_segments are the output of Algorithm 2

    return steady_segments In Algorithm 3, current_series represents the full time series of electricity consumption. steady_segments are the output of Algorithm 2. headLoc(seg(head)) and tailLoc(seg(tail)) are used in step 5 of Algorithm 3 to locate the positions of the first and...

  12. [20]

    if start > last_end / *The current steady segment starts beyond last_end, indicating an unpartitioned region in between */

  13. [21]

    Append the subsequence corresponding to the index interval [last_end, start − 1] to transition_segments

  14. [22]

    Let 0 , ntt R ∈ with 0 0, 0ntt >> and 0 ntt <

    return transition_segments 2.2 Evaluation of Electricity Consumption State Time-Series Segmentation Let S = <s1, s2 … sn> be the time series of observations recorded over the time interval ( t0, tn] on a power line, where the observed physical quantities include current, volta...

  15. [23]

    As shown in Figure 2, BayesSeg comprises a segmentation module, an evaluation module, and a parameter -optimization module

    Segmentation Parameter Optimization Based on Bayesian Optimization To achieve adaptive segmentation of user electricity consumption state time series, a Bayesian optimization-based segmentation framework, BayesSeg, is presented in Figure 2. As shown in Figure 2, BayesSeg compr...

  16. [24]

    Experiment Results and Analysis 4.1. Experimental Setup SustDataED2 [15], the dataset used in this experiment, was collected from a real -world residential scenario involving a three-person household in Portugal over a continuous 96-day period. It contains power measurements a...

    1912

  17. [25]

    making weight decisions for the user

    Conclusion and Future Work The time-series segmentation method based on tail values and the means of preceding subsequences is an effective approach for segmenting user electricity consumption time series. Effective segmentation of these state sequences significantly reduces t...

    2022

  18. [26]

    A., ABID M

    KIRMANI S., MAZID A., KHAN I. A., ABID M. A survey on IoT -enabled smart grids: Technologies, architectures, applications, and challenges[J]. Sustainability, 2022, 15(1): 717

    2022

  19. [27]

    A Survey of the Research on Non-intrusive Load Monitoring and Disaggregation[J]

    CHENG Xiang, LI Linzhi, WU Hao, et al. A Survey of the Research on Non-intrusive Load Monitoring and Disaggregation[J]. Power System Technology, 2016, 40(10): 3108-3117

    2016

  20. [28]

    Non -intrusive load monitoring: A systematic review of methods, scenario- specific challenges, and pathways to practical deployment[J]

    XIANG H., SU W., ZONG Y. Non -intrusive load monitoring: A systematic review of methods, scenario- specific challenges, and pathways to practical deployment[J]. Energies, 2026, 19(8): 1883

    2026

  21. [29]

    Research on Feature Model and Mining Method for Current Transition Sequence[J]

    ZHANG Hui, LIU Shuai, YANG Zecheng, et al. Research on Feature Model and Mining Method for Current Transition Sequence[J]. Journal of Xinjiang University (Natural Science Edition in Chinese and English), 2024, 41(01): 37-51

    2024

  22. [30]

    A low -frequency residential NILM approach based on adaptive event detection[J]

    LUO Z., LI Z., ZHANG T., et al. A low -frequency residential NILM approach based on adaptive event detection[J]. Journal of Building Engineering, 2025, 106: 112543

    2025

  23. [31]

    Unsupervised time series segmentation: A survey on recent advances[J]

    WANG C., LI X., ZHOU T., et al. Unsupervised time series segmentation: A survey on recent advances[J]. Computers, Materials & Continua, 2024, 80(2): 2657-2673

    2024

  24. [32]

    Adaptive algorithms for change point detection in financial time series[J]

    MUSAEV A., GRIGORIEV D., KOLOSOV M. Adaptive algorithms for change point detection in financial time series[J]. AIMS Mathematics, 2024, 9(12): 35238-35263

    2024

  25. [33]

    XIA S. P. C., CHU L., PEI L., et al. A boundary consistency -aware multitask learning framework for joint activity segmentation and recognition with wearable sensors[J]. IEEE Transactions on Industrial Informatics, 2023, 19(3): 2984-2996

    2023

  26. [34]

    CLaP-State detection from time series[EB/OL]

    ERMSHAUS A., SCHÄFER P., LESER U. CLaP-State detection from time series[EB/OL]. arXiv:2504.01783,

    arXiv

  27. [35]

    GHARGHABI S., YEH C. C. M., DING Y., et al. Domain agnostic online semantic segmentation for multi - dimensional time series[J]. Data Mining and Knowledge Discovery, 2019, 33(1): 96-130

    2019

  28. [36]

    L I N J . F . S . , K A R G M . , K U L I Ć D . M o v e m e n t primitive segmentation for human motion modeling: A framework for analysis[J]. IEEE Transactions on Human-Machine Systems, 2016, 46(3): 325-339

    2016

  29. [37]

    Time2State: An unsupervised framework for inferring the latent states in time series data[J]

    WANG C., WU K., ZHOU T., et al. Time2State: An unsupervised framework for inferring the latent states in time series data[J]. Proceedings of the ACM on Management of Data, 2023, 1(1): 1-18

    2023

  30. [39]

    Analyzing the performance of biomedical time-series segmentation with electrophysiology data[J]

    REDINA R., HEJC J., FILIPENSKA M., et al. Analyzing the performance of biomedical time-series segmentation with electrophysiology data[J]. Scientific Reports, 2025, 15(1): 11776

    2025

  31. [40]

    A residential labeled dataset for smart meter data analy tics[J]

    PEREIRA L., COSTA D., RIBEIRO M. A residential labeled dataset for smart meter data analy tics[J]. Scientific Data, 2022, 9(1): 134

    2022

  32. [41]

    Transient event detection algorithm for non-intrusive load monitoring[J]

    NIU Lulu, JIA Hongji. Transient event detection algorithm for non-intrusive load monitoring[J]. Automation of Electric Power Systems, 2011, 35(09): 30-35

    2011

  33. [42]

    Nonintrusive load monitoring (NILM) using a deep learning model with a transformer-based attention mechanism and temporal pooling[J]

    IRANI AZAD M., RAJABI R., ESTEBSARI A. Nonintrusive load monitoring (NILM) using a deep learning model with a transformer-based attention mechanism and temporal pooling[J]. Electronics, 2024, 13(2): 407

    2024

  34. [43]

    Enhancing non-intrusive load monitoring through transfer learning with transformer models[J]

    RONG J., WANG C., ZHOU Q., et al. Enhancing non-intrusive load monitoring through transfer learning with transformer models[J]. Energy and Buildings, 2025, 330: 115334

    2025

  35. [44]

    Non-intrusive load monitoring model based on SimCLR and visualized color V-I trajectories[J]

    CHEN T., FAN Y., LI L., et al. Non-intrusive load monitoring model based on SimCLR and visualized color V-I trajectories[J]. Sensors, 2026, 26(4): 1230

    2026

  36. [45]

    N., et a l

    BEHROOZ T., MOSTAFA S., Mohammad R. N., et a l. A non-intrusive load monitoring algorithm based on real-time feature extraction and deep learning model[J]. International Journal of Electrical Power and Energy Systems, 2025, 168: 110637

    2025

  37. [46]

    Non-intrusive load monitoring based on time-enhanced multidimensional feature visualization[J]

    CHEN T., YUAN Y., GAO J., et al. Non-intrusive load monitoring based on time-enhanced multidimensional feature visualization[J]. Scientific Reports, 2025, 15:4800

    2025

  38. [47]

    Change -point detection with deep learning: A review[J]

    XU R., SONG Z., WU J., et al. Change -point detection with deep learning: A review[J]. Frontiers of Engineering Management, 2025, 12(1): 154-176

    2025

  39. [48]

    Automatic change -point detection in time series via deep learning[J]

    LI J., FEAR NHEAD P., FRYZLEWICZ P., et al. Automatic change -point detection in time series via deep learning[J]. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 2024, 86(2): 273-285

    2024

  40. [49]

    Online neural ne tworks for change-point detection[J]

    HUSHCHYN M., ARZYMATOV K., DERKACH D. Online neural ne tworks for change-point detection[J]. Machine Learning, 2026, 115: 56

    2026

  41. [50]

    Short-term power load forecasting based on Seq2Seq model integrating Bayesian optimization, temporal convolutional network and attention[J]

    DAI Y., YU W. Short-term power load forecasting based on Seq2Seq model integrating Bayesian optimization, temporal convolutional network and attention[J]. Applied Soft Computing, 2024, 166: 112248

    2024

  42. [51]

    A hybrid neural network based on Bayesian optimization for non-intrusive load disaggregation[C]

    BIE Y., TAN M., LI Z., et al. A hybrid neural network based on Bayesian optimization for non-intrusive load disaggregation[C]. 2022 6th Asian Conference on Artificial Intelligence Technology (ACAIT), Changzhou, China, 2022:1-8, doi: 10.1109/ACAIT56212.2022.10137824

    2022

  43. [52]

    R., KHALID S., et al

    YOUSAF M., SINGH A. R., KHALID S., et al. Enhancing HVDC transmission line fault detection using disjoint bagging and Bayesian optimization with artificial neural networks and scientometric insights[J]. Scientific Reports, 2024, 14: 23610

    2024

  44. [53]

    Evaluation metrics and statistical tests for machine learning[J]

    RAINIO O., TEUHO J., KLEN R. Evaluation metrics and statistical tests for machine learning[J]. Scientific Reports, 2024, 14: 6086

    2024

  45. [54]

    A closer look at classification evaluation metrics and a critical reflection of common evaluation practice[J]

    OPITZ J. A closer look at classification evaluation metrics and a critical reflection of common evaluation practice[J]. Transactions of the Association for Computational Linguistics, 2024, 12: 820-836

    2024

  46. [55]

    An experimental evaluation of anomaly detection in time series[J]

    ZHANG A., DENG S., CUI D., et al. An experimental evaluation of anomaly detection in time series[J]. Proceedings of the VLDB Endowment, 2024, 17(3): 483-496

    2024

  47. [2025]

    CLaP -- State Detection from Time Series

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.