Controlling temperature and humidity in greenhouses is a complex challenge due to its non-linear nature and dependence on external weather conditions. Conventional control methods often experience energy inefficiency and delayed responses to drastic changes. This study proposes a hybrid approach by combining Long Short-Term Memory (LSTM) optimized using Genetic Algorithm (GA) for temperature prediction, and Fuzzy Logic Controller (FLC) for actuator decision-making. Genetic Algorithm is employed to find optimal hyperparameters (number of neurons and batch size) in the LSTM architecture. Experimental results demonstrate the GA-LSTM model's capability in predicting temperature with high accuracy, yielding an R² score of 0.9881 and Root Mean Square Error (RMSE) of 1.1273°C. These accurate predictions are subsequently used as input to the FLC to regulate exhaust fan speed and mist pump status. Simulations demonstrate the system's capability in making energy-efficient decisions—activating actuators only when conditions are predicted to deviate from ideal values—while remaining responsive to extreme temperature anomalies.

Background: Diabetes mellitus affects 463 million people worldwide and necessitates continuous blood glucose monitoring. Current glucose prediction systems often lack efficiency, and real-time prediction is essential for timely clinical intervention. Objectives: This study aims to develop and validate a novel Convolutional Recurrent Neural Network (CRNN) enhanced with bio-inspired algorithms to improve blood glucose prediction and enable real-time detection of hypoglycemia and hyperglycemia. Materials and methods: The proposed framework employs a CRNN architecture that combines Convolutional Neural Networks (CNNs) for feature extraction with Long Short-Term Memory (LSTM) layers for temporal sequence learning. The model was trained and evaluated using the HUPA-UCM diabetes dataset. Additionally, the study benchmarks the proposed model against 19 traditional Machine Learning (ML) algorithms and compares it with state-of-the-art methods from the literature. Results: The proposed approach demonstrates superior predictive capability, consistently delivering promising results across multiple evaluation frameworks. The model achieves clinically acceptable prediction intervals, confirming its effectiveness in enhancing the accuracy and reliability of blood glucose prediction for diabetes management. Conclusion: The findings demonstrate that the proposed CRNN model, enhanced with bio-inspired algorithms, provides an effective and reliable solution for real-time blood glucose prediction. By outperforming conventional ML methods and achieving clinically acceptable accuracy levels, the model shows strong potential for integration into intelligent diabetes management systems to support timely clinical decisions and improve patient outcomes.

GFASNet: Gait feature attention-driven deep sequential network for dementia-related gait pattern analysis.

Evaluating the impact of Numerical Weather Prediction variables on wind power forecasting: A case study of the Alpha Ventus offshore wind farm

The critical trigger of 2-methylisoborneol in Taihu Lake: Flow velocity.

A long-term prediction model with Gaussian linear fuzzy granules based on convolutional neural networks and long short-term memory

A data-driven model for ship trajectory prediction with integrated hydrometeorological conditions: A case study in complex waters

Early deployment of deep learning models for lithium-ion battery state-of-health prediction with limited initial data

DeepHybridCPI: A hybrid deep learning framework for compound-protein interaction prediction.

This work integrates Bayesian regime detection with conditional neural processes for 24-hour electricity price forecasting in the German, French, and Norwegian markets. Regimes are inferred via a disentangled sticky hierarchical Dirichlet process hidden Markov model (DS-HDP-HMM). For each regime, an independent conditional neural process (CNP) learns localized mappings from input contexts to 24-dimensional hourly price trajectories; final forecasts are produced as regime-weighted mixtures of the regime-specific CNP outputs. Temporal robustness and cross-market generalization are evaluated on Germany (2021–2023) and on France and Norway (2023). We benchmark against deep neural networks (DNN), the Lasso estimated autoregressive (LEAR) model, extreme gradient boosting (XGBoost), Bayesian long short-term memory (BLSTM), and the temporal fusion transformer (TFT), and assess downstream value through battery storage optimization. Results indicate that the proposed regime-aware CNP often delivers higher profits or lower costs, while DNN can be exceptionally competitive in specific cost-minimization settings. Because point accuracy does not necessarily translate into operational optimality, we apply the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to aggregate forecasting and operational criteria. TOPSIS ranks the CNP as the leading model for 2023 and, overall, as the most balanced and consistently preferred solution across the considered markets.

Enhancing pulmonary embolism diagnosis: a squeeze-and-attention U-Net for precise detection and segmentation in CT angiography.

Accurate estimation of the State of Charge (SOC) is essential for enhancing the efficiency and reliability of Battery Management Systems (BMS) in Internet of Things (IoT) applications. This study introduces the Pattern-Aware Transformer Model (PATM), an interpretable framework for SOC prediction in Float-Nominal (FN), Constant-Current (CC), and Energy Release (ER) scenarios. PATM extends the standard Transformer architecture by incorporating a pattern embedding mechanism that explicitly encodes operating conditions and directs adaptive attention allocation. A feature engineering pipeline that combines mutual information (MI) ranking and principal component analysis (PCA) reduces dimensionality while preserving physically relevant variables. On real-world data, PATM achieves an RMSE of 2.08 × 10 −3 and an R 2 of 0.9998, outperforming the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) baselines. Compared with single-scenario CC modeling, multi-scenario learning reduces RMSE, MAE, MSE, and MAPE by 54.9%, 80.1%, 79.6%, and 75.9%, respectively. Ablation studies further demonstrate that removing the embedding module increases RMSE by 2.4%, MAE by 17.8%, and MSE by 4.9%, while leaving R 2 nearly unchanged. This indicates that the embedding mechanism enhances cross-scenario robustness and error stability. SHapley Additive exPlanations (SHAP) analysis and attention visualizations reveal the model’s dependence on physically relevant factors, including temperature gradients, voltage fluctuations, and internal resistance. • This study combines MI ranking and PCA for efficient feature selection, preserving interpretability while reducing redundancy. • This study introduces a multi-scenario approach that integrates data from diverse operating phases, enhancing adaptability and accuracy. • This study introduces the Pattern-Aware Transformer Model (PATM), and experimental results show that it outperforms LSTM and GRU across multiple metrics, demonstrating its accuracy and robustness.

A training sample reduction strategy for data-driven surrogate modeling of wind-induced structural responses

• Comparison of multi-horizon net load forecasts using real data from Cypriot PV prosumers. • Univariate models (both statistical and RNN) are effective when meteorological data is unavailable. • LSTM achieves top accuracy for medium-term univariate net load forecasting. • ARIMA and TSLM outperform RNN models when exogenous inputs are included. • Exogenous weather data improves Transformers, harms RNNs in forecasting tasks. Considering the European Union (EU) climate neutrality objectives, the increasing adoption of residential photovoltaic (PV) systems presents new challenges for grid reliability, especially in isolated electricity systems such as in islands. This study evaluates recursive and direct multi-horizon forecasting strategies of household net load (defined as electricity consumption minus PV generation), using real-world 30-min data from 68 PV households in Cyprus (one year). Forecasts were produced for horizons from 1 to 73 days, and benchmarked across recursive statistical models, including seasonal autoregressive-integrated-moving average (ARIMA) and Time-series Linear model (TSLM), and direct multi-output deep learning (DL) models, including Long Short-Term Memory (LSTM) and Transformer architectures. Models were evaluated in both univariate (net load-only) and multivariate settings, with the latter incorporating exogenous variables such as solar irradiance, air temperature, and humidity. The results show that for models without exogenous parameters, ARIMA with seasonal adjustment had the best performance in the short-term (RMSE = 145 W at 5 days), while LSTM outperformed in the medium-term forecasts (RMSE = 433 W at 66 days). When exogenous parameters are included, statistical models, particularly ARIMA and TSLM with calendar–weather interactions consistently outperformed across all forecast horizons, including the medium-term (RMSE = 364 W at 73 days). Prediction-interval analysis further indicates horizon-widening uncertainty for recursive statistical forecasts, whereas direct deep learning ensembles tend to produce comparatively stable interval widths, with regime-dependent changes during PV-active periods. These findings provide practical guidance on horizon-dependent model selection and the conditional value of exogenous inputs for planners and operators managing PV-rich residential systems.

This study presents an integrated framework for calibrating material parameters and their associated uncertainties in continuum finite element (FE) simulations of progressive damage in thin wood veneer laminates. Compact tension tests on two laminate layups, [ 0 / 90 ] 2 s and [ ± 45 ] 2 s , serve as the basis for calibrating FE input parameters both along and perpendicular to the grain direction. To enable efficient uncertainty quantification, a deep Long Short-Term Memory (LSTM) neural network is developed to rapidly predict full force vs displacement curves from these fracture tests. The FE input parameters are treated as random variables, and Bayesian inference with Markov Chain Monte Carlo sampling is applied to the LSTM surrogate models to estimate their distributions based on variability observed in experiments. Validation against experimental data demonstrates that the calibrated parameters accurately simulate damage progression, including uncertainty, in both compact tension and open-hole tension tests of quasi-isotropic [ 90 / 45 / 0 / − 45 ] s laminates, with mean prediction errors below 7%. • Development of integrated FEA–ML–Bayesian framework for progressive fracture simulation including uncertainties. • Curve-based uncertainty quantification using deep LSTM surrogate modelling for rapid model evaluation. • Experimentally validated prediction of variability across layups and mechanical tests.

Deep liquid neural network for prediction of weather-impacted flight delay

Deep learning-based prediction of the hysteretic behavior of buckling-restrained braces for seismic design using analysis-of-mean-based optimal hyperparameters

Current risk stratification for lower extremity deep vein thrombosis remains limited, often failing to identify high-risk patients for impending pulmonary embolism (PE) and leading to non-guideline-concordant overtreatment. We aimed to develop and validate a novel artificial intelligence (AI) system that processes dynamic digital subtraction angiography (DSA) radiomics, with the potential to guide precision therapy during endovascular intervention. In a retrospective cohort study of 168 patients treated at a single vascular surgery center (2019-2023), we developed a hybrid deep learning model integrating a transformer-UNet for spatial feature extraction and a long short-term memory (LSTM) network for temporal hemodynamic analysis. This model processed intraprocedural dynamic DSA sequences to quantify novel thrombus kinematic parameters (eg, displacement velocity, oscillation angle θ) and hemodynamic parameters venous (quantitative flow ratio). The model's performance for predicting subsequent PE was compared against the Wells score. Its impact on clinical decision-making and 12-month outcomes was evaluated rigorously. The AI model demonstrated significantly superior discriminative performance for predicting PE compared with the Wells score (area under the curve, 0.88; 95% confidence interval [CI], 0.85-0.92 vs 0.76; 95% CI, 0.70-0.83; P = .026). Implementation of the AI-guided strategy was associated with markedly improved clinical outcomes at the 12-month follow-up: a 54% lower incidence of PE (3.4% vs 11.1%; relative risk [RR], 0.46; 95% CI, 0.08-0.82; P = .005), a 62% lower incidence of severe post-thrombotic syndrome (Villalta score ≥10; 8.0% vs 21.0%; RR, 0.38; 95% CI, 0.17-0.86; P = .008), and a lower prevalence of preexisting inferior vena cava filters in the AI-stratified high-risk group (25.3% vs 44.4%; RR, 0.57; 95% CI, 0.36-0.89; P < .001), without a significant increase in major bleeding events (2.3% vs 7.4%; P = .096). An AI-guided risk stratification system based on dynamic DSA radiomics accurately identifies thrombus instability and hemodynamic impairment in real time and suggests its potential to help enable more personalized therapeutic decisions during intervention. In this retrospective analysis, AI-based risk stratification was associated with a significantly lower incidence of PE and severe post-thrombotic syndrome while safely curbing the overuse of inferior vena cava filters, representing a transformative advancement in the precision management of acute deep vein thrombosis.

Poincaré feature-based classification of electroencephalography signals for multiple sclerosis diagnosis.

From classical models to attention-based transformers: A comparative study of injury prediction pipelines in female varsity soccer.