Benchmark Model Research Articles

Sea Surface Temperature Prediction Using ConvLSTM-Based Model with Deformable Attention

Sea surface temperature (SST) prediction has received increasing attention in recent years due to its paramount importance in the various fields of oceanography. Existing studies have shown that neural networks are particularly effective in making accurate SST predictions by efficiently capturing spatiotemporal dependencies in SST data. Among various models, the ConvLSTM framework is notably prominent. This model skillfully combines convolutional neural networks (CNNs) with recurrent neural networks (RNNs), enabling it to simultaneously capture spatiotemporal dependencies within a single computational framework. To overcome the limitation that CNNs primarily capture local spatial information, in this paper we propose a novel model named DatLSTM that integrates a deformable attention transformer (DAT) module into the ConvLSTM framework, thereby enhancing its ability to process more complex spatial relationships effectively. Specifically, the DAT module adaptively focuses on salient features in space, while ConvLSTM further captures the temporal dependencies of spatial correlations in the SST data. In this way, DatLSTM can adaptively capture complex spatiotemporal dependencies between the preceding and current states within ConvLSTM. To evaluate the performance of the DatLSTM model, we conducted short-term SST forecasts in the Bohai Sea region with forecast lead times ranging from 1 to 10 days and compared its efficacy against several benchmark models, including ConvLSTM, PredRNN, TCTN, and SwinLSTM. Our experimental results show that the proposed model outperforms all of these models in terms of multiple evaluation metrics short-term SST prediction. The proposed model offers a new predictive learning method for improving the accuracy of spatiotemporal predictions in various domains, including meteorology, oceanography, and climate science.

Machine Learning-Based Optimization Models for Defining Storage Rules in Maritime Container Yards

This paper proposes an integrated approach to define the best consignment strategy for storing containers in an export yard of a maritime terminal. The storage strategy identifies the rules for grouping homogeneous containers, which are defined simultaneously with the assignment of each group of containers to the available blocks (bay-locations) in the yard. Unlike recent literature, this study focuses specifically on weight classes and their respective limits when establishing the consignment strategy. Another novel aspect of this work is the integration of a data-driven algorithm and operations research. The integrated approach is based on unsupervised learning and optimization models and allows us to solve large instances within a few seconds. Results obtained by spectral clustering are treated as input datasets for the optimization models. Two different formulations are described and compared: the main difference lies in how containers are assigned to bay-locations, shifting from a time-consuming individual container assignment to the assignment of groups of containers, which offers significant advantages in computational efficiency. Experimental tests are organized into three campaigns to evaluate the following: (i) The computational time and solution quality (i.e., space utilization) of the proposed models; (ii) The performance of these models against a benchmark model; (iii) The practical effectiveness of the proposed solution approach.

Automated Evaluation Method for Risk Behaviors of Quay Crane Operators at Ports Using Virtual Reality

Currently, the operational risk assessment of quay crane operators at ports relies on manual evaluations based on experience, but this method lacks objectivity and fairness. As port throughput continues to grow, the port accident rate has also increased, making it crucial to scientifically evaluate the risk behaviors of operators and improve their safety awareness. This paper proposes an automated evaluation method based on a Deep Q-Network (DQN) to assess the risk behaviors of quay crane operators in virtual scenarios. A risk simulation module has been added to the existing automated quay crane remote operation simulation system to simulate potential risks during operations. Based on the collected data, a DQN-based benchmark model reflecting the operational behaviors and decision-making processes of skilled operators has been developed. This model enables a quantitative evaluation of operators’ behaviors, ensuring the objectivity and accuracy of the assessment process. The experimental results show that, compared with traditional manual scoring methods, the proposed method is more stable and objective, effectively reducing subjective biases and providing a reliable alternative to conventional manual evaluations. Additionally, this method enhances operators’ safety awareness and their ability to handle risks, helping them identify and avoid risks during actual operations, thereby ensuring both operational safety and efficiency.

Dynamic Testing and Finite Element Model Adjustment of the Ancient Wooden Structure Under Traffic Excitation

In situ dynamic testing is conducted to study the dynamic characteristics of the wooden structure of the North House main hall. The velocity response signals on the measurement points are obtained and analyzed using the self-interaction spectral method and stochastic subspace method, yielding natural frequencies, mode shapes, and damping ratios. This study reveals that the natural frequencies and damping ratios are highly consistent between the two methods. Therefore, to eliminate errors, the average of the results from both modal identification methods is taken as the final measured modal parameters of the structure. The natural frequencies of the first and second order in the X direction were 2.097 Hz and 3.845 Hz and in the Y direction were 3.955 Hz and 5.701 Hz. The modal frequency in the Y direction of the structure exceeds that in the X direction. Concurrently, a three-dimensional finite element model was established using ANSYS 2021R1, considering the semi-rigid properties of mortise–tenon connections, and validated based on in situ dynamic testing. The sensitivity analysis indicates adjustments to parameters such as beam–column elastic modulus, tenon–mortise joint stiffness, and roof mass for finite element model refinement. Modal parameter calculations from the corrected finite element model closely approximate the measured modal results, with maximum errors of 9.41% for the first two frequencies, both within 10% of the measured resonant frequencies. The adjusted finite element model closely matches the experimental results, serving as a benchmark model for the wooden structure of North House main hall. The validation confirms the rationality of the benchmark finite element model, providing valuable insights into ancient timber structures along transportation routes.

A Spatial-Temporal Aggregated Graph Neural Network for Docked Bike-Sharing Demand Forecasting

Predicting the number of rented and returned bikes at each station is crucial for operators to proactively manage shared bike relocation. Although existing research has proposed spatial-temporal prediction models that significantly advance traffic prediction, these models often neglect the unique characteristics of shared bike systems (BSS). Spatially, the entire bike-sharing system (BSS) experiences peak activity during morning and evening rush hours, whereas, during other periods, activity is localized to local stations, with some recording no rides, highlighting the need to distinguish between global and local spatial information across different times. Temporally, the historical riding records for each station exhibit non-stationary patterns, necessitating the analysis of both global trends and local fluctuations. Existing Graph Neural Network (GNN) approaches to predicting shared bike demand primarily capture static spatial-temporal data and fail to account for the dynamic nature of bike flows. Moreover, these studies focus on global spatial-temporal information without considering local nuances, making it challenging to capture spatiotemporal dynamics in fluctuating BSS. To address these challenges, we introduce the Spatial-Temporal Aggregated Graph Neural Network (STAGNN). Our model first constructs a dynamic adjacent matrix to describe the evolving connections between stations, followed by local and global information layers to capture spatial-temporal information from large-scale shared bike networks accurately. Our methodology has been validated through experiments on four real-world datasets, comparing it against benchmark models to demonstrate superior prediction accuracy. Additionally, we conduct extended experiments on four datasets during the morning and evening rush hours, and the results also affirm the efficacy of the STAGNN in enhancing prediction performance.

Optimised Deep Learning for Time-Critical Load Forecasting Using LSTM and Modified Particle Swarm Optimisation

Short-term electric load forecasting is critical for power system planning and operations due to demand fluctuations driven by variable energy resources. While deep learning-based forecasting models have shown strong performance, time-sensitive applications require improvements in both accuracy and convergence speed. To address this, we propose a hybrid model that combines long short-term memory (LSTM) with a modified particle swarm optimisation (mPSO) algorithm. Although LSTM is effective for nonlinear time-series predictions, its computational complexity increases with parameter variations. To overcome this, mPSO is used for parameter tuning, ensuring accurate forecasting while avoiding local optima. Additionally, XGBoost and decision tree filtering algorithms are incorporated to reduce dimensionality and prevent overfitting. Unlike existing models that focus mainly on accuracy, our framework optimises accuracy, stability, and convergence rate simultaneously. The model was tested on real hourly load data from New South Wales and Victoria, significantly outperforming benchmark models such as ENN, LSTM, GA-LSTM, and PSO-LSTM. For NSW, the proposed model reduced MSE by 91.91%, RMSE by 94.89%, and MAPE by 74.29%. In VIC, MSE decreased by 91.33%, RMSE by 95.73%, and MAPE by 72.06%, showcasing superior performance across all metrics.

VS-LTGARCHX: A Flexible Variable Selection in Log-TGARCHX Models

Abstract The log-TGARCHX model is less restrictive in terms of the inclusion of exogenous variables and asymmetry lags compared to the GARCHX model. Nevertheless, adding less (or more) covariates than necessary may lead to under- or overfitting, respectively. In this context, we propose a new algorithm, called VS-LTGARCHX, which incorporates a variable selection procedure into the log-TGARCHX estimation process. Furthermore, the VS-LTGARCHX algorithm is applied to extremely volatile BTC markets using 42 conditioning variables. Interestingly, our results show that the VS-LTGARCHX models outperform benchmark models, namely the log-GARCH(1,1) and log-TGARCHX(1,1) models, in one-step-ahead forecasting.

Hybrid modeling approaches for agricultural commodity prices using CEEMDAN and time delay neural networks

Improving the forecasting accuracy of agricultural commodity prices is critical for many stakeholders namely, farmers, traders, exporters, governments, and all other partners in the price channel, to evade risks and enable appropriate policy interventions. However, the traditional mono-scale smoothing techniques often fail to capture the non-stationary and non-linear features due to their multifarious structure. This study has proposed a CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise)-TDNN (Time Delay Neural Network) model for forecasting non-linear, non-stationary agricultural price series. This study has evaluated its suitability in comparison with the other three major EMD (Empirical Mode Decomposition) variants (EMD, Ensemble EMD and Complementary Ensemble EMD) and the benchmark (Autoregressive Integrated Moving Average, Non-linear Support Vector Regression, Gradient Boosting Machine, Random Forest and TDNN) models using monthly wholesale prices of major oilseed crops in India. Outcomes from this investigation reflect that the CEEMDAN-TDNN hybrid models have outperformed all other forecasting models on the basis of evaluation metrics under consideration. For the proposed model, an average improvement of RMSE (Root Mean Square Error), Relative RMSE and MAPE (Mean Absolute Percentage Error) values has been observed to be 20.04%, 19.94% and 27.80%, respectively over the other EMD variant-based counterparts and 57.66%, 48.37% and 62.37%, respectively over the other benchmark stochastic and machine learning models. The CEEMD-TDNN and CEEMDAN-TDNN models have demonstrated superior performance in predicting the directional changes of monthly price series compared to other models. Additionally, the accuracy of forecasts generated by all models has been assessed using the Diebold-Mariano test, the Friedman test, and the Taylor diagram. The results confirm that the proposed hybrid model has outperformed the alternative models, providing a distinct advantage.

Assessing volatility persistence in fractional Heston models with self-exciting jumps

We derive a new fractional Heston model with self-exciting jumps. We study volatility persistence and demonstrate that the quadratic variation necessarily exhibits less memory than the integrated variance, which preserves the degree of long-memory of the instantaneous volatility. Focusing on realized volatility measures, we find that traditional long-memory estimators are dramatically downward biased, in particular for low-frequency intraday sampling. Conveniently, our Monte Carlo experiments reveal that some noise-robust local Whittle-type estimators offer good finite sample properties. We apply our theoretical results in a risk forecasting study and show that our frequency-domain forecasting procedure outperforms the traditional benchmark models.

Electric Vehicle Charging Station Recommendations Considering User Charging Preferences Based on Comment Data

User preferences are important for electric vehicle charging station (EVCS) recommendations, but they have not been deeply analyzed. Therefore, in this study, user charging preferences are identified and applied to EVCS recommendations using a hybrid model that integrates LightGBM and singular value decomposition (SVD). In the model, LightGBM is used to predict user ratings according to users’ comments regarding charging orders, and the feature importance reported by each user is output. Then, a co-occurrence matrix between users and charging stations (EVCSs) is constructed and decomposed using SVD. Based on the decomposed results, the final evaluated scores of each user for EVCSs can be calculated. Upon ranking the EVCSs according to the scores, the EVCS recommendation results are obtained, taking into account the users’ charging preferences. The sample data consist of 28,306 orders from 508 users at 241 charging stations in Linyi, Shandong, China. The experimental results show that the proposed hybrid model outperforms the benchmark models in terms of precision, recall, and F1 score, and its F1 score can be increased by 96% compared with that of the traditional item-based collaborative filtering method with charging counts for EVCS recommendations.

An Enhanced Deep Learning Model for Effective Crop Pest and Disease Detection

Traditional machine learning methods struggle with plant pest and disease image recognition, particularly when dealing with small sample sizes, indistinct features, and numerous categories. This paper proposes an improved ResNet34 model (ESA-ResNet34) for crop pest and disease detection. The model employs ResNet34 as its backbone and introduces an efficient spatial attention mechanism (effective spatial attention, ESA) to focus on key regions of the images. By replacing the standard convolutions in ResNet34 with depthwise separable convolutions, the model reduces its parameter count by 85.37% and its computational load by 84.51%. Additionally, Dropout is used to mitigate overfitting, and data augmentation techniques such as center cropping and horizontal flipping are employed to enhance the model’s robustness. The experimental results show that the improved algorithm achieves an accuracy, precision, and F1 score of 87.09%, 87.14%, and 86.91%, respectively, outperforming several benchmark models (including AlexNet, VGG16, MobileNet, DenseNet, and various ResNet variants). These findings demonstrate that the proposed ESA-ResNet34 model significantly enhances crop pest and disease detection.

A long short-term memory enhanced realized conditional heteroskedasticity model

This paper examines the potential of using realized volatility measures for capturing financial markets’ uncertainty. Earlier studies show the usefulness of the high-frequency data based Generalized AutoRegressive Conditional Heteroskedasticity (RealGARCH) model for enhancing volatility forecasting accuracy; however, this model focuses only on linear and short-term dependencies of realized volatility measures on the underlying volatility. Recognizing the critical economic implications of this limitation, the long short-term memory neural network is integrated into RealGARCH, aiming to explore the full impact of realized volatility on volatility modeling and forecasting via capturing the nonlinear and long-term effects. A comprehensive empirical study using 31 indices from 2004 to 2021 is conducted. The results demonstrate that our proposed framework achieves superior in-sample and out-of-sample performance compared to several benchmark models. Importantly, it retains interpretability and effectively adapts to the stylized facts observed in volatility, emphasizing its significant potential for enhancing economic decision-making and risk management.

Predicting the volatility of major energy commodity prices: The dynamic persistence model

Time variation and persistence are crucial properties of volatility that are often studied separately in energy volatility forecasting models. Here, we propose a novel approach that allows shocks with heterogeneous persistence to vary smoothly over time, and thus model the two together. We argue that this is important because such dynamics arise naturally from the dynamic nature of shocks in energy commodities. We identify such dynamics from the data using localised regressions and build a model that significantly improves volatility forecasts. Such forecasting models, based on a rich persistence structure that varies smoothly over time, outperform state-of-the-art benchmark models and are particularly useful for forecasting over longer horizons.

Predicting Continuous Locomotion Modes via Multidimensional Feature Learning From sEMG.

Walking-assistive devices require adaptive control methods to ensure smooth transitions between various modes of locomotion. For this purpose, detecting human locomotion modes (e.g., level walking or stair ascent) in advance is crucial for improving the intelligence and transparency of such robotic systems. This study proposes Deep-STF, a unified end-to-end deep learning model designed for integrated feature extraction in spatial, temporal, and frequency dimensions from surface electromyography (sEMG) signals. Our model enables accurate and robust continuous prediction of nine locomotion modes and 15 transitions at varying prediction time intervals, ranging from 100 to 500 ms. Experimental results showcased Deep-STP's cutting-edge prediction performance across diverse locomotion modes and transitions, relying solely on sEMG data. When forecasting 100 ms ahead, Deep-STF achieved an improved average prediction accuracy of 96.60%, outperforming seven benchmark models. Even with an extended 500ms prediction horizon, the accuracy only marginally decreased to 93.22%. The averaged stable prediction times for detecting next upcoming transitions spanned from 31.47 to 371.58 ms across the 100-500 ms time advances. Although the prediction accuracy of the trained Deep-STF initially dropped to 71.12% when tested on four new terrains, it achieved a satisfactory accuracy of 92.51% after fine-tuning with just 5 trials and further improved to 96.27% with 15 calibration trials. These results demonstrate the remarkable prediction ability and adaptability of Deep-STF, showing great potential for integration with walking-assistive devices and leading to smoother, more intuitive user interactions.

Probabilistic oil price forecasting with a variational mode decomposition-gated recurrent unit model incorporating pinball loss

Prediction methods have garnered significant attention in intelligent decision-making. Most existing approaches to predicting crude oil prices prioritize accuracy and stability while providing precise prediction intervals (PIs) that can offer valuable insights. Thus far, we introduce a novel hybrid model to forecast future crude oil prices. Our approach leverages the variational mode decomposition (VMD) to simplify the complexity of the original time series, yielding a set of subseries. These subseries are then modeled using a deep neural network architecture called a gated recurrent unit (GRU). To address the prediction uncertainty, we employed the pinball loss function rather than the mean square error (MSE) to guide the proposed VMD-GRU. This adaptation extends the traditional GRU-based point forecasting to probabilistic forecasting by estimating quantiles. We evaluated our proposed model on a well-established crude oil price series by conducting both single- and multi-step-ahead forecasting analyses. Our findings underscore the efficacy of the combined model, demonstrating its superior predictive performance compared to benchmark models.

Short-term inbound passenger flow prediction at high-speed railway stations considering the departure passenger arrival pattern

Accurate prediction of short-term inbound passenger flow at high-speed railway stations is of great significance for the refined operation of stations, the formulation of emergency plans, and the provision of intelligent services. The arrival of passengers traveling on the same train at the same station shows a similar pattern, which is called the departure passenger arrival pattern (DPAP). The short-term inbound passenger flow at the station is composed of the short-term inbound passenger flow of all waiting trains within the same period. Inspired by this, this paper develops an ensemble prediction model based on the time series decomposition modeling strategy to introduce the DPAP to the short-term inbound passenger flow prediction at stations. Firstly, we propose a new framework for studying the DPAP to calculate the fitted station short-term inbound passenger flow, which is only affected by the DPAP. During this process, we find that 7minutes is the optimal time granularity. Secondly, based on the singular spectrum analysis, we prove that the DPAP is the determining factor affecting the station short-term inbound passenger flow. Finally, we propose an ensemble prediction model that considers the DPAP to achieve short-term inbound passenger flow prediction at stations. The model consists of two parts: the deterministic and stochastic components prediction, where the former is predicted by the fitted station short-term inbound passenger flow, and the latter is achieved by the combination of historical stochastic components and weather type with the help of the Seq2Seq model based on time attention mechanism. Using real inbound passenger flow data, we compare the proposed model with 13 benchmark models and the results show that under different training and prediction steps, our model achieves optimal prediction performance, whether in all-day period and the busiest period of the station. Through further ablation experiments, it has been proven that the introduction of the DPAP effectively improves the prediction accuracy. Our model can provide scientific support for the intelligent operation of stations and the refined management of passenger flow.

Hybrid deep learning model with VMD-BiLSTM-GRU networks for short-term traffic flow prediction

Accelerating urbanization and the rapid development of intelligent transportation systems have rendered short-term traffic flow prediction an important research field. Accurate prediction of traffic flow is beneficial for the optimization of traffic planning, improvement of road utilization, reduction of traffic congestion, and reduction in the incidence of traffic accidents. However, data pertaining to traffic flow are typically influenced by a multitude of factors, resulting in data that exhibit a considerable degree of nonlinearity and complexity. To address the issue of noise in raw traffic flow data, this study proposes a hybrid model that combines variational modal decomposition (VMD), a bidirectional long short-term memory network (BiLSTM), and a gated recurrent unit (GRU) for short-term traffic flow prediction. To validate the effectiveness of the model, an experimental validation was conducted based on traffic flow data from UK highways, and the performance of the model was compared with common benchmark models. The experimental results demonstrate that the proposed method yields superior prediction results in terms of mean absolute error, coefficient of determination, and root-mean-square error compared to existing prediction techniques, thereby substantiating its efficacy in short-term traffic flow prediction.

Rearrangement and breakup amplitudes from the solution of Faddeev-AGS equations by pseudo-state discretization of the two-particle continuum

The AGS equations for rearrangement transition operators in the three-particle problem are turned into a set of effective multi-channel two-body equations using the pseudo-state discretization of the two-particle resolvent. The resulting effective equations are LS-type integral equations in the spectator degrees of freedom, much like the LS equations of multichannel inelastic scattering. In particular, the effective potential matrix is real, energy-independent and non-singular, while the propagator matrix has only simple poles. Difficulties associated with the moving singularities of the effective potential matrix in the usual separable-T approach to AGS equations are avoided. After regularization of the kernel via subtraction procedures well known from two-particle scattering, the set of coupled LS-type equations in the spectator momenta are solved rather straightforwardly by the Nyström method. Solutions of effective two-body equations are then used to calculate the breakup amplitudes using the well-known relationship between rearrangement and breakup amplitudes. Calculations using a local momentum-space basis on a benchmark model of the n+d collision show that rather accurate results for elastic and breakup amplitudes can be obtained with rather small bases.

BrainSim: AI-Driven Brain MRI Simulation for Alzheimer’s Disease Diagnosis

Alzheimer's Disease, a progressive neurodegenerative disorder with no cure, presents a formidable challenge to global healthcare. The simulation of brain states, encompassing both rejuvenation and aging, aids clinicians in understanding disease progression and early detection. Recent advances in deep adversarial neural networks, effective in generating age-varied facial images, are now applied to brain MRI simulations across different time states. However, existing methods for brain MRI synthesis predominantly focus on aging simulation, lacking rejuvenation capabilities, and relying heavily on longitudinal data. Access to rejuvenated brain scans is essential as it can help researchers track the structural changes and biomarkers in the early stage of AD, facilitating AD research and personalized treatments. Additionally, previous methods are often 2D-based, failing to capture complex 3D spatial information obtained by 3D scans. To overcome these limitations, I propose BrainSim, a novel longitudinal brain MRI synthesis framework. Specifically, BrainSim leverages a bidirectional cycle-consistent generative neural network to generate rejuvenated and aged MRI scans simultaneously using cross-sectional scans only. BrainSim incorporates a 3D conditional U-Net as its fundamental model, facilitating the direct generation of 3D MRI scans conditioned on the subject's health state and sex. BrainSim was evaluated against benchmark models using longitudinal data and a pre-trained age predictor to assess the quality of generated images. Experiments using the ADNI dataset show BrainSim's state-of-the-art performance in brain aging and rejuvenation tasks, marking significant advancements in MRI synthesis for Alzheimer's research and drug development.

Hierarchical Attention and Contextual Embedding for Robust Sarcasm Detection

Objective: The study aims to develop a technique for sarcasm detection in social media text to enrich the sentiment analysis. Methods: The methodology used in the present research is as follows: emotion-embedded vectors for capturing the emotional content of the text, dynamic contextual modulation for the adaption of the model to the given context, and Hierarchical attention mechanism for segmentation of the text at the different level of abstraction. Findings: Evaluation with the test set proved that the proposed model achieved accuracy rates 89% higher than the benchmark models. Including these several complicated methods helped achieve greater accuracy and F1 measures, contributing to sarcasm detection efficiency. Novelty: This approach ensures that vital sarcastic alerts are detected, providing a better text analysis and increasing sentiment analysis's accuracy and robustness. Keywords: Sarcasm detection, Sentiment analysis, Natural language processing, Deep learning, Opinion mining