Temporal Dependence Research Articles

Capsular attention Conv-LSTM network (CACN): A deep learning structure for crop yield estimation based on multispectral imagery

Precise prediction of agricultural production output is crucial for farmers, policymakers, and the Farming-related industry. This article introduces a novel methodology to crop yield forecasting using a capsular neural network equipped with Conv-LSTM and attention mechanism. Our model combines the strengths of 3DCNN, and Conv-LSTM, which can capture the temporal dependencies and 3D features of crop yield data, and attention mechanism, which Can prioritize the most significant characteristics for making predictions. We evaluated CACN on a sizable collection of data of soybean crop yield in the United States from 2003 to 2019 and evaluated against various cutting-edge deep learning models. The outcomes indicate that our suggested approach surpasses other models in performance in terms of RMSE, correlation coefficient, and prediction error map. Specifically, our model achieved approximately 14 % improvement in terms of RMSE, compared to the state-of-the-art model Deep-Yield. Our model also demonstrated the ability to extract more meaningful features and capture the complex relationships between crop yield data and meteorological variables. Overall, our proposed method shows great potential for accurate and efficient crop yield forecasting and can be applied to other crops and regions.

A Systematic Literature Review on The Role of LSTM Networks in Capturing Temporal Dependencies in Data Mining Algorithms

Data mining plays a crucial role in extracting meaningful insights from large and complex data sets, with broad applications in sectors like finance, health care, and market analysis. Traditional techniques—such as classification, clustering, association rule mining, regression, and anomaly detection—are effective for analyzing structured data but struggle with sequential data due to the challenges of modeling temporal dependencies. Long Short Term Memory (LSTM) networks, a specialized form of Recurrent Neural Networks (RNNs), provide a solution to these challenges. By incorporating memory cells and gating mechanisms, LSTM effectively manage long-term dependencies and address issues like vanishing and exploding gradients. This paper reviews the impact of LSTM networks on data mining, analyzing over 60 key publications. By synthesizing concepts and recent advancements, the review underscores how LSTMs enhance the ability of data mining algorithms to capture and predict temporal patterns, reflecting current research trends and innovations

Forecasting stock index closing points using ARIMA-GARCH with a rolling data window

This research develops and examines the efficacy of a hybrid ARIMA-GARCH model, augmented by a rolling data window approach, to enhance the accuracy of stock index prediction, specifically focusing on the NEPSE index. Accurate predictions of stock indices are of paramount importance to investors, analysts and policy makers to navigate and circumvent the market uncertainties. The AutoRegressive Integrated Moving Average (ARIMA) model captures linear trends and temporal dependencies in time series data, while the General AutoRegressive Conditional Heteroskedasticity (GARCH) model addresses volatility clustering—ubiquitous character of financial time series—thereby providing a comprehensive framework for prediction. Utilizing a dataset comprised of daily closing points of NEPSE index approximately three years and nine months, the study identifies the ARIMA (5,1,0)-GARCH (1,1) model as optimal fit, upon integrating 180-day rolling data window. This model achieved a Mean Percentage Error of -0.0058% and a correlation of 0.995, which is indicative of superior fit to the underlying time series data. These findings underscore the hybrid model’s capacity to adaptively respond to dynamic market conditions and acclimatize prediction parallel to most recent market trends and volatility. This research is useful for optimizing investment strategies for those invested in Nepalese stocks. Also, this research lays a foundational framework for future investigations into application of this advanced forecasting method in other emerging markets, financial instruments and indices.

Enhancing IoT Security Using GA-HDLAD: A Hybrid Deep Learning Approach for Anomaly Detection

The adoption and use of the Internet of Things (IoT) have increased rapidly over recent years, and cyber threats in IoT devices have also become more common. Thus, the development of a system that can effectively identify malicious attacks and reduce security threats in IoT devices has become a topic of great importance. One of the most serious threats comes from botnets, which commonly attack IoT devices by interrupting the networks required for the devices to run. There are a number of methods that can be used to improve security by identifying unknown patterns in IoT networks, including deep learning and machine learning approaches. In this study, an algorithm named the genetic algorithm with hybrid deep learning-based anomaly detection (GA-HDLAD) is developed, with the aim of improving security by identifying botnets within the IoT environment. The GA-HDLAD technique addresses the problem of high dimensionality by using a genetic algorithm during feature selection. Hybrid deep learning is used to detect botnets; the approach is a combination of recurrent neural networks (RNNs), feature extraction techniques (FETs), and attention concepts. Botnet attacks commonly involve complex patterns that the hybrid deep learning (HDL) method can detect. Moreover, the use of FETs in the model ensures that features can be effectively extracted from spatial data, while temporal dependencies are captured by RNNs. Simulated annealing (SA) is utilized to select the hyperparameters necessary for the HDL approach. In this study, the GA-HDLAD system is experimentally assessed using a benchmark botnet dataset, and the findings reveal that the system provides superior results in comparison to existing detection methods.

AI-driven optimization of agricultural water management for enhanced sustainability

Optimizing agricultural water resource management is crucial for food production, as effective water management can significantly improve irrigation efficiency and crop yields. Currently, precise agricultural water demand forecasting and management have become key research focuses; however, existing methods often fail to capture complex spatial and temporal dependencies. To address this, we propose a novel deep learning framework that combines remote sensing technology with the UNet-ConvLSTM (UCL) model to effectively integrate spatial and temporal features from MODIS and GLDAS datasets. Our model leverages the high-resolution spatial data from UNet and the temporal dependencies captured by ConvLSTM to significantly improve prediction accuracy. Experimental results demonstrate that our UCL model achieves the best R2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$R^2$$\\end{document} compared to existing methods, reaching 0.927 on the MODIS dataset and 0.935 on the GLDAS dataset. This approach highlights the potential of AI and remote sensing technologies in addressing critical challenges in agricultural water management, contributing to more sustainable and efficient food production systems.

Revolutionising Diabetic Retinopathy Diagnosis with Modified Regularisation Long Short-Term Memory Framework

The diagnosis of Diabetic Retinopathy (DR) demands a paradigm shift towards more accurate and efficient solutions to overcome vision impairment. Therefore, the current study introduces a new Modified Regularisation Long Short-term Memory (MR-LSTM) framework approach for DR diagnosis. The proposed framework leverages the power of deep learning and provides a dynamic and robust solution for the early detection of DR, which in turn preserves a patient’s vision. The proposed framework uses a DR Debrecen Dataset from the UCI database with 21 distinct features relevant to retinal health, and employs a series of data preprocessing steps, including data cleaning, normalisation, and transformation, to ensure data quality and compatibility. The MR-LSTM framework excels at capturing temporal dependencies in sequential retinal images, offering a unique advantage in understanding the progression of DR. The MR-LSTM framework is implemented using Python libraries, and the results are compared with those of other popular models. It is observed that the MR-LSTM framework outperforms other models and achieves an accuracy of 97.12 percent and an F1 Score of 98.49. Furthermore, the Receiver Operating Characteristic (ROC) curve reveals an area under the curve of 0.97, highlighting the exceptional ability to discriminate between positive and negative cases of the proposed framework. By revolutionising DR diagnosis with the proposed MR-LSTM framework, it can achieve accurate, timely, and accessible solutions in the fight against vision-threatening conditions.

Based on Deep Learning Model and Flink Streaming Computing Short Term Photovoltaic Power Generation Prediction for Suburban Distribution Network

With the advancement of global energy internet construction, accurate prediction of new energy generation power such as photovoltaic is an important foundation for ensuring the safety and economic working of new power systems. A short-term photovoltaic power generation prediction method for suburban distribution networks based on deep learning model fusion and Flink flow calculation is proposed to address the challenges of complex power grids, diversified disturbance factors, and isolated monitoring points. This method uses Bi directional Long Short Term Memory(BiLSTM) to extract cross sequential nonlinear characteristic of photovoltaic power generation time series data. Compared with standard LSTM, BiLSTM can consider both historical and future information simultaneously, thus extracting richer extracted features from power generation time series data. This method also integrates attention mechanism to capture the importance distribution of historical temporal features for power generation prediction, effectively solving the problem of long-term temporal dependence in standard LSTM models. The Flink streaming computing framework embeds a trained BiLSTM-Attention photovoltaic power generation prediction model, enabling real-time prediction and monitoring analysis of photovoltaic power generation at various monitoring points in the suburban distribution network. This article uses a dataset of a suburban photovoltaic power station for validation, and trains the model with historical power generation data, meteorological factors, weather types, seasons, and other information as inputs. The BiLSTM-Attention fusion model studys the temporal characteristics of power generation, and has high accuracy in predicting short-term photovoltaic power generation in different scenarios. The Flink streaming computing platform can not only process high throughput predicted power data, but also has low time delay.

Multi-Attention Recurrent Neural Network for Multi-Step Prediction of Chlorophyll Concentration

Abstract: Chlorophyll prediction facilitates the comprehension of red tide characteristics and enables early warning. In practice, it is formulated as a multivariate time series forecasting problem aimed at forecasting future chlorophyll concentrations by considering both exogenous factors and chlorophyll. However, the multi-step prediction of chlorophyll concentration poses a formidable challenge due to the intricate interaction between factors and the long temporal dependence between input sequences. In this work, we propose a Multi-attention Recurrent Neural Network (MaRNN) for the multi-step prediction of chlorophyll concentration. The MaRNN comprises an encoder incorporating two-stage spatial attention and a decoder employing temporal attention. The encoder first learns the significance of exogenous factors for prediction in the first phase, and subsequently captures the spatial correlation between the exogenous sequence and chlorophyll sequence in the second phase. The decoder further excavates input sequences that exhibit a strong correlation with the task through temporal attention module, thereby enhancing the prediction accuracy of the model. Experiments conducted on two real-world datasets reveal that MaRNN not only surpasses state-of-the-art methods in performance, but also offers interpretability for chlorophyll prediction.

Significant of Blood Glucose Time-Series Forecasting with Temporal Fusion Transformer Model for Type 1 Diabetes

Blood glucose (BG) prediction has advanced to a state of the art thanks to deep learning models, which have been demonstrated to enhance type 1 diabetes (T1D) therapy. Nevertheless, most current models are limited to single-horizon prediction and have several practical drawbacks, including poor interpretability. In this study, we develop a novel approach to optimize the forecasting model in the training processes using the Optuna function, cross-validation, and the latest dataset. We offer a novel deep learning framework for multi-horizon BG prediction: the temporal fusion transformer (TFT). TFT employs a self-attention mechanism to extract long-term temporal dependencies and enables a model with an auto-tuning adjustment approach on hyperparameters and a cross-validation function on univariate and multivariate input models. On a clinical dataset with T1D subjects, namely ShanghaiT1DM (16 subjects), D1NAMO (9 subjects), and OhioT1DM (6 subjects) datasets, it achieved an average root mean square error (RMSE) of the three datasets used. The TFT model gave a lower error score than the baseline models of neural hierarchical interpolation for time series (N-HiTS)and long short-term memory (LSTM) with the values of 10.08+0.31 mg/dL and 12.34+0.62 mg/dL on the univariate input model and 9.18+1.21 mg/dL and 14.33+0.52 mg/dL on the multivariate input model for the prediction horizons of 30 and 60 minutes, respectively. This result explained that the TFT model was adequate for carrying out multi-horizon BG value-level forecasting and potentially deploying on-edge devices to improve clinical efforts to manage BG levels for T1D patients in real-time application.

Machine learning for base transceiver stations power failure prediction: A multivariate approach

The widespread deployment of cellular networks has improved communication access, driving economic growth and enhancing social connections across diverse regions. Base Transceiver Stations (BTSs), are foundational to mobile networks but are vulnerable to power failures, disrupting service delivery and causing user inconvenience. This paper proposes a machine-learning-based framework for preemptive BTS power failure prediction using multivariate time-series data from power and environmental monitoring systems. We employ a combination of deep learning architectures, including Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and hybrid CNN-LSTM models, to achieve accurate and timely predictions of BTS power failures. CNNs were selected for extracting dependencies among features of a multivariate time-series data, while LSTMs effectively capture temporal dependencies, making them suitable for predicting power failures.The proposed models exhibit noteworthy predictive performance, with the LSTM network emerging as the most accurate model (MSE: 0.001, MAPE: 2.528), followed by the hybrid CNN-LSTM (MSE: 0.001, MAPE: 2.843) and the CNN (MSE: 0.223, MAPE: 2.843). This work demonstrates deep learning’s effectiveness in preemptive BTS failure prediction, enabling proactive maintenance and improved network resilience.

Research on power battery anomaly detection method based on improved TimesNet

Health monitoring and abnormality detection of power batteries for new energy vehicles has been one of the hot topics in recent years. Accurate and efficient power battery anomaly detection is crucial to ensure stable operation of the battery system and energy saving. However, power battery data are often non-linear and unstable due to external factors, such as temperature conditions, which pose challenges for anomaly detection. Existing methods for collecting battery data’s temporal and spatial features are available, but many of them yield suboptimal detection accuracy and feature extraction efficiency due to their disregard for the multi-periodicity of the data. This paper proposes a novel network structure for power battery anomaly detection based on an improved TimesNet. Firstly, the original battery data undergo preprocessing, and the feature correlation coefficient matrix is established using the MIC algorithm. Secondly, the improved TimesNet network is employed to convert the one-dimensional time feature sequence into a two-dimensional tensor, enabling the extraction of temporal features and dependencies at various cycle scales. Finally, the two-dimensional tensor is transformed into a one-dimensional time series, which undergoes information-weighted aggregation to obtain the final anomaly detection results. To assess the effectiveness and generalization of the proposed model, experiments are conducted using Battery and four public datasets for anomaly detection. The experimental results demonstrate that the proposed model outperforms the transformer, autoformer, TimesNet, and Dlinear models, achieving an improvement of 1%–19% in the F1 value and 1%–3% in the ACC compared to the other models.

Enhancing IoT Security through an Artificial Neural Network Approach

This study aims to fortify Internet of Things (IoT) security through the strategic implementation of Artificial Neural Networks (ANNs). With the rapid expansion of IoT devices, traditional security measures have struggled to cope with the dynamic and complex nature of these environments. ANNs, known for their adaptability, are explored as a promising solution to enhance security. The central objective is to significantly improve the accuracy of IoT security measures by optimizing ANN architectures. Using a curated dataset with key environmental parameters, the study evaluates three ANN models—Backpropagation Neural Network (BPNN), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM). The evaluation metrics include accuracy, precision, recall, and F1-score across different train-test splits. Results show that LSTM consistently outperforms BPNN and MLP, demonstrating superior accuracy and the ability to capture temporal dependencies within IoT security data. Implications stress the importance of aligning model selection with specific application goals, considering factors like computational efficiency. In conclusion, this research contributes valuable insights into the practical implementation of ANNs for IoT security, guiding future optimization efforts and addressing real-world deployment challenges to safeguard sensitive data and ensure system resilience in the evolving IoT landscape.

Functional Principal Component Analysis for Continuous Non-Gaussian, Truncated, and Discrete Functional Data.

Mobile health studies often collect multiple within-day self-reported assessments of participants' behavior and well-being on different scales such as physical activity (continuous scale), pain levels (truncated scale), mood states (ordinal scale), and the occurrence of daily life events (binary scale). These assessments, when indexed by time of day, can be treated and analyzed as functional data corresponding to their respective types: continuous, truncated, ordinal, and binary. Motivated by these examples, we develop a functional principal component analysis that deals with all four types of functional data in a unified manner. It employs a semiparametric Gaussian copula model, assuming a generalized latent non-paranormal process as the underlying generating mechanism for these four types of functional data. We specify latent temporal dependence using a covariance estimated through Kendall's bridging method, incorporating smoothness in the bridging process. The approach is then extended with methods for handling both dense and sparse sampling designs, calculating subject-specific latent representations of observed data, latent principal components and principal component scores. Simulation studies demonstrate the method's competitive performance under both dense and sparse sampling designs. The method is applied to data from 497 participants in the National Institute of Mental Health Family Study of Mood Spectrum Disorders to characterize differences in within-day temporal patterns of mood in individuals with the major mood disorder subtypes, including Major Depressive Disorder and Type 1 and 2 Bipolar Disorder. Software implementation of the proposed method is provided in an R-package.

EEG-based patient-specific seizure prediction based on Spatial–Temporal Hypergraph Attention Transformer

Background and Objectives:Patient-specific seizure prediction based on Electroencephalogram (EEG) is a challenging and significant neuromedicine study, which could potentially safeguard drug-resistant patients from unforeseen risks in their daily lives. However, prevalent seizure prediction frameworks based on convolutional neural networks (CNNs) prioritize local information while overlooking long-range global dependencies and high-order spatial correlations prevalent among brain leads. Methods:To address the limitations above, we propose a novel Spatial–Temporal Hypergraph Attention Transformer (STHAT) framework for extracting the key features in EEG signals. Aimed at the characteristics of epilepsy EEG data, our framework comprises two primary branches: (1) Long-range Temporal Feature Dependencies: This branch aims to capture temporal dependencies within EEG signals by leveraging Swin Transformer and 2D-CNN. Utilizing shifting windows and convolution operations, we encode local and global content dependencies within segmented EEG samples. (2) High-Order Spatial Information Correlations: We construct a raw graph structure using the Pearson correlation coefficient, enabling the extraction of initial node embeddings through graph convolution operations (GCO). Subsequently, a dynamic hypergraph is formed using K-Nearest Neighbors (KNN) and K-means clustering, allowing us to employ a Hypergraph Attention Network (HAN) to capture structural context relationships among brain regions. By integrating information from both branches, we utilize a linear layer to derive the final seizure prediction outcome. Experiment and Results:Extensive evaluations are conducted on the widely used CHB-MIT dataset. Our comprehensive experimental results demonstrate the effectiveness of the proposed framework in achieving superior performance in automatic seizure data prediction, outperforming state-of-the-art algorithms. The STHAT framework effectively exploits both temporal and spatial information within EEG signals, enhancing seizure prediction accuracy.

Forecasting Orange Juice Futures: LSTM, ConvLSTM, and Traditional Models Across Trading Horizons

This study evaluated the forecasting accuracy of various models over 5-day and 10-day trading horizons to predict the prices of orange juice futures (OJ = F). The analysis included traditional models like Autoregressive Integrated Moving Average (ARIMA) and advanced neural network models such as Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Backpropagation Neural Network (BPNN), Support Vector Regression (SVR), and Convolutional Long Short-Term Memory (ConvLSTM), incorporating factors like the Commodities Index and the S&P500 Index. We employed loss function metrics and various tests to assess model performance. The results indicated that for the 5-day horizon, the LSTM and ConvLSTM consistently outperformed the other models. LSTM achieved the lowest error rates and demonstrated superior capability in capturing temporal dependencies, especially in single-factor and S&P500 Index predictions. ConvLSTM also performed strongly, effectively modeling spatial and temporal data patterns. In the 10-day horizon, similar trends were observed. LSTM and ConvLSTM models had significantly lower errors and better alignment with actual values. The BPNN model performed well when all factors were included, and the SVR model maintained consistent accuracy, particularly for single-factor predictions. The Diebold–Mariano (DM) test indicated significant differences in forecasting accuracy, favoring advanced neural network models. In addition, incorporating multiple influencing factors further improved predictive performance, enhancing investment outcomes and reducing risk.

A hybrid transformer and attention based recurrent neural network for robust and interpretable sentiment analysis of tweets

Sentiment analysis is a pivotal tool in understanding public opinion, consumer behavior, and social trends, underpinning applications ranging from market research to political analysis. However, existing sentiment analysis models frequently encounter challenges related to linguistic diversity, model generalizability, explainability, and limited availability of labeled datasets. To address these shortcomings, we propose the Transformer and Attention-based Bidirectional LSTM for Sentiment Analysis (TRABSA) model, a novel hybrid sentiment analysis framework that integrates transformer-based architecture, attention mechanism, and recurrent neural networks like BiLSTM. The TRABSA model leverages the powerful RoBERTa-based transformer model for initial feature extraction, capturing complex linguistic nuances from a vast corpus of tweets. This is followed by an attention mechanism that highlights the most informative parts of the text, enhancing the model’s focus on critical sentiment-bearing elements. Finally, the BiLSTM networks process these refined features, capturing temporal dependencies and improving the overall sentiment classification into positive, neutral, and negative classes. Leveraging the latest RoBERTa-based transformer model trained on a vast corpus of 124M tweets, our research bridges existing gaps in sentiment analysis benchmarks, ensuring state-of-the-art accuracy and relevance. Furthermore, we contribute to data diversity by augmenting existing datasets with 411,885 tweets from 32 English-speaking countries and 7,500 tweets from various US states. This study also compares six word-embedding techniques, identifying the most robust preprocessing and embedding methodologies crucial for accurate sentiment analysis and model performance. We meticulously label tweets into positive, neutral, and negative classes using three distinct lexicon-based approaches and select the best one, ensuring optimal sentiment analysis outcomes and model efficacy. Here, we demonstrate that the TRABSA model outperforms the current seven traditional machine learning models, four stacking models, and four hybrid deep learning models, yielding notable gain in accuracy (94%) and effectiveness with a macro average precision of 94%, recall of 93%, and F1-score of 94%. Our further evaluation involves two extended and four external datasets, demonstrating the model’s consistent superiority, robustness, and generalizability across diverse contexts and datasets. Finally, by conducting a thorough study with SHAP and LIME explainable visualization approaches, we offer insights into the interpretability of the TRABSA model, improving comprehension and confidence in the model’s predictions. Our study results make it easier to analyze how citizens respond to resources and events during pandemics since they are integrated into a decision-support system. Applications of this system provide essential assistance for efficient pandemic management, such as resource planning, crowd control, policy formation, vaccination tactics, and quick reaction programs.

Estimation of Hydraulic and Water Quality Parameters Using Long Short-Term Memory in Water Distribution Systems

Predicting essential water quality parameters, such as discharge, pressure, turbidity, temperature, conductivity, residual chlorine, and pH, is crucial for ensuring the safety and efficiency of water supply systems. This study employs long short-term memory (LSTM) networks to address the challenge of capturing temporal dependencies in these complex processes. Our approach, using a robust LSTM-based model, has demonstrated significant predictive accuracy, as evidenced by substantial R-squared values (e.g., 0.86 for discharge and 0.97 for conductivity). These models have proven particularly effective in handling non-linear patterns and time-series data, which are prevalent in water quality metrics. The results indicate the potential for LSTMs not only to enhance the real-time monitoring of water systems but also to aid in the strategic planning and management of water supply systems. This study’s findings can serve as a basis for further research into the integration of AI in environmental engineering, particularly for predictive tasks in complex, dynamic systems.

Probabilistic modeling of multivariate extreme non-Gaussian wind loads and its applications in envelope engineering

Probabilistic modeling for non-Gaussian wind load fields under extreme winds is crucial to the quantitative risk and reliability analysis of envelope structures. Based on the multi-point synchronous wind load data during typhoons, a probabilistic model is proposed to characterize multivariate non-Gaussian wind loads to improve the accuracy of the risk analysis considering data dimension reduction. In the proposed model, the Pareto distribution function is utilized to fit the marginal distribution, and the copula function is employed to construct the dependent structure of wind loads from different locations. In the marginal fitted results, the extreme wind load value in the edge region is about 0.10–0.15 higher than those from inner measurement points at quantile p = 95 %. In addition, the variation gradient from the different variable directions is approximately equal, and a sharp variation occurs at the point (0.10, 0.10) for bivariate joint distributions. Engineering applications given in this article are composed of three parts: calculating regional exceeding probability, estimating wind load vector under a specific regional exceeding probability, and determining the location of joint extreme wind loads using the joint gradient value. Finally, risk analysis of existing roof coverings is carried out, the edge corner region is weaker, and two engineering recommendations are given. For determining the regional design wind load, a comprehensive calculation framework considering the temporal dependence of different locations is established, and the reduction rates are in the range (5 %, 25 %).

Integration of Generative-Adversarial-Network-Based Data Compaction and Spatial Attention Transductive Long Short-Term Memory for Improved Rainfall–Runoff Modeling

This work presents a novel approach to rainfall–runoff modeling. We incorporate GAN-based data compaction into a spatial-attention-enhanced transductive long short-term memory (TLSTM) network. The GAN component reduces data dimensions while retaining essential features. This compaction enables the TLSTM to capture complex temporal dependencies in rainfall–runoff patterns more effectively. When tested on the CAMELS dataset, the model significantly outperforms benchmark LSTM-based models. For 8-day runoff forecasts, our model achieves an NSE of 0.536, compared to 0.326 from the closest competitor. The integration of GAN-based feature extraction with spatial attention mechanisms improves predictive accuracy, particularly for peak-flow events. This method offers a powerful solution for addressing current challenges in water resource management and disaster planning under extreme climate conditions.

Comparative analysis and application of soft sensor models in domestic wastewater treatment for advancing sustainability

ABSTRACT This study focuses on the development and evaluation of soft sensor models for predicting NH3-N values in a wastewater treatment process. The study compares the performance of linear regression (LR), neural networks (NN) and random forest regression (RFR) models. The proposed methodology involves optimizing the sequencing batch reactor process using artificial intelligence and an automatic control system. Real-time NH3-N values are obtained by inputting data from electronic conductivity and temperature sensors into the prediction models. Once the predicted NH3-N value falls below the effluent standard, the cycle ends, improving energy efficiency and sustainability by cutting down the agitator and aerator. The research results demonstrate that the RNN-based NH3-N soft sensor built in this study exhibits the best performance, which is promising for wastewater treatment process optimization and evaluation. The results show that sensor model NNR[0.5Y]H exhibits exceptional performance, utilizing recurrent neural network with 5-step input delays. Sensor NNR[0.5Y]H exhibits an R2 of 0.921, an RMSE of 6.110, and an MAE of 4.558. Based on the findings, recurrent neural network (RNN) variants emerge as the most effective modeling technique due to their ability to capture temporal dependencies and handle variable-length sequences. This study provides satisfied performance results for the NNR[0.5Y]H soft sensor model in NH3-N monitoring and process optimization in wastewater treatment, highlighting the effectiveness of recurrent neural networks and their contribution to improving interpretability, accuracy, and adaptability of soft sensor models.