Root Mean Square Error Research Articles

Novel load prediction in microservice architecture using attention mechanism-based deep LSTM networks

Load balancing in microservice architecture is essential for optimizing resource utilization and maintaining high availability. Traditional load balancing algorithms like First-Come-First-Serve (FCFS) and Round Robin often lead to inefficiencies due to their inability to account for server capabilities and varying request sizes. Machine Learning (ML) offers a promising solution by predicting future load patterns and distributing requests more effectively. In this study, we propose an innovative, novel attention mechanism-based Long Short-Term Memory (LSTM) network for web server load prediction. Our methodology involves detailed data preprocessing, sequence creation, and scaling to prepare the NASA HTTP dataset for model training. The attention mechanism enhances the LSTM network’s ability to focus on relevant input sequences, significantly improving predictive accuracy. Compared to traditional algorithms such as linear regression, polynomial regression, L2 regularization, decision tree regression, XGBoost, and ARIMA, our model achieves the lowest Mean Squared Error (MSE) of 187,293.59 and Root Mean Squared Error (RMSE) of 432.77, with a strong R-squared value of 0.8532. This superior performance highlights the model’s effectiveness in capturing both short-term and long-term dependencies in the data. This novel predictive model can be used to integrate into dynamic and efficient load balancing frameworks. Accurate future load predictions from AMDLN in the microservices environment optimize resource utilization and save infrastructure costs by providing accurate future load predictions for scaling up and scaling down of microservices.

A deep learning approach for reconstructing hourly surface air temperature in Qinghai for the period 2006-2015

The Qinghai-Tibet Plateau, often referred to as the “Roof of the World”, has temperature variations that significantly affect both the local and surrounding ecosystems. Qinghai Province, located at the northeastern gateway to the plateau, serves as a microcosm of the plateau’s broader temperature changes. Therefore, it is crucial to study the temperature changes in Qinghai Province. However, existing meteorological data, including ground observations, atmospheric reanalysis data, and satellite remote sensing data, suffer from low accuracy due to limitations in time and space. In this research, we introduce a deep learning approach to integrate multi-source temperature data and reconstruct a 10-year near-surface temperature dataset with a temporal resolution of 1 hour and a spatial resolution of 0.01°. The reconstruction of near-surface temperature was performed by inputting two atmospheric reanalysis products-European Centre for Medium-Range Weather Forecasts Regional Reanalysis Version 5 (ERA5) and Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2)-along with auxiliary variables that influence temperature, into the newly developed Temporal Gated Positional Model (TGPM) deep learning model. The reconstructed dataset from the TGPM model was then compared to ground observation data and evaluated against datasets generated by five other machine learning methods. The results show that the fusion dataset generated by the TGPM model closely fits the target data, outperforming the products from the other five methods across all evaluation metrics. The correlation coefficient (CC), coefficient of determination (R2), mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), and relative bias (RB) for the TGPM near-surface temperature dataset were 0.9949, 0.9896, 0.7713°C, 1.0901°C, 1.1888 °C2, and 4.02%, respectively. Overall, the deep learning method proposed in this study is more suitable for near-surface temperature reconstruction in Qinghai Province, providing a robust foundation for further research on temperature variations and ecological conservation in both Qinghai and the broader Qinghai-Tibet Plateau. The reconstructed dataset also offers actionable insights for climate policy formulation, disaster risk reduction, and ecosystem management in high-altitude regions.

Research on Trajectory Tracking of a Two-Link Robot Based on Hybrid GA-SQP Optimized Feedforward-PID Control

The increasing deployment of robotic systems in industrial applications has driven widespread use of two-link robots, valued for their high speed and precision. However, their inherent nonlinear dynamics and strong coupling effects present substantial challenges to achieving high-precision trajectory tracking. To address these issues, this paper proposes a feedforwardPID control strategy optimized using a hybrid Genetic AlgorithmSequential Quadratic Programming (GASQP) approach. The proposed method combines the anticipatory capabilities of feedforward control with the corrective feedback of PID control, enabling automatic and efficient parameter tuning. Simulation results demonstrate that, in comparison to conventional PID control, the proposed approach enhances trajectory tracking accuracy by approximately 39.61%. Specifically, the GASQP-optimized controller reduces the Root Mean Square Error (RMSE) to 0.48mm for an Archimedean spiral trajectory, and further to 0.01mm for a Sine-like trajectory, confirming its adaptability across various trajectory profiles. Torque analysis further highlights the complementary interaction between feedforward and PID components, substantiating the methods effectiveness. These results underscore the proposed strategys potential to significantly improve trajectory tracking accuracy and robustness for two-link robots, especially in complex dynamic environments.

Evaluating Bias Correction Methods Using Annual Maximum Series Rainfall Data from Observed and Remotely Sensed Sources in Gauged and Ungauged Catchments in Uganda

This research addresses the challenge of bias in Remotely Sensed Rainfall (RSR) datasets used for hydrological planning in Uganda’s data-scarce, ungauged catchments. Four bias correction methods, Quantile Mapping (QM), Linear Transformation (LT), Delta Multiplicative (DM), and Polynomial Regression (PR), were evaluated using daily rainfall data from four gauged stations (Gulu, Soroti, Jinja, Mbarara). QM consistently outperformed other methods based on statistical metrics, e.g., for National Oceanic and Atmospheric Administration Climate Prediction Center (NOAA_CPC) RSR data at Gulu, Root-Mean-Square Error (RMSE) was reduced from 29.20 mm to 19.00 mm, Mean Absolute Error (MAE) reduced from 22.44 mm to 12.84 mm, and Percent Bias (PBIAS) reduced from −19.23% to 1.05%, and improved performance goodness-of-fit tests (KS = 0.03, p = 1.00), while PR, though statistically strong, failed due to overfitting. A bias correction framework was developed for ungauged catchments, using predetermined bias factors derived from observed station data. Validation at Arua (tropical savannah) and Fort Portal (tropical monsoon) demonstrated significant improvements in RSR data when the bias correction framework was applied. At Arua, bias correction of Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) data reduced RMSE from 49.14 mm to 21.41 mm, MAE from 45.74 mm to 17.38 mm, and PBIAS from −59.83% to −8.18%, while at Fort Portal, bias correction of the CHIRPS dataset reduced RMSE from 28.35 mm to 15.02 mm, MAE from 25.28 mm to 11.35 mm, and PBIAS from −46.2% to 4.74%. Our research concludes that QM is the most effective method, and that the framework is a tool for improving RSR data in ungauged catchments. Recommendations for future work includes machine learning integration and broader regional validation.

Mining autonomous student patterns score on LMS within online higher education

Higher education institutions actively integrate information and communication technologies through learning management systems (LMS), which are crucial for online education. This study used data mining techniques to predict the autonomous scores of students in the online Law and Psychology programs at the Technical University of Manabi. The process involved data integration and selection of more than 16,000 records, preprocessing, transformation with RobustScaler, predictive modelling that included recursive feature elimination with cross-validation to select features (RFEcv), and hyperparameter fitting to achieve the best fit, and finally, evaluation of the models using metrics of root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R2). The feature selection framework suggested by RFEcv contributed to the performance of the models. The variables analyzed focused on download rate, homework submission rate, test performance rate, median daily accesses, median days of access per month, observation of comments on teacher-reviewed assignments, length of final exam, and not requiring the supplemental exam. Hyperparameter adjustment improved the performance of the models after applying RFEcv. The models evaluated showed minimal differences in RMSE ([0.5411 .. 0.6025]). The gradient boosting model achieved the best performance of R2 = 0.6693, MAE = 0.4041 and RMSE = 0.5411 with the Law online program data, as with the Psychology online program data, with an R2 = 0.6418, MAE = 0.4232 and RMSE = 0.6025, while the combination of both data sets reflected the best performance with the extreme gradient boosting (XGBoost) model with the values of R2 = 0.6294, MAE = 0.4295 and RMSE = 0.5985. Future research and implementations could include autonomous score data through plugins and reports integrated into LMSs. This approach may provide indicators of interest for understanding and improving online learning from a personalized, real-time perspective.

Significant Improvement in Short-Term Green-Tide Transport Predictions Using the XGBoost Model

Accurately predicting the drift trajectory of green tides is crucial for assessing potential risks and implementing effective countermeasures. This paper proposes a short-term green-tide drift prediction method that combines green-tide patch characteristics, 1 h interval drift distances from GOCI-II images, and driving-factor data using the XGBoost machine learning model to enhance prediction accuracy. The results demonstrate that the proposed method outperforms the traditional OpenDrift model in short-term predictions. Specifically, at time intervals of 3, 5, and 7 h, the root mean square errors (RMSEs) of the OpenDrift model in the zonal direction are 1.81 km, 2.89 km, and 3.55 km, respectively, whereas the RMSEs of the proposed method are 0.80 km, 0.98 km, and 1.20 km, respectively; in the meridional direction, the RMSEs of the OpenDrift model are 1.77 km, 2.67 km, and 3.10 km, while the RMSEs for the proposed method are 0.82 km, 1.10 km, and 1.25 km, respectively. Furthermore, the proposed XGBoost method more-accurately tracks the actual positions of green-tide patches compared to the OpenDrift model. Specifically, at the 25 h interval, the proposed method continues to accurately predict patch positions, while the OpenDrift model exhibits significant deviations. This study demonstrates that the proposed method, by learning drift patterns from historical data, effectively predicts the short-term drift process of green tides. It provides valuable support for early warning systems, thereby helping to mitigate the ecological and economic impacts of green-tide disasters.

Predicting climate change impacts on groundwater aquifer levels in the Henan North China Plain

Monitoring GWL over extended periods is crucial for comprehending the fluctuations of groundwater resources in the present context for ongoing global changes. This study analyzed the effects of climate variations on the GWL in Henan Province North China Plain using two deep-learning models Bidirectional Long Short-Term Memory (BidLSTM) and Gated Recurrent Unit (GRU). These models predicted monthly variations in GWL at 85 monitoring wells across the area using a dataset from 1980 to 2015. For validation and evaluation, both models were quantitatively calibrated using training set (1980–2015) to predict GWL from 2016 to 2100. The dataset was partitioned, with 80% allocated for training and 20% for testing. The result interpreted that in AHP3 well, GWL declined to 120 m in 1980 due to reduced precipitation 57 mm and Et 62 mm, while temperature stayed at 10 °C as of 2070, In the Zhengzhou and Keifing regions GWL declined by 98 m in the 1980 s despite rising precipitation 72 mm and Et 60 mm, due to insufficient recharge by 2100, GWL is expected to reach 140 m, driven by climate changes, including a temperature increase to 17 °C. The results indicated significant changes with the effect of precipitation, significant increase in temperature and surface Et. Anthropogenic activity also impacted GWL in the area. The trained models demonstrated good performance, with a prediction error of 0.0350, 0.0346 m, and the root mean square error (RMSE) was recorded at 0.1870, 0.1860 m. By accurately predicting GWLs, the BidLSTM model can help ensure that groundwater resources are used sustainably and efficiently.

Research on rock burst prediction based on an integrated model

Rockburst is a significant safety threat in coal mining, influenced by complex nonlinear dynamic characteristics and multi-factor coupling. This study proposes a rockburst risk prediction method based on the SSA-CNN-MoLSTM-Attention model. The model integrates the local feature extraction capability of convolutional neural networks (CNN), the temporal modeling advantages of the modified long short-term memory network (MoLSTM), and the enhanced feature recognition capability of the attention mechanism. Additionally, the sparrow search algorithm (SSA) is employed to optimize hyperparameters, further improving the model’s performance. Unlike traditional approaches that rely on time-axis-based analysis, this study uses the working face advancement distance as the basis for prediction, which better reveals the potential spatial correlations of rockburst occurrences, aligning with engineering practice needs.Validation using microseismic monitoring data from a coal mine demonstrates that the proposed model achieves a prediction accuracy of 93.62% and an F1-score of 93.54%. The model outperforms traditional methods in mean absolute error (MAE) and root mean square error (RMSE), providing effective insights and a reference for rockburst risk assessment and disaster prevention in mining operations.

Predicting tuberculosis drug properties using extended energy based topological indices via a python driven QSPR approach

In the present work, the physicochemical characteristics of important anti-tuberculosis (TB) drugs such as isoniazid, pyrazinamide, ethambutol, ethionamide, linezolid, and levofloxacin are explored using extended energy-based topological indexes. Based on the molecules of the drugs, we calculate the extended energies of many widely recognized indexes such as Zagreb Second Index, Harmonic Index, Randic Index, Sombor Index, Reduced Sombor Index, and Average Sombor Index. All the calculations are done using Python, and the rigorous algorithmic implementation in the form of matrix formulation and computation of the eigenvalue is also given for reproducibility. We use the linear, quadratic, and logarithmic regression models to predict nine important physicochemical parameters: the boiling point, the melting point, the flash point, the molar refractivity, the polarizability, the molar volume, the molecular weight, the logarithm of the partition coefficient, and the surface area. Among the three models, the quadratic regression always yields the best predictability, as reflected in the largest coefficient of determination () as well as the minimum root mean square error (RMSE) values. Visual analyses such as heatmaps, scatter plot matrices, bar charts, and regression plots are employed to complement the numerical findings. Also, a rigorous discourse about model validity, model significance, and limitations is discussed. The entire source code and dataset are made available through GitHub to allow verification and transparency. The Python-based QSPR methodology, in addition to elucidating the high correlation of the topological descriptors with the properties of drugs, offers a drug design and optimization process in pharmaceutical research in an efficient way.

Improving urinary oxygen monitoring with a transit time algorithm: enhancing AKI detection in cardiac surgery.

Acute kidney injury (AKI) affects 40-50% of cardiac surgery patients and is closely linked to renal medullary hypoxia. Although urinary oxygen partial pressure (PuO2) offers real-time insight into renal oxygenation, variable urine transit times through the urinary catheter can impair measurement accuracy. This study aimed to develop an algorithm that calculates transit time by modeling urine flow as discrete particles and to assess whether it improves PuO2 estimation. The proposed algorithm models urine flow as discrete particles, tracking transit time through the urinary catheter. The transit time allows correcting oxygen measurements at the catheter exit, mitigating distortions from variable flow rates. Validation used a bench-top system with a flow sensor, a 30-cm glass tube simulating a catheter, and optode-based oxygen sensors positioned inside a flask and at the catheter entry and exit. Flow rates spanned 20-450 mL/h, and flask oxygen 15-120 mmHg, with exit compared to entrance values. Without adjustment, the root mean square error (RMSE) between entrance and exit oxygen measurements was 15.71 mmHg. Incorporating the transit time correction reduced the RMSE to 5.82 mmHg. This marked improvement indicates that the corrected measurements more accurately reflect the true oxygen levels entering the catheter across various flow conditions. By accounting for dynamic urine transit times, the proposed algorithm substantially enhances the accuracy of urinary oxygen monitoring. This improvement in estimating renal oxygenation may facilitate noninvasive detection of renal hypoxia and allow for timely interventions to reduce the incidence and severity of AKI in cardiac surgery patients.

Assimilation of the total electron content obtained from GNSS to a model of the ionosphere using a hierarchical Bayesian network

Ionospheric data assimilation aims to address the uneven spatiotemporal distribution of observational data and errors in numerical models. This paper proposes an ionospheric data assimilation model using the hierarchical Bayesian network (HBN) algorithm. We use the International Reference Ionosphere (IRI) 2016 as background model. The HBN method assimilates global navigation satellite system (GNSS) observational data from approximately 260 stations within the Crustal Movement Observation Network of China (CMONOC). For this analysis, we use the total electron content (TEC) data from the Center for Orbit Determination in Europe (CODE) and BeiDou Navigation Satellite System (BDS) geostationary earth orbit (GEO) experiments. We evaluate the HBN assimilation effect through single-frequency precise point positioning (PPP). The results demonstrate that the HBN algorithm closely aligns with the BDS GEO TEC, regardless of geomagnetic conditions. Statistical results show that, with BDS GEO TEC data as the ground truth reference, the HBN model improves the correlation coefficient by approximately 14% and reduces the root mean square error (RMSE) by around 33% compared to the IRI model. The assimilation effect is significantly superior to that of the Kalman filter. Additionally, the HBN-based PPP method demonstrates slightly improved GNSS positioning accuracy compared to CODE-based PPP, with a reduction in RMSE observed under both geomagnetically disturbed and quiet conditions. Thus, the HBN method is effective for ionospheric data assimilation.

Enhanced Financial Fraud Detection Using an Adaptive Voted Perceptron Model with Optimized Learning and Error Reduction

Financial fraud detection is an important field in financial technology, and strong and effective machine learning (ML) models are needed to detect fraudulent transactions with high accuracy and reliability. Conventional fraud detection models, like probabilistic, instance-based, and tree-based models, tend to have high error rates, class imbalance problems, and poor adaptability to changing fraud patterns. These issues call for sophisticated methods that improve predictive accuracy while being computationally efficient. To overcome these limitations, this research introduces the Voted Perceptron (VP) model, which utilizes an iterative learning process to dynamically adapt decision boundaries based on misclassified examples. In contrast to traditional models with static decision rules, the VP model constantly updates its weight parameters, thus providing better fraud detection abilities. The evaluation compares VP with state-of-the-art machine learning models, such as Average One Dependency Estimator (A1DE), K-nearest Neighbor (KNN), Naïve Bayes (NB), Random Tree (RT), and Functional Tree (FT), by using important performance metrics, like Mean Absolute Error (MAE), Root Mean Square Error (RMSE), True Positive Rate (TPR), recall, and accuracy. Experimental results show that VP outperforms its rivals significantly, yielding better fraud detection performance with low error rates and high recall. Furthermore, an ablation study confirms the influence of essential VP model elements on general classification performance. These results demonstrate VP to be an extremely effective model for detecting financial fraud, with enhanced flexibility towards evolving fraud patterns, and confirm the necessity for intelligent fraud detection mechanisms within financial organizations.

Fine carrier frequency offset estimation for OFDM and MIMO-OFDM systems: A comparative study

In Orthogonal Frequency Division Multiplexing (OFDM) and Multiple-Input Multiple-Output-OFDM (MIMO-OFDM) systems, estimating Carrier Frequency Offset (CFO) is a critical challenge, particularly in degraded channel conditions where traditional methods struggle with precision and adaptability. This comparative study views various existing CFO estimation techniques and identifies three conventional methods—CFOest, CC, and AF—as benchmarks. To enhance estimation accuracy, a machine learning-based approach is proposed to effectively function across different channel conditions. Three distinct CFO estimators are developed using Kernel Support Vector Machine (KSVM), Linear Discriminant Analysis (LDA), and Artificial Neural Network (ANN), as this is a common strategy in machine learning for identifying optimal solutions. A comparative analysis of their performance demonstrates that the proposed approach outperforms traditional techniques by achieving lower Root Mean Square Error (RMSE), with the ANN-based CFO estimator performing best in larger estimation ranges, while the KSVM-based estimator excels in smaller ranges. To further enhance accuracy, a novel three-step machine learning-based approach is proposed, demonstrating significant improvements in accuracy through subsequent simulations when contrasted with conventional methods and single-step models.

Comparative analysis of machine learning techniques for temperature and humidity prediction in photovoltaic environments

This research conducts a comparative analysis of nine Machine Learning (ML) models for temperature and humidity prediction in Photovoltaic (PV) environments. Using a dataset of 5,000 samples (80% for training, 20% for testing), the models—Support Vector Regression (SVR), Lasso Regression, Ridge Regression (RR), Linear Regression (LR), AdaBoost, Gradient Boosting (GB), Decision Tree (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)—were evaluated based on Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the Coefficient of Determination (R²). For temperature prediction, XGBoost demonstrated the best performance, achieving the lowest MAE of 1.544, the lowest RMSE of 1.242, and the highest R² of 0.947, indicating strong predictive accuracy. Conversely, SVR had the weakest performance with an MAE of 4.558 and an R² of 0.674. Similarly, for humidity prediction, XGBoost outperformed other models, achieving an MAE of 3.550, RMSE of 1.884, and R² of 0.744, while SVR exhibited the lowest predictive power with an R² of 0.253. This comprehensive study serves as a benchmark for the application of ML models to environmental prediction in PV systems, a research area that is relatively important. Notably, the results underscore the performance advantage of ensemble-based approaches, especially for XGBoost and RF compared to simpler, linear-based methods such as LR and SVR, when it comes to well-dispersed environmental interactions. The proposed machine-learning based power generation analysis approach shows significant improvements in predictive analytics capabilities for renewable energy systems, as well as a means for real-time monitoring and maintenance practices to improve PV performance and reliability.

Performance evaluation and improvement of ICESat-2 and GEDI forest canopy height retrievals in Northeast China

ABSTRACT The advent of new-generation spaceborne Light Detection and Ranging (lidar) systems, exemplified by the Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) and Global Ecosystem Dynamics Investigation (GEDI), has opened up an unprecedented opportunity for observing forest canopy structures. However, forest canopy height derived from ICESat-2 ATL08 land and vegetation products and GEDI L2A geolocated elevation and height products exhibit varying accuracy across different regions. Low data accuracy limits the broader application of these systems. Moreover, the canopy height detection abilities of the two spaceborne lidar systems in the ecologically important forests of Northeast China require further investigation. In this study, airborne lidar data were used to evaluate the performance of canopy height retrievals from ICESat-2 ATL08 and GEDI L2A, as well as their influencing factors. In addition, a canopy height improvement method, based on a random forest model, was proposed to enhance the accuracy and consistency of canopy height data derived from ICESat-2 and GEDI. The results indicate that the performance of strong beams surpassed that of weak beams in detecting canopy height, with only weak beam data collected during the day recommended for exclusion in ICESat-2 applications. In contrast, for the GEDI mission, the advantages of the power beam were not pronounced, and its performance was better during the day than at night. Compared to ICESat-2, GEDI exhibited lower accuracy in canopy height detection under nighttime conditions or in evergreen needle-leaved forests, but showed greater sensitivity to slope. Moreover, the proposed method increased the coefficient of determination (R2) for ICESat-2 canopy height accuracy from 0.53 to 0.82, and reduced the root mean square error (RMSE) from 3.98 m to 2.00 m. Similarly, the R2 for GEDI improved from 0.52 to 0.80, while RMSE decreased from 4.45 m to 2.41 m. The consistency between ATL08 and GEDI L2A was also improved, with the RMSE reduced by 3.14 m. The findings of this study could provide valuable guidance for the selection and utilization of the two spaceborne lidar data. Canopy height data derived from the improved strategy may enable new opportunities for forest canopy height mapping in Northeast China and support further applications, such as the quantification of aboveground carbon stocks in forests.

Robust ensemble learning frameworks for predicting minimum miscibility pressure in pure nitrogen and gas mixtures containing nitrogen–crude oil systems: Insights from explainable artificial intelligence

AbstractMiscible gas injection techniques, such as nitrogen injection, are among the attractive enhanced oil recovery (EOR) techniques for improving oil recovery factors in oil reservoirs. A key challenge in implementing these techniques is accurately determining the minimum miscibility pressure (MMP). While laboratory experiments offer reliable results, they are costly and time‐consuming, and existing empirical correlations often have moderate accuracy, which limits their practical use. In this study, robust ensemble methods, namely light gradient boosting machine (LightGBM), extra trees (ET), and categorical boosting (CatBoost), were implemented for modelling MMP in pure nitrogen and gas mixtures containing nitrogen–crude oil systems. An extensive experimental database involving 164 data points was used to elaborate on the predictive models. The findings revealed that the proposed ensemble methods achieved outstanding accuracy in training and test datasets, with ET consistently outperforming the other models. The ET model provided the most consistent MMP predictions with a total root mean square error (RMSE) of only 0.3197 MPa and a determination coefficient of 0.9976. Additionally, the ET model exhibited very small RMSE values across a broad range of operational conditions. Furthermore, the Shapley additive explanations (SHAP) method further validated the interpretability of the ET model, allowing for clear insights into the impact of input features. This study underlines the significant potential of machine learning to enhance MMP prediction in pure nitrogen and gas mixtures containing nitrogen–crude oil systems, thereby aiding in the appropriate design of this kind of EOR process and supporting better decision‐making in reservoir management.

Research on Park Perception and Understanding Methods Based on Multimodal Text–Image Data and Bidirectional Attention Mechanism

Parks are an important component of urban ecosystems, yet traditional research often relies on single-modal data, such as text or images alone, making it difficult to comprehensively and accurately capture the complex emotional experiences of visitors and their relationships with the environment. This study proposes a park perception and understanding model based on multimodal text–image data and a bidirectional attention mechanism. By integrating text and image data, the model incorporates a bidirectional encoder representations from transformers (BERT)-based text feature extraction module, a Swin Transformer-based image feature extraction module, and a bidirectional cross-attention fusion module, enabling a more precise assessment of visitors’ emotional experiences in parks. Experimental results show that compared to traditional methods such as residual network (ResNet), recurrent neural network (RNN), and long short-term memory (LSTM), the proposed model achieves significant advantages across multiple evaluation metrics, including mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), and the coefficient of determination (R2). Furthermore, using the SHapley Additive exPlanations (SHAP) method, this study identified the key factors influencing visitors’ emotional experiences, such as “water”, “green”, and “sky”, providing a scientific basis for park management and optimization.

Innovative colorimetric thermal study of methylcellulose hydrogel via smartphone imaging

A novel colorimetric analysis of methylcellulose (MC) hydrogel was conducted using a standard smartphone camera to measure its thermo-optical properties. As demonstrated for the first time, the temperature of MC gels was directly determined from photographs by exploiting a unique one-to-one correlation between temperature and mean pixel intensity in the blue channel of RGB images, all color channels showed strong hysteresis. Two innovative procedures for gelation assessment are introduced: variance analysis and histogram analysis with normal distribution fitting. The variance analysis confirms known gelation related temperatures, validating the effectiveness of the new method. Histogram analysis reveals a significant increase in RMSE (root mean square error) near gelation related points, offering a new indicator for gelation status. This methodology underscores the untapped potential of colorimetry to extract valuable data from hydrogels in general and MC gel in particular.

Impact of COVID-19 policy changes on tropospheric NO₂ in the Yangtze River Basin: insights from GF-5 02 EMI-II observations

ABSTRACT This study evaluates the performance of the EMI-II (Environmental trace gases Monitoring Instrument-II) sensor in retrieving tropospheric NO₂ vertical column densities (VCDs) and investigates NO₂ concentration trends in the Yangtze River Basin following significant COVID-19 policy adjustments. Using comparative analyses with TROPOMI (TROPospheric Ozone Monitoring Instrument) and GEMS (Geostationary Environment Monitoring Spectrometer) products, EMI-II demonstrated robust capabilities in capturing NO₂ spatial distributions over key regions such as the Yangtze River Delta (YRD), North China Plain, and Sichuan-Chongqing urban clusters, with spatial correlation coefficients (r) reaching 0.94 and root mean square error (RMSE) as low as 1.98×10¹⁵ molecules/cm² in monthly comparisons, indicating strong agreement with established datasets. Following COVID-19 restrictions’ relaxation in December 2022 and the Spring Festival in January 2023, NO₂ levels in the Yangtze River Basin decreased by 26.94%, with specific declines of 41.90% in the YRD, 28.44% in central urban clusters, and 21.16% in the Sichuan–Chongqing region. These changes were driven by reduced vehicular emissions, industrial activity, and population mobility during the holiday period. The results underscore the interplay between anthropogenic activity, policy measures, and regional air quality, affirming EMI-II’s potential as a reliable tool for monitoring atmospheric pollutants and analyzing environmental responses to societal changes.

Precipitation Spatio-Temporal Forecasting in China via DC-CNN-BiLSTM

Accurate and reliable precipitation prediction remains a significant challenge due to an incomplete understanding of regional meteorological dynamics and limitations in forecasting routine weather events. To overcome these challenges, we propose a novel model, DC-CNN-BiLSTM, which integrates a dilation causal convolutional neural network (DC-CNN) with a Bidirectional Long Short-Term Memory (BiLSTM) network. The DC-CNN component, by fusing causal and dilated convolutions, extracts multi-scale spatial features from time series data. In parallel, the BiLSTM module leverages bidirectional memory cells to capture long-term temporal dependencies. This integrated approach effectively links localized meteorological inputs with broader hydrological responses. Experimental evaluation demonstrates that the DC-CNN-BiLSTM model significantly outperforms traditional models. Specifically, the model improves the Root Mean Square Error (RMSE) by 9.05% compared to ConvLSTM and by 32.3% compared to ConvGRU, particularly in forecasting medium- to long-term precipitation. In conclusion, our results validate the benefits of incorporating advanced spatio-temporal feature extraction techniques for precipitation forecasting, ultimately improving disaster preparedness and resource management.