Optimal Window Length Research Articles

Time-varying cable force identification in cable-stayed bridges by a high-resolution time-frequency method

Accurate monitoring of time-varying cable force plays a vital role in the condition assessment of cables. However, most current methods for identifying the cable force can only provide an average cable force over a period. The vibration method correlates the cable force with cable frequency via a formula, transforming the time-varying cable force identification into the instantaneous frequency (IF) identification of the cable. Synchroextracting transform (SET) is a powerful time-frequency method with the ability to identify the cable’s IF from the measured response. However, the performance of SET in identifying the IF is heavily influenced by the choice of window length, limiting its practical application. In this study, an improved SET (ISET) is proposed to overcome this limitation and is utilized to identify the time-varying force in bridge cables. In ISET, the variational mode decomposition is first employed to decompose the measured response into several mono-component modes. A hyperparameter optimization algorithm, adopting Rényi entropy as the evaluation index, is then introduced to determine the optimal window length for each mono-component mode. Furthermore, the proposed ISET with the optimal window length is used to identify the cable’s IF, thereby calculating the time-varying cable force using the vibration method formula. Cases involving finite element model of the cable, along with laboratory experiments on a scaled cable, show that the proposed method successfully addresses the limitations of SET and accurately identifies time-varying cable force. Finally, to validate the practicality of the proposed method, the field test data from an unsymmetric cable-stayed bridge is adopted. The results demonstrate that the proposed method is straightforward and efficient, capable of accurately monitoring the time-varying cable force in both new and existing cables of cable-stayed bridges.

Key environmental parameters and numerical prediction model of jellyfish bloom in Qingchuan Bay Nuclear Power Plant, China

In recent years, there has been a notable increase in cooling water intake blockage caused by marine organism blooms at coastal nuclear power plants worldwide, resulting in shutdowns of nuclear power plants and large economic losses. A sizable portion of these incidents were caused by blooms from jellyfish, a planktonic invertebrate with a unique growth pattern. Suitable external conditions are conducive to the rapid growth of jellyfish, and blooms can occur within a few days. In order to better predict jellyfish bloom and enable nuclear power plants to prepare for it in advance, this study explores the numerical relationship between jellyfish biomass and environmental parameters. A series of time windows (evaluation intervals) were defined and constructed by a time-series recursive approach, which solved the problem of poor correlation between jellyfish biomass and environmental parameters at non-bloom points. The optimal time window length D = 10 was obtained, and the key environmental parameters affecting jellyfish biomass were screened as sea surface temperature, salinity, voltage value, dissolved oxygen, and chlorophyll. According to the ADF and KPSS tests, the key parameters have no significant time dependence, which ensures the stability and reliability of the subsequent predictions. The jellyfish bloom prediction model was derived by calculating the score F through the recursive Principal Component Analysis of the key environmental parameter in the time interval preceding the prediction point. The sudden change moments of score F correspond well to the jellyfish bloom moments, and the sudden change moments are all advanced for a period of time compared to the bloom time, which can provide valuable time for the nuclear power plant to organize manpower to deal with the blockage. Finally, a maximum score F threshold model was proposed to be coupled with the jellyfish bloom prediction model to provide a more robust basis for early warning of jellyfish at nuclear power plants.

Classification of African ground pangolin behaviour based on accelerometer readouts: validation of bio-logging methods

BackgroundUnderstanding how free-ranging animals behave can help in the design of optimal management strategies to both conserve species and enable individuals to express natural behaviours, maximising welfare. Animal-borne accelerometers passively collect data on body acceleration which can be interpreted to quantify behaviour. Accelerometers are increasingly used in behavioural research, however validation of accelerometer readouts to enable classification of discrete behaviours is required for each species. Pangolins are a heavily trafficked clade of mammals, all of which are considered vulnerable to extinction. They are also under-researched, with little known about their behaviour in the wild. In this study, we present the first validation of behavioural classification based on accelerometer readouts for a pangolin species; the ground pangolin (Smutsia temminckii).ResultsWe present a standardised protocol for attaching accelerometers to pangolins to minimise the impact of devices on welfare. We match the readouts from accelerometers to behaviours defined through video observations. Using a random forest classification, we defined five discrete behaviours (walking, digging, feeding, investigating ground, and stationary, accuracy of 85%) and three activity levels (low, medium, and high, accuracy of 94%) from accelerometer readouts. We determine optimal sampling frequency and window length (50 Hz and five seconds for discrete behaviour, 10 Hz and seven seconds for activity level). We then deploy accelerometers and classify the behaviour of three free-ranging pangolins for between two and four days. We find considerable variation in peak daily activity between free-ranging pangolins with different individuals displaying nocturnal and crepuscular behaviour. We also find that pangolins spend the majority of their time (between 62 and 71%) at rest.ConclusionThe methods we present will enable the quantification of ground pangolin behaviour in the wild to improve our understanding of the species’ ecology and help inform conservation efforts. This will also help to improve our fundamental understanding of animal behaviour and ecology.

Intermittent fault detection in nonstationary processes via a Wald‐based control chart

SummaryIn this article, the problem of intermittent fault (IF) detection is investigated for nonstationary processes in the multivariate statistics framework. By combining the moving window technique with maximum likelihood estimation (MLE), the moving window Wald‐based control chart is proposed to realize the detection of IFs in nonstationary processes. The computational efficiency and the convergence properties are discussed for the designed iterative algorithm of MLE. Then, necessary and sufficient conditions are presented to guarantee the detectability of IFs with the consideration of window lengths. Moreover, the alarm delays are analyzed for the appearance and disappearance of IFs. In virtue of the above analysis, the optimal window length is derived by minimizing the supremum of alarm delays. In order to estimate the time of IFs' appearance and disappearance, an algorithm is designed with the inspiration of simulated annealing strategy. Finally, a simulation on rotary steerable drilling tool system is provided to verify the effectiveness of the proposed method.

Predicting the failure of rock using critical slowing down theory on acoustic emission characteristics

The joint surfaces and size affect the mechanical behavior and failure precursors of shale. Therefore, this paper studies the mechanical behavior, the response characteristics of acoustic emission (AE) and critical slowing down (CSD) characteristics of AE for shale with different joint surfaces and size. The results show that with an increase in the number of joint surfaces and the size of the sample, the crack initiation stress, uniaxial compressive strength (UCS), peak strain, and fractal dimension decrease. During the plastic stage, the occurrence of a sharp rise in AE ringing count, RA, a low-frequency, high-amplitude AE signal, and a rapid drop in AF can be characteristics of impending sample failure. Variance and autocorrelation coefficients can be used to characterize the CSD phenomenon. The optimal window length and lag step are determined to be 400 and 100, respectively. The CSD characteristics of the autocorrelation coefficient and the variance of multi-AE parameters are obtained. The variance-based early warning time provides short-term early warning, while the autocorrelation coefficient-based early warning time provides early warning. The research results can provide theoretical references for the mechanical design and failure prediction of underground rock engineering.

Seismic data filtering using deconvolutive short-time Fourier transform

Cohen class distributions are well-known time-frequency distributions that have been widely used to analyze different signals. However, they are impractical for signal filtering because they only provide amplitude spectra and suffer from cross terms. In this paper, deconvolutive short-time Fourier transform (DSTFT) is developed by estimating phase spectra and updating moduli to address residual cross terms. The DSTFT moduli is used as weights and apply a weighted least-squares technique to estimate a high-resolution and almost cross-term-free Wigner-Ville distribution. Through numerical tests, it has been found that choosing the optimal window length can minimize cross terms in short-time Fourier transform and DSTFT spectrograms. Furthermore, using thresholding or reweighting techniques can effectively eliminate weights associated with noise. The performance of our method is demonstrated using synthetic and two real seismic wavefield separation problems, including ground-roll removal in seismic shot records and polarization analysis in seismology. The results indicate the high performance of our method in estimating phase spectra and filtering seismic data.

Developing a multivariate time series forecasting framework based on stacked autoencoders and multi-phase feature

Time series forecasting across different domains has received massive attention as it eases intelligent decision-making activities. Recurrent neural networks and various deep learning algorithms have been applied to modeling and forecasting multivariate time series data. Due to intricate non-linear patterns and significant variations in the randomness of characteristics across various categories of real-world time series data, achieving effectiveness and robustness simultaneously poses a considerable challenge for specific deep-learning models. We have proposed a novel prediction framework with a multi-phase feature selection technique, a long short-term memory-based autoencoder, and a temporal convolution-based autoencoder to fill this gap. The multi-phase feature selection is applied to retrieve the optimal feature selection and optimal lag window length for different features. Moreover, the customized stacked autoencoder strategy is employed in the model. The first autoencoder is used to resolve the random weight initialization problem. Additionally, the second autoencoder models the temporal relation between non-linear correlated features with convolution networks and recurrent neural networks.Finally, the model's ability to generalize, predict accurately, and perform effectively is validated through experimentation with three distinct real-world time series datasets. In this study, we conducted experiments on three real-world datasets: Energy Appliances, Beijing PM2.5 Concentration, and Solar Radiation. The Energy Appliances dataset consists of 29 attributes with a training size of 15,464 instances and a testing size of 4239 instances. For the Beijing PM2.5 Concentration dataset, there are 18 attributes, with 34,952 instances in the training set and 8760 instances in the testing set. The Solar Radiation dataset comprises 11 attributes, with 22,857 instances in the training set and 9797 instances in the testing set. The experimental setup involved evaluating the performance of forecasting models using two distinct error measures: root mean square error and mean absolute error. To ensure robust evaluation, the errors were calculated at the identical scale of the data. The results of the experiments demonstrate the superiority of the proposed model compared to existing models, as evidenced by significant advantages in various metrics such as mean squared error and mean absolute error. For PM2.5 air quality data, the proposed model's mean absolute error is 7.51 over 12.45, about ∼40% improvement. Similarly, the mean square error for the dataset is improved from 23.75 to 11.62, which is ∼51%of improvement. For the solar radiation dataset, the proposed model resulted in ∼34.7% improvement in means squared error and ∼75% in mean absolute error. The recommended framework demonstrates outstanding capabilities in generalization and outperforms datasets spanning multiple indigenous domains.

Continuous Locomotion Mode and Task Identification for an Assistive Exoskeleton Based on Neuromuscular-Mechanical Fusion.

Human walking parameters exhibit significant variability depending on the terrain, speed, and load. Assistive exoskeletons currently focus on the recognition of locomotion terrain, ignoring the identification of locomotion tasks, which are also essential for control strategies. The aim of this study was to develop an interface for locomotion mode and task identification based on a neuromuscular-mechanical fusion algorithm. The modes of level and incline and tasks of speed and load were explored, and seven able-bodied participants were recruited. A continuous stream of assistive decisions supporting timely exoskeleton control was achieved according to the classification of locomotion. We investigated the optimal algorithm, feature set, window increment, window length, and robustness for precise identification and synchronization between exoskeleton assistive force and human limb movements (human-machine collaboration). The best recognition results were obtained when using a support vector machine, a root mean square/waveform length/acceleration feature set, a window length of 170, and a window increment of 20. The average identification accuracy reached 98.7% ± 1.3%. These results suggest that the surface electromyography-acceleration can be effectively used for locomotion mode and task identification. This study contributes to the development of locomotion mode and task recognition as well as exoskeleton control for seamless transitions.

Variance of entropy for testing time-varying regimes with an application to meme stocks

Shannon entropy is the most common metric for assessing the degree of randomness of time series in many fields, ranging from physics and finance to medicine and biology. Real-world systems are typically non-stationary, leading to entropy values fluctuating over time. This paper proposes a hypothesis testing procedure to test the null hypothesis of constant Shannon entropy in time series data. The alternative hypothesis is a significant variation in entropy between successive periods. To this end, we derive an unbiased sample entropy variance, accurate up to the order O(n-4)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$O(n^{-4})$$\\end{document} with n the sample size. To characterize the variance of the sample entropy, we first provide explicit formulas for the central moments of both binomial and multinomial distributions describing the distribution of the sample entropy. Second, we identify the optimal rolling window length to estimate time-varying Shannon entropy. We optimize this choice using a novel self-consistent criterion based on counting significant entropy variations over time. We corroborate our findings using the novel methodology to assess time-varying regimes of entropy for stock price dynamics by presenting a comparative analysis between meme and IT stocks in 2020 and 2021. We show that low entropy values correspond to periods when profitable trading strategies can be devised starting from the symbolic dynamics used for entropy computation, namely periods of market inefficiency.

Predicting Chlorophyll-a Concentration and Harmful Algal Blooms in Lake Okeechobee Using Time-Series MODIS Satellite Imagery and Long Short-Term Memory

Highlights A 10-year dataset of time-series MODIS imagery and in situ Chl-a concentration were curated for Lake Okeechobee. LSTM significantly outperformed KNN, SVR, and RF for Chl-a prediction and subsequent HAB detection. The optimal window length was found to be 13 days with a 4-day temporal resolution for the LSTM model. KNN, SVR, and RF models were not effective at utilizing the temporal dynamics of the input features. Abstract. Harmful algal blooms (HABs) in inland waterbodies are a global concern due to their negative impact on human, animal, and ecosystem health. Chlorophyll-a (Chl-a) concentration is an important water quality parameter for monitoring HABs. While statistical and machine learning (ML) models have been widely studied to predict Chl-a concentration and HABs based on single-time-point satellite data, this work assessed whether long short-term memory (LSTM) can improve both tasks by leveraging temporal features in time-series MODIS satellite images compared to three classical ML models, including k-nearest neighbor (KNN), support vector regression (SVR), and random forest (RF). A dataset of daily MODIS images and monthly in situ Chl-a concentration measurements from 2011 to 2020 was curated for Lake Okeechobee, Florida. A window size of 13 days with a temporal resolution of four days was found to produce the optimal performance for LSTM, which significantly outperformed KNN, SVR, and RF for Chl-a prediction with a root mean square error of 11.95 µg/L, a mean absolute error of 8.55 µg/L, and a R2 value of 0.43. The superior performance of LSTM for Chl-a prediction was likely due to its ability to leverage the temporal dynamics in the features associated with HAB development. The Chl-a predictions were further used to determine HAB events, showing better accuracy and a significantly higher F1 score for LSTM over the other models. The study suggested that combining LSTM with high-temporal-resolution time-series data should be preferred over applying common ML models on time-series or single-time-point remote sensing data for Chl-a and HAB monitoring. Keywords: Cyanobacteria, LSTM, Machine Learning, Remote Sensing, Water Quality.

Bedtime regularity predicts positive affect among veterans with posttraumatic stress disorder: an ecological momentary assessment study

BackgroundRegularizing bedtime and out-of-bed times is a core component of behavioral treatments for sleep disturbances common among patients with posttraumatic stress disorder (PTSD). Although improvements in subjective sleep complaints often accompany improvements in PTSD symptoms, the underlying mechanism for this relationship remains unclear. Given that night-to-night sleep variability is a predictor of physical and mental well-being, the present study sought to evaluate the effects of bedtime and out-of-bed time variability on daytime affect and explore the optimal window lengths of over which variability is calculated.MethodsFor about 30 days, male U.S. military veterans with PTSD (N = 64) in a residential treatment program provided ecological momentary assessment data on their affect and slept on beds equipped with mattress actigraphy. We computed bedtime and out-of-bed time variability indices with varying windows of days. We then constructed multilevel models to account for the nested structure of our data and evaluate the impact of bedtime and out-of-bed time variability on daytime affect.ResultsMore regular bedtime across 6–9 days was associated with greater subsequent positive affect. No similar effects were observed between out-of-bed time variability and affect.ConclusionsMultiple facets of sleep have been shown to differently predict daily affect, and bedtime regularity might represent one of such indices associated with positive, but not negative, affect. A better understanding of such differential effects of facets of sleep on affect will help further elucidate the complex and intertwined relationship between sleep and psychopathology.Trial registrationThe trial retrospectively was registered on the Defense Technical Information Center website: Award # W81XWH-15–2-0005.

A novel algorithm for robust estimation of ants' speed on convoluted trajectories derived from their gait pattern

AbstractAccurate measurements of travel distance and the corresponding speed are crucial for the analysis of animal movements. Particularly, the trajectories of ants were used in numerous behavioral studies. However, measurements of travel distance involve the dilemma of setting the proper time window: Estimates from a short time window are vulnerable to spatial errors in observation, while estimates from a long time window lead to an underestimation of the travel distance. To overcome these difficulties, we propose a novel algorithm that successively interpolates two consecutive points of an ant's trajectory for a given time window by embracing the alternating tripod gait and other gait patterns of ants. We demonstrate that this algorithm is more reliable compared with the conventional method of travel distance estimation based on the sum of the consecutive straight‐line displacement (SLD). After obtaining speed estimates for a range of sampling time windows, we applied a fitting method that can estimate the actual speed without prior knowledge of spatial error distribution. We compared results from several methods of speed estimation extracted from the empirical and simulated data of ant trajectories. The accuracy of our algorithm was comparable with or much higher than the accuracy of the sum over the consecutive SLD with optimal window length. Hence, subjective selection of the sampling time window can be avoided by using the proposed algorithm. We provide software that enables empirical scientists to utilize the proposed methods rather than the conventional SLD method of distance and speed estimation.

Differentiable Short-Time Fourier Transform Window Length Selection Driven by Cyclo-Stationarity

The Short-Time Fourier transform is widely applied in the condition monitoring of rotating machinery. Even so, selecting the optimal window length for the Short-Time Fourier Transform remains a challenge. This work presents a procedure for adapting the Short Time Fourier Transform algorithm to be differentiable with respect to window length by using continuous window functions defined over the entire input signal duration. Thanks to this modification, a differentiable loss criterion can be defined to measure the Short-Time Fourier quality, and the gradient of the loss criterion with respect to window length can be computed. The optimal window length for a given loss criterion can then be efficiently solved for using a gradient-based optimization algorithm. Results from a simulated bearing dataset and three experimental bearing datasets are used to compare the optimal spectrograms obtained using different loss criteria. Specifically, a sparsity-based loss criterion is compared with two loss criteria inspired by the characteristic cyclo-stationarity machine of faults in rotating machinery. The results demonstrate the effectiveness of the differentiable window length selection method and highlight the importance of selecting appropriate loss criteria for defining STFT quality. Loss criteria that account for the cyclo-stationary nature of the signals are shown to be less likely to target single high-amplitude impulsive events compared to the sparsity-based loss criterion.

Identification and analysis of autism spectrum disorder via large-scale dynamic functional network connectivity.

Autism spectrum disorder (ASD) is a prevalent neurodevelopmental disorder with severe cognitive impairment. Several studies have reported that brain functional network connectivity (FNC) has great potential for identifying ASD from healthy control (HC) and revealing the relationships between the brain and behaviors of ASD. However, few studies have explored dynamic large-scale FNC as a feature to identify individuals with ASD. This study used a time-sliding window method to study the dynamic FNC (dFNC) on the resting-state fMRI. To avoid arbitrarily determining the window length, we set a window length range of 10-75 TRs (TR = 2 s). We constructed linear support vector machine classifiers for all window length conditions. Using a nested 10-fold cross-validation framework, we obtained a grand average accuracy of 94.88% across window length conditions, which is higher than those reported in previous studies. In addition, we determined the optimal window length using the highest classification accuracy of 97.77%. Based on the optimal window length, we found that the dFNCs were located mainly in dorsal and ventral attention networks (DAN and VAN) and exhibited the highest weight in classification. Specifically, we found that the dFNC between DAN and temporal orbitofrontal network (TOFN) was significantly negatively correlated with social scores of ASD. Finally, using the dFNCs with high classification weights as features, we construct a model to predict the clinical score of ASD. Overall, our findings demonstrated that the dFNC could be a potential biomarker to identify ASD and provide new perspectives to detect cognitive changes in ASD.

A New Method for Heart Disease Detection: Long Short-Term Feature Extraction from Heart Sound Data.

Heart sounds have been extensively studied for heart disease diagnosis for several decades. Traditional machine learning algorithms applied in the literature have typically partitioned heart sounds into small windows and employed feature extraction methods to classify samples. However, as there is no optimal window length that can effectively represent the entire signal, windows may not provide a sufficient representation of the underlying data. To address this issue, this study proposes a novel approach that integrates window-based features with features extracted from the entire signal, thereby improving the overall accuracy of traditional machine learning algorithms. Specifically, feature extraction is carried out using two different time scales. Short-term features are computed from five-second fragments of heart sound instances, whereas long-term features are extracted from the entire signal. The long-term features are combined with the short-term features to create a feature pool known as long short-term features, which is then employed for classification. To evaluate the performance of the proposed method, various traditional machine learning algorithms with various models are applied to the PhysioNet/CinC Challenge 2016 dataset, which is a collection of diverse heart sound data. The experimental results demonstrate that the proposed feature extraction approach increases the accuracy of heart disease diagnosis by nearly 10%.

VARMA model parameterization using MLLE approach for intraday wind power forecasting application

AbstractAccurate intraday wind power forecasting (WPF) is required by system operators to mitigate challenges arising from small‐scale grid integration of intermittent wind power. This low‐level integration of wind energy into the system urges quality deterministic forecast. Multivariate statistical models considering spatiotemporal correlations are commonly used for such WPF. However, prudent parameterization is a necessity to improve deterministic forecasts even in smaller datasets. This article proposes a computationally efficient vector autoregressive moving average (VARMA) based intraday WPF model that uses the novel maximum log‐likelihood estimation (MLLE) approach to estimate the best‐fit model parameters. MLLE approach reduces the spatiotemporal fitting error by approximately 15% with computational time deduction by more than 25%. Furthermore, an advanced accuracy in parameter estimation (AIPE) technique supports the MLLE approach to obtain the optimal training window length for computing distribution parameters with narrow confidence intervals. Proposed centralized forecast model is implemented on a group of Wind Farms (WFs) in SE Australia. Pre‐processing of wind power generation data is ensured using the Yeo–Johnson transformation and stationarity concept. Proposed model is accurate and computationally efficient as compared to benchmark models in terms of accuracy and computational time. An appreciated forecast quality is achieved using the proposed model having only 0.20 MW and 0.29 MW mean normalized RMSE for 3 and 10 WFs studies, respectively. Possible applicability of the proposed model includes an overall improvement in intraday decision‐making of wind energy producers and system operators.

ICanClean Improves Independent Component Analysis of Mobile Brain Imaging with EEG.

Motion artifacts hinder source-level analysis of mobile electroencephalography (EEG) data using independent component analysis (ICA). iCanClean is a novel cleaning algorithm that uses reference noise recordings to remove noisy EEG subspaces, but it has not been formally tested in a parameter sweep. The goal of this study was to test iCanClean’s ability to improve the ICA decomposition of EEG data corrupted by walking motion artifacts. Our primary objective was to determine optimal settings and performance in a parameter sweep (varying the window length and r2 cleaning aggressiveness). High-density EEG was recorded with 120 + 120 (dual-layer) EEG electrodes in young adults, high-functioning older adults, and low-functioning older adults. EEG data were decomposed by ICA after basic preprocessing and iCanClean. Components well-localized as dipoles (residual variance < 15%) and with high brain probability (ICLabel > 50%) were marked as ‘good’. We determined iCanClean’s optimal window length and cleaning aggressiveness to be 4-s and r2 = 0.65 for our data. At these settings, iCanClean improved the average number of good components from 8.4 to 13.2 (+57%). Good performance could be maintained with reduced sets of noise channels (12.7, 12.2, and 12.0 good components for 64, 32, and 16 noise channels, respectively). Overall, iCanClean shows promise as an effective method to clean mobile EEG data.

Parameter optimization of moving horizon estimation based on hybrid intelligent algorithm

AbstractIn the application of moving horizon estimation (MHE) algorithm, the window length will affect the estimation accuracy and the computing efficiency. For this kind of problem, a method of parameter optimization is proposed to obtain suitable window length. Firstly, in order to facilitate online solution, the optimization problem involved in the algorithm is transformed into a quadratic programming (QP) problem in matrix form. Secondly, for the time index and the estimated residual index that measure different properties, the normalization idea is adopted to incorporate them into the same dimension to design the fitness function, and a genetic optimization algorithm based on simulated annealing mechanism is given to search for the optimal window length. Finally, the proposed parameter optimization method is verified by two cases. The results show that the parameter optimization method has the advantages of excellent local search ability and sufficient convergence, and the window length obtained by this method can better take into account the two performance indexes of the MHE algorithm and improve the estimation performance.

An adaptive lightweight seq2subseq model for non‐intrusive load monitoring

Non-intrusive load monitoring (NILM) is an important technology for deeply mining consumers' internal electricity consumption information, which can improve the level of awareness of the load and significantly improve the demand-side management capability of the smart grid. For a long time, traditional NILM faces the challenge of huge data, long training time and low accuracy. To address this issue, this paper proposes an adaptative lightweight seq2subseq model. The model adaptively determines the optimum window length of the active power segment to extract the appliance characteristics and thus reduces the redundant data to construct the appliance-level-seq2subseq model. The proposed model significantly improves the ability to extract and utilize information. At the same time, transfer learning is used to ban the training of appliance models with similar characteristics, further reducing model complexity and training time. The experimental results based on the REDD dataset verify the method's effectiveness. Compared with the seq2seq and seq2point models, the proposed model achieved better accuracy and F1 score, up to 98.10% and 90.29%, respectively. At the same time, the proposed method is further verified on UK-DALE, which proves that the proposed method has good portability and robustness.

On the Kurtosis of Modulation Formats for Characterizing the Nonlinear Fiber Propagation

Knowing only two high-order statistical moments of modulation symbols, often represented by the fourth moment called “kurtosis”, the overestimation of nonlinear interference (NLI) in a Gaussian noise (GN) model due to Gaussian signaling assumption can be corrected through an enhanced GN (EGN) model. However, in some modern optical communication systems where the transmitted modulation symbols are statistically correlated, such as in systems that use probabilistic constellation shaping (PCS) with finite-length sphere shaping, the kurtosis-based EGN model produces significant inaccuracies in analytical prediction of NLI. In this paper, we show that for correlated modulation symbols, the NLI can be more accurately estimated by substituting a statistical measure called windowed kurtosis into the EGN model, instead of the conventional kurtosis. Remarkably, the optimal window length for windowed kurtosis is found to be consistent with the self-phase modulation (SPM) and cross-phase modulation (XPM) characteristic times in various system configurations. The findings can be used in practice to analytically evaluate and design NLI-tolerant modulation formats.