Jianhong Gan

Accurate correction of gridded numerical weather prediction precipitation forecasts remains challenging because many end-to-end deep learning correction models treat meteorological variables as undifferentiated data channels and therefore provide limited physical interpretability. Inspired by general principles of biological environmental sensing, selective information processing, and regulatory constraint learning, this study proposes PCPNet, a bio-inspired and physically constrained precipitation correction framework. The framework does not imitate a specific biological organ or species; instead, it abstracts three information-processing principles into a meteorological correction task. First, key weather-system cues, including low-level shear lines, trough-ridge effects, upper-level jet-stream forcing, vorticity-divergence-related vertical motion, and water-vapor flux convergence, are quantified as structured diagnostic fields. This transforms the subjective synoptic diagnosis of forecasters into automated grid-based sensing features. Second, these diagnostic cues are fused with numerical weather prediction variables and terrain descriptors in an encoder–attention–decoder network, allowing the model to emphasize dynamically important precipitation-triggering regions. Third, water-vapor conservation and terrain-forcing relationships are embedded as differentiable constraint losses, providing training-time constraint-based regulation that guides the corrected precipitation field toward physically consistent solutions. The method is evaluated from 2021 to 2023 in Hunan Province, China, using hourly numerical weather prediction model outputs as input features, China Meteorological Administration Land Data Assimilation System gridded analysis data as the training target, and independent meteorological station observations for strict cross-validation. PCPNet reduces the mean absolute error by 22.1% compared with the uncorrected China Meteorological Administration Land Data Assimilation System gridded precipitation products and outperforms Linear Regression, Bagging, Boosting, Multi-Layer Perceptron, TabNet, and Tree-based Progressive Regression Models by 12.9%, 13.5%, 16.9%, 10.8%, 14.9%, and 15.9%, respectively. The single-day event analysis provides an initial demonstration of heavy precipitation recovery capability, while comprehensive validation across long-term continuous weather events is planned for future operational deployment to further verify model stability. These results indicate that bio-inspired sensing and regulatory constraint learning can improve both the accuracy and interpretability of precipitation nowcasting correction.

Agricultural cold chain logistics is characterized by inherent challenges—product perishability, high carbon emissions, and stringent time windows—which are further exacerbated by dynamic disruptions. Existing methods suffer from slow adaptability, unstable multi-objective convergence, and severe cold-start issues. This work falls within the broad scope of biomimetics—the science of emulating nature’s time-tested strategies to solve complex engineering problems—and bio-inspired data-driven methods and their applications in engineering control, optimization, and artificial intelligence. The proposed H-MODRL framework embodies core biomimetic principles: the Genetic Algorithm (GA) mimics Darwinian natural selection and genetic inheritance, the Sparrow Search Algorithm (SSA) abstracts the cooperative foraging and anti-predation behaviors of sparrow populations in nature, and the Arrhenius-based freshness-decay model captures the biochemical kinetics governing perishable biological products. By synergistically integrating these biological evolution principles, swarm intelligence, and deep learning, the framework tackles real-world logistics complexity in a manner directly inspired by living systems. This study presents a well-organized hybrid optimization framework (H-MODRL) that couples a three-stage hybrid evolutionary mechanism, synergistically integrating heuristic warm-start, evolutionary policy guidance, and deep reinforcement learning decision-making. First, an improved genetic algorithm combined with the earliest deadline first strategy constructs a feasible initial population satisfying hard time-window constraints. Second, a large neighborhood search-enhanced chaotic sparrow search algorithm builds a high-quality elite guidance set for policy learning. Third, a physics-based multi-objective proximal policy optimization model embedded with Arrhenius equation-derived freshness-decay kinetics performs online decision-making. Experiments demonstrate that pre-computed all-pairs shortest paths and an O(1) hash-based dynamic-disruption indexing mechanism support fast online replanning. On heterogeneous simulated terrains based on real Chinese geospatial data, H-MODRL outperforms state-of-the-art algorithms across four objectives—logistics cost, carbon emissions, terminal freshness, and delivery time—while exhibiting compact, low-variance performance distributions, thereby validating its engineering robustness and practical value in complex agricultural cold chain environments.

Radar image extrapolation serves as a fundamental methodology in meteorological forecasting, facilitating precise short-term weather prediction through spatiotemporal sequence analysis. Conventional approaches remain constrained by progressive image degradation and artifacts, substantially limiting operational forecasting reliability. This research introduces E-HEOA—an enhanced deep learning architecture with integrated hyperparameter optimization. Our framework incorporates two fundamental innovations: (a) a hybrid metaheuristic optimizer merging a Gaussian-mutated ESOA and Cauchy-mutated HEOA for autonomous learning rate and dropout optimization and (b) embedded ConvLSTM2D modules for enhanced spatiotemporal feature preservation. Experimental validation on 170,000 radar echo samples demonstrates superior performance, demonstrating considerable enhancement in almost all aspects relative to the baseline model while establishing new state-of-the-art benchmarks in prediction fidelity, convergence efficiency, and structural similarity metrics.

Natural gas pipeline leak monitoring suffers from severe environmental noise, non-stationary signals, and complex multi-source variable couplings, limiting prediction accuracy and robustness. Inspired by biological perceptual systems, particularly their multimodal integration and dynamic attention allocation, we propose GL-TransLSTM, a biomimetic hybrid deep learning model. It synergistically combines Transformer’s global self-attention (emulating selective focus) and LSTM’s gated memory (mimicking neural temporal retention). The architecture incorporates a multimodal fusion pipeline; raw sensor data are first decomposed via CEEMDAN to extract multi-scale features, then processed by an enhanced LSTM-Transformer backbone. A novel physics-informed gated attention mechanism embeds gas diffusion dynamics into attention weights, while an adaptive sliding window adjusts temporal granularity. This study makes evaluations on an industrial dataset with methane concentration, temperature, and pressure, GL-TransLSTM achieves 99.93% accuracy, 99.86% recall, and 99.89% F1-score, thereby significantly outperforming conventional LSTM and Transformer-LSTM baselines. Experimental results demonstrate that the proposed biomimetic framework substantially enhances modeling capacity and generalization for non-stationary signals in noisy and complex industrial environments through multi-scale fusion, physics-guided learning, and bio-inspired architectural synergy.

Accurate and automated identification of water bodies from satellite imagery is critical for environmental monitoring, water resource management, and disaster response. Current deep learning approaches, however, suffer from a strong dependence on manual hyperparameter tuning, which limits their automation capability and robustness in complex, multi-scale scenarios. To overcome this limitation, this study proposes a fully automated segmentation framework that synergistically integrates an enhanced U-Net model with a novel hybrid evolutionary optimization strategy. Extensive experiments on public Kaggle and Sentinel-2 datasets demonstrate the superior performance of our method, which achieves a Pixel Accuracy of 96.79% and an F1-Score of 94.75, outperforming various mainstream baseline models by over 10% in key metrics. The framework effectively addresses the class imbalance problem and enhances feature representation without human intervention. This work provides a viable and efficient path toward fully automated remote sensing image analysis, with significant potential for application in large-scale water resource monitoring, dynamic environmental assessment, and emergency disaster management.

The accuracy of numerical model outputs strongly depends on the quality of the initial wind field, yet ground observation data are typically sparse and provide incomplete spatial coverage. More importantly, many current mainstream correction models rely on reanalysis grid datasets like ERA5 as the true value, which relies on interpolation calculation, which directly affects the accuracy of the correction results. To address these issues, we propose a new deep learning model, PPWNet. The model directly uses sparse and discretely distributed observation data as the true value, which integrates observation point positions and a physical consistency term to achieve a high-precision corrected wind field. The model design is inspired by biological intelligence. First, observation point positions are encoded as input and observation values are included in the loss function. Second, a parallel dual-branch DenseInception network is employed to extract multi-scale grid features, simulating the hierarchical processing of the biological visual system. Meanwhile, PPWNet references the PointNet architecture and introduces an attention mechanism to efficiently extract features from sparse and irregular observation positions. This mechanism reflects the selective focus of cognitive functions. Furthermore, this paper incorporates physical knowledge into the model optimization process by adding a learned physical consistency term to the loss function, ensuring that the corrected results not only approximate the observations but also adhere to physical laws. Finally, hyperparameters are automatically tuned using the Bayesian TPE algorithm. Experiments demonstrate that PPWNet outperforms both traditional and existing deep learning methods. It reduces the MAE by 38.65% and the RMSE by 28.93%. The corrected wind field shows better agreement with observations in both wind speed and direction, confirming the effectiveness of incorporating position information and a physics-informed approach into deep learning-based wind field correction.

Convolutional neural networks (CNNs) and their improved models (like DenseNet-121) have achieved significant results in image classification tasks. However, the performance of these models is still constrained by issues such as hyperparameter optimization and gradient vanishing and exploding. Owing to their unique exploration and exploitation capabilities, evolutionary algorithms offer new avenues for addressing these problems. Simultaneously, to prevent these algorithms from falling into a local optimum during the search process, this study designs a novel interpolation algorithm. To achieve better image classification performance, thus enhancing classification accuracy and boosting model stability, this paper utilizes a hybrid algorithm based on the horned lizard algorithm with quadratic interpolation and the giant armadillo optimization with Newton interpolation (HGAO) to optimize the hyperparameters of DenseNet-121. It is applied to five datasets spanning different domains. The learning rate and dropout rate have notable impacts on the outcomes of the DenseNet-121 model, which are chosen as the hyperparameters to be optimized. Experiments are conducted using the HGAO algorithm on five image datasets and compared with nine state-of-the-art algorithms. The performance of the model is evaluated based on accuracy, precision, recall, and F1-score metrics. The experimental results reveal that the combination of hyperparameters becomes more reasonable after optimization with the HGAO algorithm, thus providing a crucial improvement. In the comparative experiments, the accuracy of the image classification on the training set increased by up to 0.5%, with a maximum reduction in loss of 0.018. On the test set, the accuracy rose by 0.5%, and the loss decreased by 54 points. The HGAO algorithm provides an effective solution for optimizing the DenseNet-121 model. The designed method boosts classification accuracy and model stability, which also dramatically augments hyperparameter optimization effects and resolves gradient difficulties.

Hyper-parameters play a critical role in neural networks; they significantly impact both training effectiveness and overall model performance. Proper hyper-parameter settings can accelerate model convergence and improve generalization. Among various hyper-parameters, the learning rate is particularly important. However, optimizing the learning rate typically requires extensive experimentation and tuning, as its setting is often dependent on specific tasks and datasets and therefore lacks universal rules or standards. Consequently, adjustments are generally made through trial and error, thereby making the selection of the learning rate complex and time-consuming. In an attempt to surmount this challenge, evolutionary computation algorithms can automatically adjust the hyper-parameter learning rate to improve training efficiency and model performance. In response to this, we propose a black widow optimization algorithm based on Lagrange interpolation (LIBWONN) to optimize the learning rate of ResNet18. Moreover, we evaluate LIBWONN’s effectiveness using 24 benchmark functions from CEC2017 and CEC2022 and compare it with nine advanced metaheuristic algorithms. The experimental results indicate that LIBWONN outperforms the other algorithms in convergence and stability. Additionally, experiments on publicly available datasets from six different fields demonstrate that LIBWONN improves the accuracy on both training and testing sets compared to the standard BWO, with gains of 6.99% and 4.48%, respectively.

With the rapid development of deep learning in artificial intelligence, sequential recommendation systems play an increasingly important role in e-commerce, social media, digital entertainment, and other fields. This work systematically reviews the research progress of deep learning in sequential recommendation systems from a methodological perspective. This paper focuses on analyzing three dominant technical paradigms: contrastive learning, graph neural networks, and attention mechanisms, elucidating their theoretical innovations and evolutionary trajectories in sequential recommendation systems. Through empirical investigation, we categorize the prevailing evaluation metrics, benchmark datasets, and characteristic distributions of typical application scenarios within this domain. This work further proposes promising avenues for sequential recommendation systems in the future.