Short-term Memory Network Research Articles

Development of PM2.5 Forecast Model Combining ConvLSTM and DNN in Seoul

Accurate prediction of PM2.5 concentrations is essential for public health management, especially in areas affected by long-range pollutant transport. This study presents a hybrid model combining convolutional long short-term memory (ConvLSTM) and deep neural networks (DNNs) to enhance PM2.5 forecasting in Seoul, South Korea. The hybrid model leverages ConvLSTM’s ability to capture spatiotemporal dependencies and DNN’s strength in feature extraction, enabling it to outperform standalone CMAQ and DNN models. For the T1 forecast (6 h averages), the ConvLSTM-DNN model exhibited superior performance, with an RMSE of 7.2 µg/m3 compared to DNN’s 8.5 µg/m3 and CMAQ’s 10.1 µg/m3. The model also maintained high categorical accuracy (ACC) and probability of detection (POD) for critical PM2.5 levels while reducing false alarms (FARs), particularly in bad and very bad events. Although its performance decreases over extended forecast periods, the ConvLSTM-DNN model demonstrates its utility as a robust forecasting tool. Future work will focus on optimizing the network structure to improve long-term forecast accuracy.

A Deep Learning Approach for Predicting Aerial Suppressant Drops in Wildland Firefighting Using Automatic Dependent Surveillance–Broadcast Data

This study utilizes Automatic Dependent Surveillance–Broadcast (ADS-B) data sourced by the OpenSky Network to curate a dataset aimed at enhancing the precision of aerial suppressant drop predictions in wildland firefighting. By amalgamating ADS-B data with Automated Telemetry Unit (ATU) drop information, this research constructs a reliable base for analyzing the spatial aspects of aerial firefighting operations. Using sequential machine learning models, specifically Long Short-Term Memory (LSTM) networks and 1D Convolutional Neural Networks (1DCNN), the study interprets complex flight dynamics to predict drop locations. The dataset, covering 2017 to 2023, is labeled and segmented to reflect accurate suppressant release events, facilitating the distinction between drop and non-drop activities in fixed-wing aircraft. The LSTM model demonstrated strong predictive performance with an F1 score of 0.922, effectively identifying suppressant drop events with high accuracy. This model’s reliable predictions can significantly improve situational awareness in real-time aerial firefighting operations, enabling more informed decision-making and better coordination of resources during wildfire events.

Predicting Movie Box Office Based on Machine Learning, Deep Learning, and Statistical Methods

Abstract. The global film industry has been proved to impose a significant impact on both culture and the economy, with box office revenue serving as a crucial indicator of a film's commercial success. This study utilizes data from the Kaggle "TMDB Box Office Prediction" competition, encompassing 3,000 films released between 1990 and 2018, to predict movie box office revenue using Random Forest, XGBoost algorithms, and deep learning models such as Bidirectional Long Short-Term Memory (Bidirectional LSTM) and Simple Recurrent Neural Network (SimpleRNN). The goal is to develop a model that accurately predicts movie box office. By comprehensively considering multiple factors such as budget, popularity and film characteristics, this study not only significantly improves the accuracy of box office prediction, but also provides a scientific basis for the formulation of film market strategies. The results demonstrate that the Bidirectional LSTM excels in handling complex time-series data, showing strong trend-capturing capabilities, while XGBoost exhibits greater robustness in dealing with complex data and outliers. These findings can not only provide guidance for making more effective strategies on film production and distribution, but also provide new directions for future research, such as delving into the impact of social media on box office and developing more sophisticated predictive models to adapt to changing market dynamics.

Enhancement algorithm for surface electromyographic-based gesture recognition based on real-time fusion of muscle fatigue features

This study aims to optimize surface electromyography-based gesture recognition technique, focusing on the impact of muscle fatigue on the recognition performance. An innovative real-time analysis algorithm is proposed in the paper, which can extract muscle fatigue features in real time and fuse them into the hand gesture recognition process. Based on self-collected data, this paper applies algorithms such as convolutional neural networks and long short-term memory networks to provide an in-depth analysis of the feature extraction method of muscle fatigue, and compares the impact of muscle fatigue features on the performance of surface electromyography-based gesture recognition tasks. The results show that by fusing the muscle fatigue features in real time, the algorithm proposed in this paper improves the accuracy of hand gesture recognition at different fatigue levels, and the average recognition accuracy for different subjects is also improved. In summary, the algorithm in this paper not only improves the adaptability and robustness of the hand gesture recognition system, but its research process can also provide new insights into the development of gesture recognition technology in the field of biomedical engineering.

Sparse graph structure fusion convolutional network for machinery remaining useful life prediction

Effective prediction of machinery remaining useful life (RUL) is prominent to achieve intelligent preventive maintenance in manufacturing systems. In this paper, a sparse graph structure fusion convolutional network (SGSFCN) is proposed for more accurate end-to-end RUL prediction of machine. A novel node-level graph structure called time series shapelet distance graph (TSSDG) is designed to convert the time series to node feature. The SGSFCN model is proposed to learn degradation information from the graph structure. In SGSFCN, a sparse graph structure (SGS) layer and a fusion graph structure (FGS) layer preceding the graph convolutional network (GCN) are designed to learn the SGS from node representation and fuse the original graph structure, enabling the graph structure and node update iteratively in subsequent layers. Concurrently, a bidirectional long short-term memory network (BiLSTM) layer is integrated to capture the global temporal dependencies. The method is validated by two test rig data, and results demonstrate that the proposed method offers significantly higher prediction accuracy of RUL compared to several state-of-art methods.

Predicting the Deformation of a Concrete Dam Using an Integration of Long Short-Term Memory (LSTM) Networks and Kolmogorov–Arnold Networks (KANs) with a Dual-Stage Attention Mechanism

Predicting the Deformation of a Concrete Dam Using an Integration of Long Short-Term Memory (LSTM) Networks and Kolmogorov–Arnold Networks (KANs) with a Dual-Stage Attention Mechanism

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.

Methodology and Experimental Verification for Predicting the Remaining Useful Life of Milling Cutters Based on Hybrid CNN-LSTM-Attention-PSA

In modern manufacturing, the prediction of the remaining useful life (RUL) of computer numerical control (CNC) milling cutters is crucial for improving production efficiency and product quality. This study proposes a hybrid CNN-LSTM-Attention-PSA model that combines convolutional neural networks (CNN), long short-term memory (LSTM) networks, and attention mechanisms to predict the RUL of CNC milling cutters. The model integrates cutting force, vibration, and current signals for multi-channel feature extraction during cutter wear. The model’s hyperparameters are optimized using a PID-based search algorithm (PSA), and comparative experiments were conducted with different predictive models. The experimental results demonstrate the proposed model’s superior performance compared to CNN, LSTM, and hybrid CNN-LSTM models, achieving an R2 score of 99.42% and reducing MAE, RMSE, and MAPE by significant margins. The results validate that the proposed method has significant reference and practical value for RUL prediction research of CNC milling cutters.

Deep learning-based approach for identifying vortex-induced vibrations in stay cables

Due to variations in wind speed profiles along the length of bridge stay cables, vortex-induced vibrations (VIV) exhibit multimodal characteristics, presenting challenges for VIV identification. Currently, the VIV identification is concentrated on the stable stage of VIV, lacking an available early warning system for detecting the initial developing stage of VIV. In this study, a deep learning-based approach that integrates energy distribution ratio features derived from frequency band wavelet packet decomposition to recognize VIV of stay cable was proposed. Firstly, vibration characteristics induced by vortices in cable-stayed bridges were analyzed based on field monitoring data from the bridge health monitoring system, aiming to propose suitable feature indicators for VIV identification. Secondly, using root mean square as label classification, a deep learning model was constructed, incorporating convolutional neural networks, long short-term memory networks, and attention mechanisms. Finally, four different stages in the evolution of stay cables VIV were identified utilizing field monitoring datasets to analyze the optimal parameter. Meanwhile, effective early warning recognition was achieved through the classification and recognition of confusion matrix. This study provides technical support for early warning systems and structural condition assessment concerning bridge stay cable VIV.

Predicting future evapotranspiration based on remote sensing and deep learning

Study regionThe watersheds of the four flux sites in the United States were selected as the study areas for this research. Study focusThis study validates the efficiency of Convolutional Long Short-Term Memory Network (ConvLSTM) models for site-scale ETa prediction. We enhanced the ConvLSTM model by adding a Spatial Pyramid Pooling module (SPPM) and a Multi-head Self-Attention Module (MSA-Module), creating the Multi-head Self-Attention ConvLSTM (MSA-ConvLSTM) model, which we applied to predicting regional-scale actual evapotranspiration (ETa). This study aims to investigate whether the MSA-ConvLSTM model can enhance the accuracy of predicting regional-scale ETa, considering multiple feature variables. Furthermore, we evaluated different performance indicators, discussed possible reasons for errors in regional ETa prediction, and conducted sensitivity analysis of the model characteristics. New hydrological insights for the regionThe MSA-ConvLSTM model accurately predicts the future state of ETa. The average R2 was 0.81, which is 11.6 % and 5.5 % higher than those of the ConvLSTM and Self-Attention ConvLSTM (SA-ConvLSTM) models, respectively. The average RMSE is 11.94 mm/m, which is 21.5 % and 13.7 % lower than ConvLSTM and SA-ConvLSTM, respectively. The average MAE is 9.46 mm/m, which is 21.3 % and 13 % lower than ConvLSTM and SA-ConvLSTM, respectively. Incorporating of a multi-head self-attention module enhances the model’s capacity for comprehensive understanding of input data features. This improvement allows the model to better adapt to feature relationships at varying scales and angles, enhancing its representational capacity and enabling effective adaptation to complex environmental changes.

Fault diagnosis of floating offshore wind turbine based on bidirectional long short-term memory networks with sliding window

Owing to its ability to harvest more power in deeper waters than other types of offshore wind turbines, the floating offshore wind turbine (FOWT) has attracted significant interest from both academics and industry. However, deterioration in system performance and unscheduled shutdowns will significantly increase operation and maintenance costs. To avoid system damage and ensure operation safety, there is a great demand for developing FOWT fault diagnosis to detect and identify faults as early as feasible. Due to the enormous complexity of FOWT including dynamics and nonlinearity, effective FOWT fault diagnosis poses a significant challenge. This paper develops a novel FOWT fault diagnosis method using bidirectional long short-term memory (BiLSTM) with sliding window. Firstly, the dynamics are embedded by stacking time-dependent measurements using the sliding window technique. Then, the augmented data is fed into a BiLSTM network to exploit the nonlinearity of FOWT. Consequently, a softmax classifier is used for fault diagnosis. A high-fidelity 10 MW spar FOWT benchmark based on the NREL Fatigue, Aerodynamics, Structures, and Turbulence (FAST) model is used to validate the capability and effectiveness of the proposed method. Experimental results show that the proposed method outperforms other related methods in diagnosing critical faults in FOWTs under various operating conditions.

Precision Agriculture using ML for Soil and Weather Prediction

Agriculture is the backbone of many economies and plays a critical role in global food security. However, with growing challenges posed by climate change, erratic weather patterns, and soil degradation, the need for precise and predictive techniques in agriculture has never been more urgent. This paper focuses on the application of machine learning (ML) techniques in soil fertility prediction and weather forecasting, two critical components that can significantly impact agricultural productivity. By analysing soil properties, such as moisture, pH, and nutrient levels, and combining them with accurate weather predictions, our system aims to help farmers make informed decisions regarding crop selection, irrigation scheduling, and pest control. Leveraging algorithms like Naive Bayes for soil classification and Long Short-Term Memory (LSTM) networks for weather prediction, we provide a comprehensive solution for precision agriculture. This system not only enhances productivity but also promotes sustainable agricultural practices by optimizing resource use and reducing wastage

Enhancing Smart Grid Stability through a Hybrid Biometric Pattern Recognition and Long Short-Term Memory Approach

The stability of power grids is critical in ensuring the consistent and efficient delivery of electricity. However, traditional predictive models often fall short in addressing grid data's intricate and ever-changing nature, making it challenging to maintain grid reliability. This paper introduces a novel hybrid approach that combines Biometric Pattern Recognition (BPR) with Long Short-Term Memory (LSTM) networks to enhance the prediction of smart grid stability. This approach employs BPR techniques to extract essential features from smart grid data by leveraging the pattern recognition capabilities typically used in biometric systems. These techniques are particularly effective in identifying and isolating the most relevant patterns within the complex datasets generated by smart grids. On the other hand, the LSTM-based model is designed to handle the temporal dependencies and nonlinear patterns characteristic of grid data. LSTMs, known for their proficiency in time-series analysis, are well-suited for capturing the sequential nature of grid data, enabling more accurate predictions over time. Integrating BPR and LSTM in this hybrid model addresses several limitations in existing predictive methods. By combining the strengths of both techniques, the model enhances the accuracy of predictions and improves the overall reliability of grid stability assessments. Extensive experiments were conducted using real-world datasets to validate the effectiveness of the proposed hybrid model. The results demonstrate a significant improvement in prediction accuracy, with the BPR-LSTM model achieving a 98.25% increase in accuracy compared to traditional prediction methods. This improvement underscores the potential of the BPR-LSTM hybrid approach to play a pivotal role in advancing the stability and reliability of smart grid systems.

Short-Term Prediction Model of Wave Energy Converter Generation Power Based on CNN-BiLSTM-DELA Integration

Wave energy is a promising source of sustainable clean energy, yet its inherent intermittency and irregularity pose challenges for stable grid integration. Accurate forecasting of wave energy power is crucial for reliable grid management. This paper introduces a novel approach that utilizes a Bidirectional Gated Recurrent Unit (BiGRU) network to fit the power matrix, effectively modeling the relationship between wave characteristics and energy output. Leveraging this fitted power matrix, the wave energy converter (WEC) output power is predicted using a model that incorporates a Convolutional Neural Network (CNN), a Bidirectional Long Short-Term Memory (BiLSTM) network, and deformable efficient local attention (DELA), thereby improving the accuracy and robustness of wave energy power prediction. The proposed method employs BiGRU to transform wave parameters into power outputs for various devices, which are subsequently processed by the CNN-BiLSTM-DELA model to forecast future generation. The results indicate that the CNN-BiLSTM-DELA model outperforms BiLSTM, CNN, BP, LSTM, CNN-BiLSTM, and GRU models, achieving the lowest mean squared error (0.0396 W) and mean absolute percentage error (3.7361%), alongside the highest R2 (98.69%), underscoring its exceptional forecasting accuracy. By enhancing power forecasting, the method facilitates effective power generation dispatch, thereby mitigating the adverse effects of randomness on the power grid.

Adaptive Vision-Based Gait Environment Classification for Soft Ankle Exoskeleton

Lower limb exoskeletons have been developed to improve functionality and assist with daily activities in various environments. Although these systems utilize sensors for gait phase detection, they lack anticipatory information about environmental changes, which limits their adaptability. This paper presents a vision-based intelligent gait environment detection algorithm for a lightweight ankle exosuit designed to enhance gait stability and safety for stroke patients, particularly during stair negotiation. The proposed system employs YOLOv8 for real-time environment classification, combined with a long short-term memory (LSTM) network for spatio-temporal feature extraction, enabling the precise detection of environmental transitions. An experimental study evaluated the classification algorithm and soft ankle exosuit performance through three conditions using kinematic analysis and muscle activation measurements. The algorithm achieved an overall accuracy of over 95% per class, which significantly enhanced the exosuit’s ability to detect environmental changes, and thereby improved its responsiveness to various conditions. Notably, the exosuit increased the ankle dorsiflexion angles and reduced the muscle activation during the stair ascent, which enhanced the foot clearance. The results of this study indicate that advanced spatio-temporal feature analysis and environment classification improve the exoskeleton’s gait assistance, improving adaptability in complex environments for stroke patients.

Predicting Disease Outbreaks using Machine Learning Models on Public Health Data

Timely prediction of disease outbreaks is crucial for effective public health responses. This study explores the use of machine learning models to forecast outbreaks by analyzing public health data, including epidemiological, demographic, and environmental factors. The goal is to create a real-time prediction framework to aid health authorities in early interventions. Models such as time-series forecasting, Random Forest, and Long Short-Term Memory (LSTM) networks are applied to predict diseases like influenza, dengue, and COVID-19. These models are evaluated using accuracy, precision and recall. External factors like weather, population mobility, and public sentiment are also examined for their role in disease spread. The results highlight key factors driving outbreaks and demonstrate how machine learning can enhance public health surveillance by providing early warning systems. This research contributes to the development of scalable, data-driven tools for outbreak prediction, supporting proactive public health strategies and minimizing future impacts.

Aluminum dust concentration detection based on LSTM-Kalman filter

Industrial dust emissions present serious hazards, including respiratory issues and explosion risks. Traditional dust concentration detection methods are often compromised by environmental factors. This study introduces a novel FFT-KF-LSTM-NET model to predict high-concentration aluminum powder levels, utilizing 128 sets of electrostatic induction measurement data. By applying the Fast Fourier Transform (FFT) to eliminate low-frequency interference, integrating Long Short-Term Memory (LSTM) networks for advanced time series analysis, and using the Kalman Filter (KF) for rapid model convergence, this approach significantly enhances prediction accuracy. The model achieves an 82% improvement in Mean Squared Error (MSE), reducing it to 0.1336, outperforming traditional methods. Furthermore, by modeling the voltage signal generated by charged dust particles in the electrostatic induction sensor's sensing area and using the first derivative of the voltage signal as a learning feature, the model's prediction speed is increased from 1.5 s-2 s–0.5 s, with improved anti-interference capabilities. These advancements position the FFT-KF-LSTM-NET model as a crucial tool for real-time, reliable dust concentration detection in industrial environments.

Prediction of reference crop evapotranspiration based on improved convolutional neural network (CNN) and long short-term memory network (LSTM) models in Northeast China

The accurate prediction of reference crop evapotranspiration (ET0) is essential to better manage crop irrigation water consumption and improve crop water use efficiency. To effectively improve the accuracy of ET0 simulated by machine learning models, five meteorological stations in Hailaer, Harbin, Hohhot, Changchun, and Dalian were taken as representative stations, daily and monthly ET0 data from 1952 to 2020 were used, and empirical mode decomposition (EMD) and wavelet threshold denoising (WD) were considered. The convolutional neural network (CNN) and long short-term memory network (LSTM) models were improved, and two new hybrid neural network models (EMD–WD–CNN and EMD–WD–LSTM) were constructed. Using the ET0 calculated using the FAO-56 Penman–Monteith (P–M) formula as the standard value, the applicability of the improved machine learning model was evaluated. Results showed the following: i) the daily ET0-PM minimum values of five stations were close to 0, the average values were not significantly increased, and the maximum values significantly fluctuated (the fluctuations in Hailaer and Hohhot showed an upward trend, and the fluctuations in Harbin, Changchun, and Dalian showed a downward trend). The annual average monthly ET0-PM varied seasonally, with the peak in June in the Hailaer station and May in all other stations (the peak in Hohhot was the largest, and the peak in Dalian was the smallest). ii) The daily and monthly ET0 values predicted by the proposed EMD–WD–CNN and EMD–WD–LSTM models were highly consistent with the calculated results of the P–M model, showing high accuracy on the daily and monthly ET0 of the simulated five stations (daily: mean absolute error (MAE), 0.30–0.41 mm/day; root mean square error (RMSE), 0.46–0.60 mm/day; R2, 0.86–0.95; monthly: MAE, 5.66–13.71 mm/month; RMSE, 8.97–18.04 mm/month; R2, 0.91–0.95). iii) The EMD–WD–CNN model was suitable for daily scale ET0 simulation and prediction in Northeast China and monthly scale in Harbin, Changchun, and Hohhot. The EMD–WD–LSTM model was suitable for monthly ET0 simulation and prediction in Hailaer and Dalian in Northeast China. The mixed models of EMD–WD–CNN and EMD–WD–CNN can effectively improve the prediction accuracy of ET0 and can provide a new method for agricultural development and irrigation regulation in Northeast China.

A LSTM-based time series prediction model for China's power supply and its performance evaluation

With the rapid development of China's economy and the acceleration of urbanisation, the stability and sustainability of power supply has become increasingly important. In order to cope with the continuous growth of power demand, accurate power supply prediction has become an essential research direction. In this paper, a time series prediction model for power supply is constructed based on the long short-term memory network (LSTM). Power supply data is organised using methods such as data pre-processing and correlation analysis, and the LSTM model is used to predict future power supply. The performance of the prediction results was evaluated, and the results showed that the model can effectively capture the changing trends of power supply, with high reliability and accuracy, providing important support for scientific management and policy formulation in the power industry.

Machine Learning Models Informed by Connected Mixture Components for Short- and Medium-Term Time Series Forecasting

This paper presents a new approach in the field of probability-informed machine learning (ML). It implies improving the results of ML algorithms and neural networks (NNs) by using probability models as a source of additional features in situations where it is impossible to increase the training datasets for various reasons. We introduce connected mixture components as a source of additional information that can be extracted from a mathematical model. These components are formed using probability mixture models and a special algorithm for merging parameters in the sliding window mode. This approach has been proven effective when applied to real-world time series data for short- and medium-term forecasting. In all cases, the models informed by the connected mixture components showed better results than those that did not use them, although different informed models may be effective for various datasets. The fundamental novelty of the research lies both in a new mathematical approach to informing ML models and in the demonstrated increase in forecasting accuracy in various applications. For geophysical spatiotemporal data, the decrease in Root Mean Square Error (RMSE) was up to 27.7%, and the reduction in Mean Absolute Percentage Error (MAPE) was up to 45.7% compared with ML models without probability informing. The best metrics values were obtained by an informed ensemble architecture that fuses the results of a Long Short-Term Memory (LSTM) network and a transformer. The Mean Squared Error (MSE) for the electricity transformer oil temperature from the ETDataset had improved by up to 10.0% compared with vanilla methods. The best MSE value was obtained by informed random forest. The introduced probability-informed approach allows us to outperform the results of both transformer NN architectures and classical statistical and machine learning methods.