Memory Model Research Articles

Machine learning-based modeling of surface water temperature dynamics in arctic lakes.

Lake surface-water temperature (LSWT) regulates physical and biochemical processes in lakes. Therefore, understanding the LSWT dynamics is important, especially in Arctic zone since the region is experiencing a warming rate that is greater than the Earth's average. However, regular measurements of LSWT in the remote Arctic lakes always face difficulties or cannot be done by satellites accurately due to the cloud cover and their limited spatiotemporal resolution. Here, we used a historically rich data (1960-2023) to develop four machine learning-based algorithms for the daily LSWT modeling in Lake Inari, situated in Arctic zone, using the air-temperature data. Our results showed that both air-temperature (0.030 °C/yr) and LSWT (0.023m °C/yr) were warming with a rate faster than those in the globe. The long-short-term memory model, with the coefficients of determination varied from 0.96 to 0.98, outperformed other algorithms in modeling of the daily LSWT dynamics in Lake Inari, followed by both support vector regression and neural network tools, and random forest model. As the air-temperature data are widely accessible through synoptic stations and remote sensing techniques, our suggested models can be simply adopted for other Arctic lakes, where the local water-temperature data are often lacking or contain large windows of missing data due to harsh atmospheric conditions and equipment failure.

MetaFollower: Adaptable personalized autonomous car following

Car-following (CF) modeling, a fundamental component in microscopic traffic simulation, has attracted increasing research interest in recent decades. In this study, we propose an adaptable personalized car-following framework —– MetaFollower, by leveraging the power of meta-learning. Specifically, we first utilize Model-Agnostic Meta-Learning (MAML) to extract common driving knowledge from various CF events. Afterward, the pre-trained model can be fine-tuned on new drivers with only a few CF trajectories to achieve personalized CF adaptation. We additionally combine Long Short-Term Memory (LSTM) and Intelligent Driver Model (IDM) to reflect temporal heterogeneity with high interpretability. MetaFollower can accurately capture and simulate the intricate dynamics of car-following behavior while considering the unique driving styles of individual drivers. We demonstrate the versatility and adaptability of MetaFollower by showcasing its ability to adapt to new drivers with limited training data quickly. To evaluate the performance of MetaFollower, we conduct rigorous experiments comparing it with both data-driven and physics-based models. The results reveal that our proposed framework outperforms baseline models in predicting car-following behavior with higher accuracy and safety. To the best of our knowledge, this is the first car-following model aiming to achieve fast adaptation by considering both driver and temporal heterogeneity based on meta-learning.

MC-VMD-CNN-BiLSTM short-term wind power prediction considering rolling error correction

Abstract Wind energy is a clean and renewable source that has the potential to alleviate the global fossil fuel crisis and environmental pollution by generating electricity. However, accurately predicting wind energy output remains challenging due to its inherent uncertainty. To enhance the accuracy of wind power prediction, a short-term wind power forecasting method for power systems, MC-VMD-CNN-BiLSTM, is proposed, which considers error rolling correction. The method begins with feature selection and outlier handling using the quadrature method. Then, wind power data is decomposed into multiple sub-sequences using the Variational Mode Decomposition (VMD) technique to reduce the raw volatility of wind power. Then, a Convolutional Neural Network (CNN) followed by a Bidirectional Long Short-Term Memory (BiLSTM) model is used for wind power prediction. Finally, the proposed method utilizes the Monte Carlo method for rolling error correction by using known errors from previous time frames to correct subsequent predictions. The MC-VMD-CNN-BiLSTM proposed in this paper considering error rolling correction is compared with ELM, SVM, PSO-BP and ARIMA models through an example analysis of the data of a city, and the proposed model in this paper reduces 61.78%, 50.35%, 62.30% and 73.05% in the NRMSE index in the spring as an example, respectively. The results show that the prediction model proposed in this paper has higher prediction accuracy compared with the traditional prediction model.

A flexible deep learning framework for liver tumor diagnosis using variable multi-phase contrast-enhanced CT scans

BackgroundLiver cancer is a significant cause of cancer-related mortality worldwide and requires tailored treatment strategies for different types. However, preoperative accurate diagnosis of the type presents a challenge. This study aims to develop an automatic diagnostic model based on multi-phase contrast-enhanced CT (CECT) images to distinguish between hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (ICC), and normal individuals.MethodsWe designed a Hierarchical Long Short-Term Memory (H-LSTM) model, whose core components consist of a shared image feature extractor across phases, an internal LSTM for each phase, and an external LSTM across phases. The internal LSTM aggregates features from different layers of 2D CECT images, while the external LSTM aggregates features across different phases. H-LSTM can handle incomplete phases and varying numbers of CECT image layers, making it suitable for real-world decision support scenarios. Additionally, we applied phase augmentation techniques to process multi-phase CECT images, improving the model’s robustness.ResultsThe H-LSTM model achieved an overall average AUROC of 0.93 (0.90, 1.00) on the test dataset, with AUROC for HCC classification reaching 0.97 (0.93, 1.00) and for ICC classification reaching 0.90 (0.78, 1.00). Comprehensive validation in scenarios with incomplete phases was performed, with the H-LSTM model consistently achieving AUROC values over 0.9.ConclusionThe proposed H-LSTM model can be employed for classification tasks involving incomplete phases of CECT images in real-world scenarios, demonstrating high performance. This highlights the potential of AI-assisted systems in achieving accurate diagnosis and treatment of liver cancer. H-LSTM offers an effective solution for processing multi-phase data and provides practical value for clinical diagnostics.

RCLNet: an effective anomaly-based intrusion detection for securing the IoMT system.

The Internet of Medical Things (IoMT) has revolutionized healthcare with remote patient monitoring and real-time diagnosis, but securing patient data remains a critical challenge due to sophisticated cyber threats and the sensitivity of medical information. Traditional machine learning methods struggle to capture the complex patterns in IoMT data, and conventional intrusion detection systems often fail to identify unknown attacks, leading to high false positive rates and compromised patient data security. To address these issues, we propose RCLNet, an effective Anomaly-based Intrusion Detection System (A-IDS) for IoMT. RCLNet employs a multi-faceted approach, including Random Forest (RF) for feature selection, the integration of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) models to enhance pattern recognition, and a Self-Adaptive Attention Layer Mechanism (SAALM) designed specifically for the unique challenges of IoMT. Additionally, RCLNet utilizes focal loss (FL) to manage imbalanced data distributions, a common challenge in IoMT datasets. Evaluation using the WUSTL-EHMS-2020 healthcare dataset demonstrates that RCLNet outperforms recent state-of-the-art methods, achieving a remarkable accuracy of 99.78%, highlighting its potential to significantly improve the security and confidentiality of patient data in IoMT healthcare systems.

Research on input parameter optimization for NOx deep learning prediction model

Optimizing the input parameters for NOx prediction model is instrumental in enhancing the precision and efficacy of the prediction model. This paper focuses on the input variables of the data-driven diesel engine NOx prediction model. According to the diesel engine NOx generation mechanism, at first relevant variables are selected, then the similarity method and parameter sensitivity method are compared to optimize the input variables. Firstly, this paper conducted a real-road emission test of a heavy-duty diesel vehicle, and proposed a data-driven deep learning model for predicting diesel engine NOx emissions based on the experimental data. The experimental data were processed using an Attention-Mechanism based Convolutional Neural Networks-Long Short-Term Memory (AM-CNN-LSTM) model. Subsequently, input parameters were selected according to the sensitivity weights assigned to each parameter in the preliminary model. This approach refined the modeling data and excluded variables not pertinent to NOx emission prediction. The AM-CNN-LSTM model prediction performance has been improved with 2% increase of model R2 as well as the model training time is reduced by 23%, and the types of parameters required to train the model are reduced by 50%. It accomplished this by effectively reducing data dimensions, discarding non-essential variables for NOx prediction, and diminishing computational complexity. Six input parameters were identified as pivotal in reconstructing the modeling data for accurate NOx prediction: engine speed, torque, EGR valve position, exhaust flow rate, air mass flow rate, and oil temperature. In optimizing the NOx prediction model, it is crucial to determine an optimal number of input parameters. Excessive parameters do not enhance model accuracy, while an insufficient number impairs the model’s ability to precisely emulate the NOx generation process, leading to diminished predictive accuracy.

Unveiling the Significance of Individual Level Predictions: A Comparative Analysis of GRU and LSTM Models for Enhanced Digital Behavior Prediction

The widespread use of technology has led to a transformation of human behaviors and habits into the digital space; and generating extensive data plays a crucial role when coupled with forecasting techniques in guiding marketing decision-makers and shaping strategic choices. Traditional methods like autoregressive moving average (ARMA) can-not be used at predicting individual behaviors because we can-not create models for each individual and buy till you die (BTYD) models have limitations in capturing the trends accurately. Recognizing the paramount importance of individual-level predictions, this study proposes a deep learning framework, specifically uses gated recurrent unit (GRU), for enhanced behavior analysis. This article discusses the performance of GRU and long short-term memory (LSTM) models in this framework for forecasting future individual behaviors and presenting a comparative analysis against benchmark BTYD models. GRU and LSTM yielded the best results in capturing the trends, with GRU demonstrating a slightly superior performance compared to LSTM. However, there is still significant room for improvement at the individual level. The findings not only demonstrate the performance of GRU and LSTM models but also provide valuable insights into the potential of new techniques or approaches for understanding and predicting individual behaviors.

Study of the driving factors of the abnormal influenza A (H3N2) epidemic in 2022 and early predictions in Xiamen, China

BackgroundInfluenza outbreaks have occurred frequently these years, especially in the summer of 2022 when the number of influenza cases in southern provinces of China increased abnormally. However, the exact evidence of the driving factors involved in the prodrome period is unclear, posing great difficulties for early and accurate prediction in practical work.MethodsIn order to avoid the serious interference of strict prevention and control measures on the analysis of influenza influencing factors during the COVID-19 epidemic period, only the impact of meteorological and air quality factors on influenza A (H3N2) in Xiamen during the non coronavirus disease 2019 (COVID-19) period (2013/01/01-202/01/24) was analyzed using the distribution lag non-linear model. Phylogenetic analysis of influenza A (H3N2) during 2013–2022 was also performed. Influenza A (H3N2) was predicted through a random forest and long short-term memory (RF-LSTM) model via actual and forecasted meteorological and influenza A (H3N2) values.ResultsTwenty nine thousand four hundred thirty five influenza cases were reported in 2022, accounting for 58.54% of the total cases during 2013–2022. A (H3N2) dominated the 2022 summer epidemic season, accounting for 95.60%. The influenza cases in the summer of 2022 accounted for 83.72% of the year and 49.02% of all influenza reported from 2013 to 2022. Among them, the A (H3N2) cases in the summer of 2022 accounted for 83.90% of all A (H3N2) reported from 2013 to 2022. Daily precipitation(20–50 mm), relative humidity (70–78%), low (≤ 3 h) and high (≥ 7 h) sunshine duration, air temperature (≤ 21 °C) and O3 concentration (≤ 30 µg/m3, > 85 µg/m3) had significant cumulative effects on influenza A (H3N2) during the non-COVID-19 period. The daily values of PRE, RHU, SSD, and TEM in the prodrome period of the abnormal influenza A (H3N2) epidemic (19–22 weeks) in the summer of 2022 were significantly different from the average values of the same period from 2013 to 2019 (P < 0.05). The minimum RHU value was 70.5%, the lowest TEM value was 16.0 °C, and there was no sunlight exposure for 9 consecutive days. The highest O3 concentration reached 164 µg/m3. The range of these factors were consistent with the risk factor range of A (H3N2). The common influenza A (H3N2) variant genotype in 2022 was 3 C.2a1b.2a.1a. It was more accurate to predict influenza A (H3N2) with meteorological forecast values than with actual values only.ConclusionThe extreme weather conditions of sustained low temperature and wet rain may have been important driving factors for the abnormal influenza A (H3N2) epidemic. A low vaccination rate, new mutated strains, and insufficient immune barriers formed by natural infections may have exacerbated this epidemic. Meteorological forecast values can aid in the early prediction of influenza outbreaks. This study can help relevant departments prepare for influenza outbreaks during extreme weather, provide a scientific basis for prevention strategies and risk warnings, better adapt to climate change, and improve public health.

TCRcost: a deep learning model utilizing TCR 3D structure for enhanced of TCR-peptide binding.

Predicting TCR-peptide binding is a complex and significant computational problem in systems immunology. During the past decade, a series of computational methods have been developed for better predicting TCR-peptide binding from amino acid sequences. However, the performance of sequence-based methods appears to have hit a bottleneck. Considering the 3D structures of TCR-peptide complexes, which provide much more information, could potentially lead to better prediction outcomes. In this study, we developed TCRcost, a deep learning method, to predict TCR-peptide binding by incorporating 3D structures. TCRcost overcomes two significant challenges: acquiring a sufficient number of high-quality TCR-peptide structures and effectively extracting information from these structures for binding prediction. TCRcost corrects TCR 3D structures generated by protein structure tools, significantly extending the available datasets. The main and side chains of a TCR structure are separately corrected using a long short-term memory (LSTM) model. This approach prevents interference between the chains and accurately extracts interactions among both adjacent and global atoms. A 3D convolutional neural network (CNN) is designed to extract the atomic features relevant to TCR-peptide binding. The spatial features extracted by the 3DCNN are then processed through a fully connected layer to estimate the probability of TCR-peptide binding. Test results demonstrated that predicting TCR-peptide binding from 3D TCR structures is both efficient and highly accurate with an average accuracy of 0.974 on precise structures. Furthermore, the average accuracy on corrected structures was 0.762, significantly higher than the average accuracy of 0.375 on uncorrected original structures. Additionally, the average root mean square distance (RMSD) to precise structures was significantly reduced from 12.753Å for predicted structures to 8.785Å for corrected structures. Thus, utilizing structural information of TCR-peptide complexes is a promising approach to improve the accuracy of binding predictions.

Prognosis of major bleeding based on residual variables and machine learning for critical patients with upper gastrointestinal bleeding: A multicenter study

BackgroundUpper gastrointestinal bleeding (UGIB) is a significant cause of morbidity and mortality worldwide. This study investigates the use of residual variables and machine learning (ML) models for predicting major bleeding in patients with severe UGIB after their first intensive care unit (ICU) admission. MethodsThe Medical Information Mart for Intensive Care IV and eICU databases were used. Conventional ML and long short-term memory models were constructed using pre-ICU and ICU admission day data to predict the recurrence of major gastrointestinal bleeding. In the models, residual data were utilized by subtracting the normal range from the test result. The models included eight algorithms. Shapley additive explanations and saliency maps were used for feature interpretability. ResultsTwenty-five ML models were developed using data from 2604 patients. The light gradient-boosting machine algorithm model using pre-ICU admission residual data outperformed other models that used test results directly, with an AUC of 0.96. The key factors included aspartate aminotransferase, blood urea nitrogen, albumin, length of ICU admission, and respiratory rate. ConclusionsML models using residuals improved the accuracy and interpretability in predicting major bleeding during ICU admission in patients with UGIB. These interpretable features may facilitate the early identification and management of high-risk patients, thereby improving hemodynamic stability and outcomes.

View-based axiomatic reasoning for the weak memory models PSO and SRA

Weak memory models describe the semantics of concurrent programs in modern multicore architectures. As these semantics deviate from the commonly assumed model of sequential consistency, reasoning techniques like Owicki-Gries-style proof calculi need to be adapted to specific memory models. To avoid having to design a new proof calculus for every new memory model, a uniform approach for axiomatic reasoning has recently been proposed. This approach bases reasoning on memory-model independent axioms about thread views and how they are changed by program actions like reads and writes. It allows to prove program correctness based on axioms only. Such proofs are valid for all memory models instantiating the axioms.In this paper, we study instantiations of the axioms for two memory models, the Partial Store Order (PSO) and the Strong Release Acquire (SRA) model. We see that both models fulfil all but one axiom, a different one though. For PSO, the missing axiom refers to message-passing abilities of memory models; for SRA, the missing axiom refers to the independence of actions on executing threads. We discuss the consequences of these missing axioms and illustrate the reasoning technique on a specific litmus test.

Regionalization Strategy Guided Long Short‐Term Memory Model for Improving Flood Forecasting

ABSTRACTFlood forecasting in data‐scarce catchments is challenging for hydrologists. To address this issue, a regional long short‐term memory model (R‐LSTM) is proposed. Given the diverse physical characteristics of sub‐catchments, this model scalarises the runoff data based on catchment attributes including area, confluence path length, slope and minimum and maximum runoff values, thereby eliminating the local influence and producing a geomorphological‐runoff factor as the model input. To assess the effectiveness of R‐LSTMs for flood forecasting in data‐scarce basins, the Jiaodong Peninsula in China was selected as the study area. The proposed R‐LSTMs are compared with local LSTMs, regional LSTMs that do not use catchment attributes, or regional LSTMs that incorporate catchment attributes in different ways. The results show that R‐LSTMs outperform the benchmarking LSTM models, especially in the simulation of flood peaks. The study indicates the potential of regionalization and the benefit of building the scalarised inputs of runoff data for regional LSTM that consider catchment attributes meticulously. The research findings can provide a reference for flood forecasting in data‐scarce regions.

A Long Short Term Memory Model for Character-based Analysis of DNS Tunneling Detection

DNS tunneling is the attempt to create a hidden tunnel through a domain name service. Such a tunnel would jeopardize the targeted network and open the door for illegal access, control, and data exfiltration. The information security research community showed the variety of techniques that have been proposed to detect the tunnel. The majority of these efforts were relying on machine learning techniques where features of tunneling are considered such as length of DNS query, size, and entropy of the query. However, an additional analysis of the lexical information of the DNS query has been depicted recently and showed remarkable performance. This paper aims to examine the role of Long Short Term Memory (LSTM) model in terms of DNS lexical analysis. Two benchmark datasets related to DNS have been used. In addition, a character mapping mechanism has been used to replace every possible character with an integer number. Consequentially, the mapped representation has been fed into an LSTM model for DNS tunneling detection. Results showed that the proposed method was able to obtain a weighted average F1-score of 98% for both datasets respectively. Such results are competitive in the context of the state of the art and demonstrate the efficacy of the lexical analysis within the DNS tunneling detection task.

Multi-Way adaptive Time Aware LSTM for irregularly collected sequential ICU data

In personalized predictive medicine, accurately modeling a patient’s illness and care processes is essential, given their inherent long-term temporal dependencies. However, electronic medical records contain episodic and irregularly timed data due to patients visiting hospitals based on treatment needs, resulting in unique patterns for each hospital stay. Consequently, when constructing a personalized predictive model, it is crucial to consider these factors in order to accurately capture the patient’s health journey.To address this challenge, we present a novel deep dynamic memory neural network called Multi-Way adaptive Time Aware LSTM (MWTA-LSTM). The primary objective of MWTA-LSTM is to leverage medical records, memorize illness trajectories and care processes, estimate current illness states, and predict future risks, thereby providing a high level of precision and predictive power. To enhance its capabilities, MWTA-LSTM extends the conventional Long Short-Term Memory (LSTM) model in two key ways. Firstly, it incorporates frequency measurement and the most recent observation(last observation) to enhance personalized predictive modeling of patient illnesses, enabling a more accurate understanding of the patient’s condition. Secondly, it parameterizes the cell state to handle irregular timing effectively, utilizing both frequency measurement and elapsed times. Furthermore, the model capitalizes on both to comprehend the impact of interventions on the course of illness on the cell state, facilitating the memorization of illness courses and improving its ability to capture the temporal dynamics of healthcare data, accommodating variations and irregularities in event and observation timing. Lastly, we introduce a novel adaptive pooling strategy that specifically targets and resolves outlier issues that can potentially occur during the analysis of EHR data. By incorporating these features, MWTA-LSTM significantly improves its ability to capture the temporal dynamics of healthcare data, accommodating variations and irregularities in event and observation timing.We validate the effectiveness of our proposed model through empirical experiments conducted on two real-world clinical datasets and three real-world time series datasets. Our results demonstrate the superiority of MWTA-LSTM over current state-of-the-art models and other robust baselines. This showcases the potential of MWTA-LSTM in advancing personalized predictive medicine by offering a more accurate and comprehensive approach to modeling patient health trajectories.

A one-time training machine learning method for general structural topology optimization

Machine learning (ML) methods have found some applications in structural topology optimization. In the existing methods, however, the ML models need to be retrained when the design domains and supporting conditions have been changed, posing a limitation to their wide applications. In this paper, we propose a one-time training ML (OTML) method for general topology optimization, where the self-attention convolutional long short-term memory (SaConvLSTM) model is introduced to update the design variables. An extension–division approach is used to enrich the training sets. By developing a splicing strategy, the training results of a small design space (i.e., a basic cell of either two- or three-dimensions) can be extended to tackling the optimization problem of a large design domain with arbitrary geometric shapes. Using the OTML method, the ML model needs to be trained for only one time, and the trained model can be used directly to solve various optimization problems with arbitrary shapes of design domains, loads, and boundary conditions. In the SaConvLSTM model, the material volume of the post-processed thresholded designs can be precisely controlled, though the control precision of the gray-scale designs might be slightly sacrificed. The effects of model parameters on the computational cost and the result quality are examined. Four examples are provided to demonstrate the high performance of this structural design method. For large-scale optimization problems, the present method can accelerate the structural form-finding process. This study holds a promise in the high-resolution structural form-finding and transdisciplinary computational morphogenesis.

Cross regional online food delivery: Service quality optimization and real-time order assignment

Online food delivery (OFD) represents a rapidly evolving e-business application that leverages cloud computing data centers, playing a crucial role in meeting the demands of urban lifestyles. With diverse order fulfillment features and increasing expectations for service quality, the task of effectively assigning riders for timely long-distance, cross-regional deliveries presents a significant engineering challenge. Previous studies often relied on traditional rider allocation methods that fail to account for varying capacities, or they utilized non-intelligent systems that did not adequately address fluctuating order demands and service delays. In this study, we introduce a robust Mixed Integer Linear Programming (MILP) optimization framework designed to minimize the total service time and delivery cost for cross-regional orders. This framework divides a large OFD area into multiple regions and utilizes both transfer vehicles and riders to optimize deliveries. To enhance the predictive accuracy of our model, we incorporate advanced machine learning techniques. Specifically, we employ the Long Short-Term Memory (LSTM) model to forecast regional order demands accurately, reflecting the dynamic nature of the marketplace. Additionally, Extreme Gradient Boosting (XGBoost) is tailored to dynamically predict travel times from restaurants to customer locations, facilitating more precise scheduling and resource allocation within the MILP framework. These machine learning techniques significantly bolster the MILP framework by providing detailed, accurate predictions that improve decision-making processes and adaptability to real-time conditions. Acknowledging the complexity of this optimization problem, we further enhance our approach by integrating a meta-heuristic algorithm, Adaptive Large Neighbor Search (ALNS), which efficiently assigns orders to the appropriate transfer vehicles and riders within polynomial time. Our Cross Regional Online Food Delivery (XROFD) system is meticulously designed to optimize both customer satisfaction and rider incentives. Simulation experiments confirm that the XROFD system not only reduces service times and delivery costs but also markedly enhances customer satisfaction and provides superior incentives for riders, outperforming existing state-of-the-art methods.

LSTM Model Integrated Remote Sensing Data for Drought Prediction: A Study on Climate Change Impacts on Water Availability in the Arid Region

Climate change is one of the trending terms in the world nowadays due to its profound impact on human health and activity. Extreme drought events and desertification are some of the results of climate change. This study utilized the power of AI tools by using the long short-term memory (LSTM) model to predict the drought index for Anbar Province, Iraq. The data from the standardized precipitation evapotranspiration index (SPEI) for 118 years have been used for the current study. The proposed model employed seven different optimizers to enhance the prediction performance. Based on different performance indicators, the results show that the RMSprop and Adamax optimizers achieved the highest accuracy (90.93% and 90.61%, respectively). Additionally, the models forecasted the next 40 years of the SPEI for the study area, where all the models showed an upward trend in the SPEI. In contrast, the best models expected no increase in the severity of drought. This research highlights the vital role of machine learning models and remote sensing in drought forecasting and the significance of these applications by providing accurate climate data for better water resources management, especially in arid regions like that of Anbar province.

Towards the development of a ‘land-river-lake’ two-stage deep learning model for water quality prediction and its application in a large plateau lake

Accurate prediction of water quality that is essential for effective lake management requires a comprehensive understanding of influential factors in the ’land-river-lake’ continuum. We develop a novel deep learning model employing a Long Short-Term Memory (LSTM) model for predicting water quality, and apply it for a large plateau lake, Lake Dianchi. The model comprises two stages: the first establishes a correlation between land and rivers by predicting nutrient loads, incorporating socioeconomic and meteorological features. The second establishes a correlation between feeding rivers and the lake by predicting nutrient concentrations, taking water quality, meteorological conditions, riverine import, and agricultural Non-Point Source (NPS) loads into consideration. Compared to single-stage deep learning models developed using classical methods, the two-stage model reduces Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE), indicating enhanced prediction accuracy. Shapley Additive Explanation (SHAP) analysis is further employed to identify key contributing features. Precipitation emerges as the primary feature in the first stage, while in the second stage, disparities in the most contributing features for two zones of the lake are revealed due to dam isolation and water diversion projects. In the main zone, the primary predicting features are temperature and agricultural NPS features, while in the dam-isolated zone, Secchi depth is the key predictor, highlighting spatial–temporal variations in driving features. Our model contributes a novel deep learning framework and valuable insights for water quality management in large lakes, emphasizing the importance of considering the dynamic interplay of features or factors in the ’land-river-lake’ system.

Comparison and integration of physical and interpretable AI-driven models for rainfall-runoff simulation

Precise streamflow forecasting in river systems is crucial for water resources management and flood risk assessment. The Tagus Headwaters River Basin (THRB) in Spain is a key hydrological hub, providing regulated flow for agricultural, urban, and energy sectors, and facilitating water transfer to the Segura River Basin, a key semi-arid region. Given the basin's near-total allocation of water resources, accurate streamflow simulations are essential to optimize the socio-economic distribution and ensure sustainable management across both interconnected basins. This study conducts a comprehensive evaluation of the Soil and Water Assessment Tool (SWAT+), support vector regression (SVR), feed forward neural network (FFNN), and long short-term memory (LSTM) models in simulating the rainfall-runoff process at four gauging stations within the THRB. An ensemble machine learning technique is then applied to assess improvements in streamflow estimation. Results revealed that AI-based models significantly surpassed the SWAT+ model in performance. Furthermore, the application of an ensemble technique enhanced the precision of rainfall-runoff modeling by 18 to 26% during the calibration period and 4.1 to 9.2% during the validation period, compared to individual AI-based models. Additionally, the SWAT+ model's precision improved by 44 to 74% and 40 to 55% for the respective periods. The use of Shapley Additive Explanations (SHAP) methodology allowed the results of the ensemble with machine learning to be more interpretable by explaining how each model contributes to the prediction. This research offers significant contributions to hydrological modeling, highlighting the importance of ensemble techniques in elevating predictive accuracy for various river basins.

Detecting drug transfers via the drop-off method: A supervised model approach using AIS data

Maritime security is of tremendous importance in countering drug trafficking, particularly through sea-based routes. In this paper, we address the pressing need for effective detection methods by introducing a novel approach utilizing Automatic Identification System (AIS) data. Our focus lies on detecting the ‘drop-off’ method, a prevalent technique for contraband smuggling at sea. Unlike existing research, primarily employing unsupervised methods, we propose a supervised model specifically tailored to this illicit activity, with a particular emphasis on its application to fishing vessels.Our model significantly reduces the number of data points requiring classification by the observer by 70% , thereby enhancing the efficiency of the drop-off detection process. By employing a Long Short-Term Memory (LSTM) model, our approach demonstrates a change from traditional methods and offers advantages in capturing complex temporal patterns inherent in ‘drop-off’ activities. The rationale behind choosing LSTM lies in its ability to effectively model sequential data, which is essential for detecting drug traffic activities at sea where patterns are subtle and dynamic.Moreover, this model holds the potential for integration into real-time surveillance systems, thereby enhancing operational capabilities in detecting and preventing drug traffic. The generalizability of our model makes for considerable potential in enhancing maritime security efforts and providing assistance in countering drug traffic on a global scale. Importantly, our model outperforms both baseline models, underscoring its effectiveness and superiority in addressing the specific challenges posed by ‘drop-off’ detection. For more information and access to the code repository, please visit this link.