Search Results (1,512)

Urban micro-meteorological models often omit building surfaces’ evening heat release. We derive physics-informed thermal boundary constraints from the Surface Energy Balance equation, supplied to a Physics-Informed Neural Network (PINN) through an Urban Thermal-Geometric Database and a Force-Restore model. A 2D heated cavity benchmark at $Ra={10}^4$, verified against an independent finite-volume reference, reduces interior temperature MAE by 56%, wall-temperature MAE by 97.5% and Nusselt error by 74% across five seeds, at a 16% horizontal-velocity cost, worse in all five. In the open-roof canyon, a five-seed PINN feasibility comparison shows material-resolved wall temperatures reduce blind temperature MAE from $0.00830\pm 0.00050$ to $0.00368\pm 0.00058$, with much lower matched heat-flux errors than an equivalent-memory homogeneous control and velocity within seed scatter. Seven independent quasi-steady snapshots follow. At 14:00, material resolution raises street-surface heat transfer by 61% while corridor-band velocity variability remains comparable, indicating redistribution rather than amplification. At 18:00 the material-resolved case retains an order-of-magnitude larger ${\sigma }_v$ (0.61 vs. 0.06); at 20:00 it reverses the street-surface heat-flux direction while the mean-temperature case approaches neutrality. It is a constructed near-zero-forcing baseline; ratios to it are not interpreted. The reversal is robust in sign but not in magnitude under slab-model and $\pm 1$ K facade-contrast perturbations. Full article

Artificial neural networks (ANNs) offer efficient surrogates for radio-frequency (RF) passive-device modeling and design, but prediction accuracy alone does not establish their engineering reliability. This review examines four overlapping analytical perspectives: data generation and sampling, forward modeling, knowledge-enhanced modeling, and inverse modeling and design. Its passive-device-focused synthesis separates computational costs by workflow stage, distinguishes physical-property guarantees from enforcement and tests, and examines the feasibility and coverage of inverse-design solutions. Direct passive-device applications are distinguished from transferable electromagnetic methods. Across the reviewed workflows, adaptive sampling reduced the need for costly labeling, while structured and hybrid representations improved response approximation within the tested domains. Multivalued and generative inverse models accommodated non-unique solutions, although candidate diversity should be assessed separately from feasibility and solution coverage. Coarse models, transfer functions, and physics-informed losses provide useful structural priors but do not automatically guarantee network properties or reliable extrapolation. The experimental results demonstrated the practical potential for the tested devices and processes. These findings call for reproducible benchmarks and process-aware validation that link modeling accuracy to physical reliability and engineering applicability. Full article

Hyperspectral unmixing aims to recover endmembers and their corresponding abundances from mixed pixels, which is a key task in quantitative hyperspectral remote sensing. Although convolutional neural networks (CNNs) are effective in modeling local spatial–spectral features, their ability to capture long-range spatial dependencies remains limited. Transformers can enhance global feature interactions but typically incur high computational complexity. To address these issues, this study proposes an abundance-guided Mamba network for hyperspectral unmixing, termed AGMNet. Built upon an interpretable autoencoder-based unmixing framework, the proposed method uses CNNs to extract local spatial–spectral features and introduces a lightweight spatial Mamba branch to incorporate whole-image contextual information. Meanwhile, an abundance estimation mechanism with physical constraints and a linear decoder consistent with the linear mixing model are employed to achieve physically interpretable endmember-abundance decomposition. Experiments on three public datasets, namely Samson, Jasper Ridge, and Urban, show that AGMNet provides accurate and stable abundance recovery, achieving the lowest average root mean square error (RMSE) across all three datasets. Its spectral angle distance (SAD) performance is more dataset-dependent: AGMNet obtains the best average SAD on Urban and competitive SAD values on Samson and Jasper Ridge, but it does not consistently achieve the lowest SAD for every endmember. These results indicate that AGMNet is particularly effective for abundance recovery while retaining physically interpretable spectral reconstruction, although further improvement of endmember spectral-shape consistency remains necessary. Full article

In strong-field ionization, complex-time saddle points are closely related to electron emission times and quantum-orbit interference patterns. However, not all solutions to the saddle-point equation carry physical contributions. Saddle coalescence can also invalidate the isolated-saddle approximation. Within the strong-field-approximation framework, we propose a topology-constrained approach to reconstruct photoelectron momentum spectra. A physics-informed neural network is employed to produce candidate roots, which are subsequently refined via a Newton-type iteration. A topological selector is defined by the non-zero support of intersection numbers to filter unphysical solutions, while local coherent continuation for pole-saddle pairs handles saddle-coalescence regions. We test our method on hydrogen atoms subjected to six-cycle elliptically polarized laser fields and compare the reconstructed spectra with direct numerical strong-field-approximation results. For 300 nm and 500 nm driving fields, the normalized relative $L_2$ errors on the linear probability scale reach $3.793\%$ and $3.520\%$, respectively. Our method improves the physical interpretability of saddle-based spectrum reconstruction within the strong-field approximation and is promising for other problems in computational physics involving saddle-point equations. Full article

The impact resistance of thin PMMA plates in deep-sea transparent structures is important for crashworthiness safety and is governed by high-strain-rate loading and hydrostatic pressure. Conventional empirical constitutive models exhibit limited adaptability to strain ratio effects and complex energy dissipation, leading to significant deviations in anti-impact assessment. This study proposes a constitutive-aware physics-informed neural network framework that integrates a data-driven constitutive neural network with embedded physical constraints. Such a proposed constitutive network replaces the theoretical model to capture rate-dependent stress response, while a physics-informed network enforces momentum conservation, energy balance, thermodynamic dissipation inequality, and pressure-induced stress limits. A dynamically weighted multimodal loss function is proposed and is used to ensure stable constraint enforcement during data training. Experimental validation based on a high-speed impact test confirms the constitutive and physics constraint network accuracy. Numerical results demonstrate that increasing hydrostatic pressure alters energy dissipation, increasing residual projectile velocities, and degrading impact resistance. The framework provides an efficient, physically consistent method for evaluating anti-impact performance of thin-walled PMMA structures in deep-sea engineering. Full article

Distribution networks are operating under increasingly variable conditions as renewable generation and electric vehicles become more widespread. Maintaining reliable and secure operation in this setting requires timely information about the electrical state of the network, yet DSSE remains difficult to implement when measurement coverage is sparse, communication infrastructure is limited, and feeder configurations are complex enough to compromise observability. This review examines how current DSSE methods deal with these conditions, including high R/X ratios, three-phase unbalance, and limited instrumentation. More than one hundred studies are considered, spanning optimization-based estimators and related mean-squared-error formulations, Kalman-filter-based dynamic approaches, artificial intelligence techniques, and centralized and distributed estimation schemes. Their reported advantages, limitations, and application areas are assessed in the context of practical deployment requirements. The review also considers directions that combine physical network models with neural networks, improve pseudo-measurement generation, account for dynamic and stochastic behavior, and distribute the estimation process through approaches such as Bayesian consensus filtering and the alternating direction method of multipliers. Comparative tables summarize the main characteristics of these methods, while the discussion gives particular attention to real-time implementation, cybersecurity, and operational constraints encountered in modern distribution networks. Full article

Physics-informed neural networks usually incorporate differential equations primarily through the loss function, while the neural architecture itself is selected independently of the underlying numerical structure. This work develops a scheme-induced approach in which an analytically modified numerical method defines the computational architecture, while selected elements of the numerical scheme are made trainable and refined using physics-informed loss functions. The specific contribution is the formulation and systematic investigation of a class of scheme-parameterized low-fidelity physics-informed neural networks (LF-PINNs) for evolutionary PDEs. In the implementation studied here, a compact parameter network directly controls the coefficient of a generalized $\theta$-scheme, while the implicit update is represented by a fixed number of Picard iterations. The approach is evaluated on six one-dimensional PDE benchmark configurations against conventional, hard-constrained, and adaptively weighted PINNs. The nine-parameter LF-PINN achieves competitive accuracy on several benchmark problems, while its trainable parameter remains explicitly connected to a coefficient of the underlying numerical scheme. The results also show that performance is problem-dependent and that a small number of trainable parameters does not necessarily imply lower computational cost. Stability, ablation, computational-cost, and generalization analyses further characterize the properties and limitations of the scheme-induced construction. Full article

The vertical wind profile within the Atmospheric Boundary Layer (ABL) is a critical variable for a wide range of applications, including wind energy, air traffic and Unmanned Aerial System (UAS) management, pollutant dispersion, and urban air mobility. However, its accurate characterization typically requires expensive vertical-profiling LiDAR instruments, which limits the number of locations that can be simultaneously monitored. Building upon previous work in which machine-learning algorithms were trained on LiDAR data from the University of León (Spain) to reconstruct full ABL wind profiles from a single near-surface measurement, this study replaces the traditional data-driven neural network approach with Physics-Informed Neural Networks (PINNs), embedding a physical constraint directly into the loss function as evaluated by automatic differentiation, and tests two such constraints under architectural control: the Ekman boundary-layer momentum balance, closed with a Monin–Obukhov eddy viscosity, and a shear-constrained formulation based on a power-law profile with an inferred exponent. With dense supervision at the target heights, neither constraint improves on an architecture-matched network without a physical term (RMSE 1.35–1.36 m/s, direction MAE 20.1–20.4${}^{\circ }$ across all three, Kruskal–Wallis $p=0.0125$). In a vertical-extrapolation protocol, training with gates up to 111 m, selecting models by their skill at 140 m, and testing blind on 170–300 m, the Ekman constraint shows no detectable advantage over its own architectural control. In contrast, the shear-constrained model does, reaching $1.80\pm 0.01$ m/s speed RMSE and $27.0^{\circ }\pm 0.3^{\circ }$ direction MAE against $2.66\pm 0.10$ m/s and $34.2^{\circ }\pm 1.8^{\circ }$ for the Ekman-constrained arm and $2.61\pm 0.12$ m/s and $33.9^{\circ }\pm 2.0^{\circ }$ for the physics-free arm, which are statistically indistinguishable from one another; the shear-constrained gain is confirmed against its own architectural control ($2.64\pm 0.08$ m/s, $33.0^{\circ }\pm 1.8^{\circ }$, matching the physics-free arm), ruling out the additional output as the explanation, and converts an erratic extrapolator into a stable one. The results are consistent with the benefit of embedding physics into the loss being therefore specific to the extrapolation regime and to the formulation tested, not a generic property of physics-informed learning. These findings are based on data from a single suburban site, and multi-site validation remains necessary before they can be generalized. By eliminating the need for continuous LiDAR operation after the training phase, the proposed methodology aims to enable cost-effective monitoring of multiple locations with a single instrument. Full article

The fusion of multi-source remote sensing data and deep learning technology has been increasingly applied in precision agriculture. This review focuses on the perception–decision–execution framework and summarizes the current application status, technical routes, and development trends of multi-source remote sensing data fusion and deep learning in precision agriculture. It examines the complementary advantages of multimodal data such as optical, thermal infrared (TIR), Light Detection and Ranging (LiDAR), and synthetic aperture radar (SAR) in characterizing crop physiological traits and structural characteristics. At the algorithmic level, this review analyzes the evolution of deep learning architectures from single convolutional neural networks (CNNs) to spatiotemporal fusion networks (STFNs), Transformer-based architectures, and emerging state-space models, with emphasis on their capabilities for heterogeneous feature extraction and cross-modal information integration. This review discusses the transformation mechanism from intelligent perception to executable decision-making and explores the application effectiveness of prescription maps and variable-rate task technology in improving resource utilization. This review proposes forward-looking views on core bottlenecks such as model generalization, physical interpretability, and edge computing. This review not only provides a theoretical analysis of precision agriculture but also provides academic recommendations for the development of intelligent agricultural equipment in the future. Full article

Accurate internal temperature estimation is of great significance to the safe and dependable working performance of lithium-ion batteries (LIBs). Nevertheless, conventional purely data-driven prediction models suffer from inherent limitations, including insufficient physical interpretability and degraded generalization capability under complex dynamic driving conditions. To tackle this challenge, the present study puts forward a physics-informed temporal convolutional network and long short-term memory (PI-TCN-LSTM) framework for high-precision LIB internal temperature estimation, which depends solely on readily obtainable external sensor measurements. The proposed hybrid architecture synergistically integrates the strengths of two temporal modules: TCN captures fast transient thermal dynamic characteristics, while LSTM effectively mines long-term thermal inertia dependencies of batteries. To make up for the absence of physical constraints in pure data-driven methods, a two-state lumped thermal model is adopted to characterize battery internal heat generation and heat transfer mechanisms. Specifically, the governing thermodynamic differential equations covering irreversible Joule heat, reversible entropic heat, and convective heat dissipation are directly embedded into the network loss function as physical constraints, achieving tight coupling of domain physical laws and data-driven learning. Comprehensive experimental validations on commercial NMC811 batteries demonstrate that the proposed PI-TCN-LSTM framework substantially outperforms conventional pure data-driven models. Notably, the developed approach demonstrates favorable generalization to the unseen FTP-75 driving cycle over a wide temperature range, maintaining RMSE values of 0.081 °C and 0.069 °C at 5 °C and 45 °C, respectively. Full article

Numerical prediction of high-order nonlinear oscillators using conventional physics-informed neural networks (PINNs) is severely hampered by spectral bias, which prioritizes low-frequency components and leads to significant errors in capturing multi-harmonic vibration responses. This work is a numerical-method-oriented investigation rather than an experimental MEMS validation study. To mitigate spectral-bias issues, we employ a Fourier multiscale physics-informed neural network (FMPINN), which integrates random Fourier feature mapping with a parallel multi-sub-network architecture. We note that the individual building blocks of FMPINN are existing components in scientific-machine-learning literature, and our core contribution is not architectural innovation. While this enhancement yields substantial accuracy improvements—reducing relative L₂ errors by one to two orders of magnitude under weak-to-moderate nonlinearity—the central contribution of this work lies in the quantitative characterization of its applicability boundary. Through six systematically designed numerical cases of a fifth-order Duffing oscillator reduced-order model motivated by MEMS resonator dynamics, we demonstrate that the performance of FMPINN saturates under strong nonlinear excitation. Contrary to intuition, simply increasing the Fourier-feature dimension cannot resolve this saturation. Spectral analysis reveals that this bottleneck stems from the initialization-determined finite effective bandwidth of the fixed, frozen frequency mapping, which induces spectral-folding-dominated artifacts in the presence of dense high-order harmonics. This provides a plausible mechanistic interpretation supported by our numerical experiments, rather than a mathematically proven fundamental bound. Our findings establish a practically useful numerical benchmark for the nonlinear stiffness threshold of Fourier-augmented PINN frameworks and underscore the need for adaptive, rather than fixed-scale, frequency embedding strategies. All evaluations are conducted against synthetic noise-free Runge–Kutta 45 benchmarks; experimental MEMS measurement data are not used in the present work. Full article

Memory-dependent evolution equations replace the time derivative of the state by a convolution with a memory kernel $K\left(t\right)$. They are used in anomalous diffusion, viscoelasticity and tumour-growth modelling, usually with K fixed in advance, most often as the power-law kernel of a Caputo derivative. We study the joint recovery of K and the state u with a physics-informed neural network (PINN) trained by a block-coordinate Variable Projection (VarPro) scheme, for four kernel parametrizations: a non-negative Prony series, a fractional (Caputo) power law, a stretched exponential and a shape-constrained spline. In the best-performing runs on a synthetic benchmark, the state is recovered with relative errors below $0.05\%$ for all four families, while kernel recovery depends strongly on the family. The fractional kernel is recovered almost exactly, the stretched exponential to $5.7\%$, and the Prony series with a kernel-function reconstruction error of 2–$4\%$ in its best runs. A local sensitivity analysis at the exact solution, which requires no training, shows that the individual Prony parameters are severely ill conditioned: the column-normalized condition number of the sensitivity matrix is more than two orders of magnitude larger when the true decay rates are a factor of four apart than when they are well separated. Fitted against the exact state, without a network, the kernel and the three active Prony modes are recovered to a rounding error, while the redundant zero-weight modes of the over-parametrized representation remain non-identifiable. A perturbation of only $0.1\%$ raises the active-mode parameter error to tens of percent while the kernel error stays near $0.1\%$, and this error tracks the condition number. A Gauss–Newton/Laplace approximation and an eight-replicate training ensemble agree that the rates are far less well determined than $K\left(t\right)$. A spline that assumes only positivity, monotonicity and convexity reaches a kernel error of about $6\%$. The results show that an accurate state does not guarantee accurate recovery of the kernel or of its parameters, and that state accuracy, kernel-function accuracy and parameter identifiability should be assessed separately. Full article

False data injection attacks (FDIAs) pose a serious threat to FACTS-assisted wide-area damping controllers because they corrupt remote measurements, degrade inter-area oscillation damping, and may evade conventional detection. Data-driven detectors that ignore closed-loop physics are particularly vulnerable under multiplicative gain-ramp attacks; the problem is further exacerbated by the memory-dependent nature of fractional-order controllers, whose long-term history effects can mask attack-induced inconsistencies in the residuals unless system dynamics, actuator behavior, attack-channel constraints, and fractional-order memory are jointly embedded in the learning objective. This paper addresses that gap through a physics-informed neural network (PINN) framework that jointly estimates latent system states and reconstructs the multiplicative attack gain by incorporating plant dynamics, SVC actuator dynamics, measurement consistency, attack-channel reconstruction, and the reconstructed fractional-order command inside the SVC actuator residual. Detection is declared when a smoothed composite residual exceeds a calibrated threshold. In addition, a supervisory switching mechanism is introduced to replace corrupted wide-area control signals with reconstructed secure equivalents after confirmed detection, thereby restoring closed-loop damping performance without interrupting controller operation. Simulation results on a FACTS-based damping control system under gain-based FDIA scenarios show that the proposed detector achieves up to 99.42% accuracy, 98.86% precision, 99.18% recall, and a 99.02% F1-score, with a false alarm rate as low as 0.53%. The recovery mechanism also substantially reduces post-attack speed-deviation overshoot relative to the unmitigated case. Full article

Accurate and reliable state of health (SOH) estimation is essential for lithium-ion battery safety and lifetime management. Physics-informed neural networks (PINNs) integrate battery degradation knowledge with deep learning to improve the physical consistency of SOH estimation. However, existing PINNs insufficiently capture degradation evolution across consecutive cycles and show limited applicability across battery chemistries and charge/discharge protocols. To address these issues, we propose a sequence-conditioned physics-informed neural network (SC-PINN) that uses a gated recurrent unit (GRU) to encode local degradation states and incorporates a degradation dynamics consistency constraint and a monotonic degradation prior to improve prediction accuracy and physical plausibility. We also develop a general charging feature extraction method for different battery chemistries and charge/discharge protocols, extracting unified 18-dimensional health features from charging segments at high state of charge (SOC). The method is validated using 317 batteries and 197,024 valid cycle samples covering four battery chemistries across nine representative tasks. SC-PINN achieves a mean absolute percentage error (MAPE) of 0.6285% and a root mean square error (RMSE) of 0.8924%, with the lowest RMSE in eight of nine tasks. Compared with other methods, the proposed method demonstrates good stability and applicability in both standard and small-sample experiments. Full article

We develop two parametric data-driven reduced models: a physics-informed neural network (PINN) and a non-intrusive tensorial reduced-order model (TROM), and apply both approaches to the parametrized one-dimensional shallow-water dam-break problem. Both reduced models do not require time integration and learn a direct solution map from space, time, and dam-break parameters to the physical state. We present a detailed comparison for out-of-sample and extrapolated parameter values. In addition, we demonstrate that it is essential to introduce shock-aware collocation to improve the robustness of the PINN model. Full article