Search Results (247)

Photoacoustic signal analysis in weakly absorbing media remains challenging because of low signal-to-noise ratios. This work proposes a deep learning framework for classifying particle size and concentration in an indirect absorption configuration. We conducted a comparative study using raw temporal signals, Savitzky–Golay filtering, and time–frequency scalograms via Continuous Wavelet Transform (CWT), and evaluated both 1D and 2D convolutional neural network architectures. Experimental validation was performed using poly(methyl methacrylate) (PMMA) microspheres (6 μm and 15 μm) and hydroxyapatite nanoparticles (<200 nm) at volume fractions as low as $6\times {10}^{-4}\%$. While raw signals led to unstable training (accuracy ≈ 47%), CWT-based representations significantly improved performance, achieving near-perfect size discrimination and over 96% accuracy in discrete volume-fraction classification. Grad-CAM analysis confirmed that the model identifies physically meaningful regions of the acoustic waveform, ensuring interpretability. The proposed framework was validated under controlled experimental conditions using discrete particle types and predefined volume-fraction classes, providing a foundation for future extensions toward continuous particle characterization. Ultimately, these findings demonstrate that combining time–frequency representations with deep learning provides a robust, physically consistent approach for particle characterization in turbid media, with significant potential for biomedical diagnostics and material analysis. Full article

Long-duration flood inundation can substantially suppress crop growth and cause yield loss, particularly in semi-arid agricultural regions increasingly affected by extreme rainfall. Timely crop damage assessment is critical for disaster response and insurance-related decision-making, but direct yield-loss observations are often unavailable during or shortly after flooding. This study proposes a phenology-guided regression framework for early crop damage assessment using multi-source SAR–optical observations. The study was conducted on the Tumochuan Plateau, Inner Mongolia, China, where severe rainfall beginning on 23 July 2025 caused widespread cropland inundation. Sentinel-2 EVI time series from 2022 to 2025 were fitted using a Savitzky–Golay (SG) filter, and annual area under the EVI curve (AUC) loss in 2025 relative to the 2022–2024 historical mean was used as a proxy for flood-induced crop damage. Optical features from Landsat-8/9 and Sentinel-2, together with SAR backscatter features from Sentinel-1, Lutan-1, and Gaofen-3, were incorporated into machine learning regression models. SAR features improved pixel-wise prediction, with the Random Forest model achieving the highest $R^2$ of 0.62 using early-period features and 0.77 using later-period features. Village-scale aggregation further improved performance, yielding an early-period $R^2$ of 0.84 across 123 and 0.78 across 122 villages. These results demonstrate the feasibility of SAR–optical and phenology-guided regression for early crop damage assessment under long-duration inundation. Full article

Accurate voltage degradation prediction is essential for health management and lifetime extension of proton exchange membrane fuel cell (PEMFC) systems. During long-term constant-current operation, stack voltage evolves nonlinearly and is influenced by coupled variations in temperature, pressure, flow rate, and humidity, while many deep learning-based models lack physical interpretability. This study proposes an interpretable Transformer-based framework for PEMFC voltage degradation prediction under constant-current operation. The framework integrates outlier correction, interpolation, Savitzky–Golay filtering, Z-score normalization, sliding-window reconstruction, Transformer-based prediction, and feature-ablation interpretation. Using multivariate sensor measurements and historical voltage as inputs, the Transformer was compared with RNN, LSTM, and GRU baselines under identical preprocessing and evaluation conditions. The models were evaluated chronologically by continuously applying the sliding-window model over the held-out final 20% of the aging sequence. The Transformer achieved the best performance, with MAE of 8.2 × 10⁻⁴, RMSE of 1.18 × 10⁻³, MAPE of 0.0256%, and R² of 0.9961. Compared with the second-best RNN model, it reduced MAE, RMSE, and MAPE by 9.89%, 7.81%, and 9.86%, respectively. Feature ablation showed that flow- and pressure-related variables contributed 50.08% and 26.78% of the total importance, respectively. Full article

Anisotropic polymers fabricated via additive manufacturing exhibit complex thermal transport profiles that are challenging to characterize using steady-state techniques. We present a transient thermal method utilizing a short rectangular current pulse excitation to determine the directional thermal diffusivity and conductivity of anisotropic materials. The measurement is conducted on finite specimens, where the low diffusivity of the polymer media results in a highly attenuated and dispersed rear-side temperature profile over an extended transient window. Conduction losses to the adjacent coolers are accounted for by solving the one-dimensional heat conduction equation on an asymmetric multi-layer sandwich structure using the implicit Crank–Nicolson method. Because thermal diffusivity and conductivity are not independent quantities ($\lambda =a\rho c$), the inverse problem is deliberately formulated to estimate the diffusivity alone: the volumetric heat capacity is predetermined and held fixed, and the conductivity follows directly as $\lambda =a\rho c$. This removes the ill-conditioning that would otherwise arise from treating $\lambda$ and a as free, independent parameters in the fit. A two-parameter non-linear least-squares fit is applied to the rear-side temperature rise following Savitzky–Golay noise filtering to estimate the directional diffusivity and effective heat flux. The method is validated using an isotropic reference standard to rule out false system anisotropy, and is subsequently applied to additively manufactured polymer specimens to resolve print-induced directionality through sequential, axis-aligned (unidirectional) measurements along the axial and transverse printing directions. The validity of the one-dimensional reduction is confirmed quantitatively by two- and three-dimensional anisotropic simulations of the exact geometry, which bound the lateral-spreading bias below $0.01\%$ even for the highest-anisotropy specimen, and the robustness of the method to sensor thermal response, signal filtering, and effective-flux estimation is quantified. A rigorous evaluation of the expanded metrological uncertainty demonstrates the high accuracy and reliability of this low-energy excitation technique for highly dispersing media, making it a viable and highly accessible alternative for evaluating material anisotropy. Full article

As the application scale of battery energy storage gradually increases, the accurate prediction of the remaining service life of large-capacity energy storage batteries is crucial for high-quality development in this field. To address the issues of insufficient reliability and poor generalization in large-capacity energy storage battery life prediction, a deep learning framework based on a long short-term memory (LSTM) neural network is developed. Early aging data from the first 150 cycles is used for the model, with outliers removed and noise reduced through Savitzky–Golay (SG) filtering. Data normalization and a sliding window method are employed for training. The model is validated on two batches of large-capacity batteries under GB/T 36276-2023 conditions at 25 °C and 45 °C, achieving the root mean square errors (RMSEs) of 0.86% and 0.50%, respectively, over 1000 cycles. Additionally, the method is tested on small-capacity batteries from an MIT dataset, achieving an RMSE of 4.3%. A transfer learning module fine-tunes the model using cycles 151–300, reducing RMSEs to 0.18%, 0.10%, and 3.1% for the three battery sets. This enhances the model’s generalization and offers a practical solution for life prediction in battery inspection and evaluation. Full article

Urbanization alters the local growth environment of vegetation, including local thermal regimes, moisture availability, and human activity intensity. Yet, the gradient-dependent changes in vegetation phenology and the factors responsible for these changes are still not fully clarified. Taking Beijing as the study area, this study used MOD13Q1 EVI records from 2005 to 2024 to reconstruct annual vegetation trajectories with Savitzky–Golay filtering. The start of the growing season (SOS) and the end of the growing season (EOS) were then identified based on a dynamic threshold criterion, and the length of the growing season (LOS) was derived from the interval between the two phenological dates. Urban–rural gradients were constructed by integrating Global Urban Boundary (GUB) and digital elevation model (DEM) data. Theil–Sen slope estimation and the Mann–Kendall test were then applied to analyze trends in vegetation phenology, while partial correlation analysis and the XGBoost-SHAP method were used to identify the relative effects of climatic and urbanization-related factors. The results showed that (1) vegetation phenology in Beijing varied markedly across space. Mountainous areas in the northwest and southwest had a later SOS, earlier EOS, and shorter LOS, whereas the central plain and southeastern regions were characterized by an earlier SOS, later EOS, and longer LOS; (2) from 2005 to 2024, the SOS advanced significantly by −0.65 days/year, EOS showed a weaker delay of 0.46 days/year, and LOS increased significantly by 0.90 days/year, indicating an overall extension of the vegetation growing season over the past two decades. (3) Urban areas had an earlier SOS, later EOS, and longer LOS, but the trends of SOS advancement, EOS delay, and LOS extension were more pronounced in suburban and rural areas; and (4) the SOS was mainly influenced by the combined effects of nighttime lights and heat-related factors, whereas the EOS was primarily affected by temperature and nighttime lights. Along the urban–rural gradient, the importance of nighttime lights gradually decreased from urban areas to suburban and rural areas, while the role of climatic factors became relatively stronger. These findings reveal the gradient differentiation of vegetation phenological status, temporal trends, and associated drivers in Beijing under rapid urbanization, and provide useful information for urban vegetation management and ecological planning. Full article

This study investigates, in a controlled setting, a hybrid neural–mechanistic correction framework applied to a parameter-distorted or biased epidemiological SIR model. The aim was to explore to what extent a residual neural component can compensate for mechanistic model error, how strongly reconstruction quality depends on Savitzky–Golay-based derivative estimation, how robust the framework remains under increasing observational noise, and how the coupling parameter $\lambda$ should be selected. The distorted mechanistic SIR model was coupled with a feedforward neural network trained on residual dynamics estimated from noisy synthetic observations. Savitzky–Golay smoothing was evaluated on a parameter grid, while $\lambda$ was determined by continuous numerical optimization over the interval $\left[0, 1\right]$. The experiments were repeated using multiple random seeds and examined at three different noise levels. The hybrid model consistently improved reconstruction compared to the distorted mechanistic baseline across the entire investigated noise range. The results indicate that performance depended on the preprocessing applied for derivative estimation. The optimized coupling parameter remained in the high range in all cases, suggesting that substantial neural correction is advantageous in the presence of parameter bias. The results show that residual neural correction can be successfully transferred to an epidemiological SIR setting. Full article

Rate of penetration (ROP) is a key indicator of drilling efficiency, governed by nonlinear coupling among mechanical, hydraulic, drilling fluid, and formation factors. This study develops an ExtraTrees model optimized by an improved Harris hawks optimization algorithm (IHHO-ET) using field while drilling data from Well Z in the Tarim Oilfield. A preprocessing workflow involving drilling section identification, abnormal condition filtering, 3 × IQR outlier removal, Savitzky–Golay smoothing, and standardization is combined with correlation and gray relational analysis under engineering mechanism constraints to select 14 input features. Logistic chaotic initialization, adaptive Gaussian mutation, and dynamic weighting are introduced into HHO, with validation set RMSE as the fitness function. To reduce the influence of random splitting and initialization, all comparison models are evaluated with repeated seeds and validation set tuning. Using R² and RMSE as primary criteria, IHHO-ET achieves R² = 0.910 ± 0.004 and RMSE = 0.871 ± 0.019 on the same-well test set. Its improvement over HHO-ET is small and not significant (p = 0.109), indicating that the IHHO strategies mainly refine search stability. Same-region leave-one-well-out validation gives an average R² = 0.696, suggesting that the model suits same-region trend prediction rather than direct closed-loop control. The proposed workflow provides a practical reference for ROP prediction. Full article

Extreme and sudden temperature fluctuations observed as a result of global climate change increase the environmental pressure on energy transmission infrastructure. These meteorological changes significantly increase the risk of failure for porcelain insulators, which exhibit low thermal resistance and are susceptible to sudden arcing and surface deformations. In this study, a hybrid CNN–BiLSTM–SE architecture augmented with the Squeeze-and-Excitation attention mechanism is proposed using surface leakage current signals to diagnose healthy, cracked, and broken structural conditions in three-unit porcelain insulators. The SE block in the architecture dynamically rescales feature maps from CNN layers on a channel-by-channel basis. Thus, it highlights the signal characteristic that is dominant for fault diagnosis just before the BiLSTM units learn temporal dependencies. Leakage current data were obtained under an experimental setup at 60 kV for 15 different conditions covering all possible combinations of healthy, cracked, and broken insulator units. The raw signals were preprocessed with the Savitzky–Golay filter to suppress noise while preserving the diagnostic waveform morphology. 24 features covering time-domain statistics, frequency-domain spectral characteristics, and wavelet-domain energy components were extracted and used as model inputs. The CNN–BiLSTM–SE architecture achieved a classification accuracy of 93.83%, surpassing the standalone CNN (88.89%), BiLSTM (87.65%), and CNN–BiLSTM (91.36%) models, as well as classical machine-learning baselines (SVM: 87.65%, Random Forest: 90.12%, Boosted Trees: 87.65%). Full article

Accurate remaining useful life (RUL) prediction for aero-engines is important for condition-based maintenance and safety-oriented health management. Long short-term memory (LSTM) networks are widely used for this task, but two limitations remain important in multi-sensor degradation modeling: hidden states generated over the full window are often under-utilized, and attention mechanisms may overemphasize locally fluctuating sensor readings. This paper proposes a Multi-Statistic Disentangled LSTM (MSD-LSTM) framework for aero-engine RUL prediction. The framework first applies Savitzky–Golay filtering to smooth high-frequency signal fluctuations. A hidden-state feature extraction module then combines feature-level disentangled extraction and Global Average Pooling to use the LSTM hidden-state sequence beyond the final recurrent output. In parallel, a Multi-Statistic Pooler summarizes each input window using minimum, maximum, standard deviation, and mean statistics, and its output is fused with a self-attention branch through a static-gating mechanism. On the NASA C-MAPSS benchmark, MSD-LSTM achieves RMSE values of 10.45 and 12.33 on FD001 and FD002, respectively, and ranks first in RMSE on three of the four sub-datasets and first in SCORE on two sub-datasets among the compared recent methods. Ablation and fusion analyses show that both the hidden-state extraction and statistic-guided fusion components contribute to stable RUL prediction. Full article

Reliable and rapid forecasting of wind turbine output power is vital for operators, particularly day-ahead and intraday market scheduling and reserve allocation. However, the inherent unpredictability, intermittency, and volatility of wind turbine output make forecasting processes difficult. To address this challenge, this study proposes a Sinusoidal Representation Network (SIREN)-based forecasting model for high-accuracy, rapid direct multi-horizon forecasting of wind turbine output power. SIREN is selected due to the periodic and symmetrical mathematical structure of its sinusoidal activation function, which allows the model to represent both low-frequency trends and high-frequency sudden changes in wind energy data. To improve data quality, compensate for asymmetric fluctuations in wind data, and provide more suitable inputs for SIREN training. Several preprocessing steps are utilized before feeding the data into the model. The proposed preprocessing step includes a moving median filter, robust scaling based on median and interquartile range, Winsorizing clipping, and a Hampel filter to reduce the effects of instantaneous noise, outliers, and local peaks without disrupting temporal continuity. Subsequently, a Savitzky–Golay smoothing is applied to attenuate high-frequency measurement noise while preserving curvature, local peaks, and physically meaningful short-term dynamics in the data. The sliding-window approach is used to formulate the multi-horizon forecasting problem directly, and a direct h-step-ahead forecasting architecture is designed, preserving structural symmetry in the time series. The SIREN is trained and tested using MATLAB with the help of two different datasets: Dataset-1 has a 10 min resolution for 1 year, and Dataset-2 has a 1 h resolution for 15 years. The forecast horizon parameter h is considered separately for each step, and the proposed SIREN is independently trained, validated, and tested for each target horizon while maintaining chronological order. The results demonstrate that the proposed model is able to yield high forecast performance for a wide spectrum of horizons ranging from 10 min to 15 days. The accuracy of the proposed model for Dataset-1 is R² of 99.6%, MSE of 0.085%, MAE of 1.7%, and MAPE of 12%, while for Dataset-2, the accuracy is R² of 98.8%, MSE of 0.3%, MAE of 3.6%, and MAPE of 23%. Ablation and sensitivity analyses are conducted to evaluate the impact of the basic components used in the proposed model on forecasting performance. In addition, combative experiments are performed using traditional time series, ML, and DL forecasting techniques to better assess the contribution of the model. The obtained results show that the SIREN-based direct forecasting approach provides strong learning capability, as well as high forecasting accuracy, for both high-resolution and low-resolution wind power data. Overall, its ability to capture the symmetric and periodic characteristics inherent in wind turbine power data makes it a promising alternative for multi-horizon wind power forecasting applications. Full article

Raw sentiment extracted from Chinese financial news is noisy and difficult to use directly for market prediction. This study proposes a mathematical filtering framework that converts noisy Chinese financial news sentiment into reliable quantitative signals for financial market prediction. Three daily sentiment measures were constructed from Chinese financial news: sentiment mean, sentiment dispersion, and polarity imbalance. Seven filtering methods were applied to each measure, including exponential smoothing, autoregressive filtering, ARIMA filtering, moving average smoothing, discrete wavelet transform, Savitzky–Golay filtering, and Kalman filtering. The seven filtered outputs were averaged to produce an ensemble-smoothed sentiment signal. Support vector machines and neural networks were then used to compare the predictive performance of raw and filtered signals for stock index log returns and realized volatility. Filtering reduced the standard deviation of sentiment mean by 48%, sentiment dispersion by 55%, and polarity imbalance by 50%, while mean levels remained stable. Filtered sentiment consistently outperformed raw sentiment across all model configurations. The improvement was larger for realized volatility than for returns: the best support vector machine reduced volatility prediction error by 16.9% and return prediction error by 5.8%. A moderate neural network with 20 hidden neurons achieved optimal performance for both outcomes. Mathematical filtering extracts stable and informative sentiment signals from Chinese financial news. Filtered sentiment is more useful than raw sentiment for predicting market volatility, and the improvement holds across multiple machine learning models. Full article

To address the key challenges in GB-RAR monitoring of super-tall buildings—namely, complex noise interference (transient pulse disturbances coupled with high-frequency random fluctuations), the difficulty of distinguishing normal wind-induced vibrations from hazardous deformations, and the propensity of single-algorithm prediction models to converge prematurely—this paper proposes an integrated monitoring data processing workflow that combines status assessment and deformation early warning, using Wuhan Greenland Center as a case study. A denoising method combining Median Absolute Deviation outlier removal and Savitzky–Golay filtering was designed for preprocessing, quantitatively validated through signal-to-noise ratio analysis. Based on filtered data, a spatio-temporal trajectory model was established to visualize and evaluate building movement. Furthermore, a GB-RAR-oriented residual-driven warning framework was developed by coupling a PSO-GA-BP deformation prediction model with adaptive sliding-window thresholding and finite-state warning decisions. Simulation results demonstrate that the PSO-GA-BP model outperforms other neural network models in prediction accuracy, and the derived early warning system exhibits strong feasibility and sensitivity. This workflow proves suitable for GB-RAR deformation monitoring of super-tall buildings, offering valuable reference for future research. Full article

Accurate apple origin identification and non-destructive internal quality evaluation are important for fruit traceability, quality grading, and post-harvest management. Unlike previous studies mainly focusing on origin classification, this study established a dual-task near-infrared spectroscopy framework integrating geographical origin classification and soluble solid content (SSC, °Brix) prediction for Fuji apples. Samples were collected from three representative production regions in China: Alar in Xinjiang, Yantai in Shandong, and Luochuan in Shaanxi. Near-infrared diffuse reflectance spectra were acquired from 375 apples, generating 3000 spectral samples for origin classification and 750 SSC-calibrated samples for sugar content prediction. For classification, six deep learning models were evaluated using standardized full-spectrum input without chemometric spectral preprocessing, and the Transformer achieved the best performance, with a test accuracy of 96.22%. For SSC regression, spectra were preprocessed using standard normal variate and Savitzky–Golay filtering. The DNN model achieved the best prediction performance, with MAE = 0.5958 °Brix, RMSE = 0.7333 °Brix, R² = 0.8646, and Pearson r = 0.9338. These results indicate that near-infrared spectroscopy combined with deep learning can support both Fuji apple origin authentication and non-destructive local tissue SSC assessment. Full article

Wild carrot (Daucus carota subsp. carota), the wild relative of cultivated carrot, is globally identified as an invasive weed that threatens hybrid carrot seed production through natural cross-pollination, resulting in compromised genetic purity. Manual identification across the large areas required to ensure genetic purity in carrot seed crops is impractical. Remote sensing offers an alternative; however, morphological similarities among wild carrot, cultivated carrot, and common weeds hinder reliable detection. Early identification, however, remains essential for preventing genetic contamination. This study evaluated leaf-level hyperspectral reflectance spectroscopy (400–2450 nm) with machine learning to discriminate wild carrot from hybrid carrots, parental lines, and 19 co-occurring weed species. Spectral data from 266 wild carrot plants across three New Zealand sites and six weeks (5–10 weeks after emergence) showed negligible spatial effects (R² = 0.034–0.055, pseudo-F = 1.46–2.39, p > 0.05) and moderate temporal variation (R² = 0.136–0.151, pseudo-F = 5.48–6.17, p < 0.001), indicating broadly stable spectral signatures suitable for model generalization. Savitzky–Golay filtering, with min–max normalization outperformed SNV, yielding high full-spectrum accuracies for wild carrot vs. other species (90.35%, κ = 0.80), wild carrot vs. weeds (96.03%, κ = 0.92), and a multi-class model (90.79%, κ = 0.88). After removing atmospheric water-absorption bands to follow airborne sensing, reduced-band models based on airborne-compatible wavelengths maintained strong performance, including 89.40% accuracy (κ = 0.79) for wild carrot vs. weeds using a 20-band Subspace Discriminant model (400–402, 527, 705–720 nm). These findings demonstrate that stable wild carrot spectra and carefully selected visible and red-edge bands can underpin cost-effective UAV/UGV-mounted hyperspectral or multispectral sensors for site-specific wild carrot management. Full article