Convolutional Neural Network Long Short-term Memory Research Articles

An Artificial Intelligence Home Monitoring System That Uses CNN and LSTM and is Based on the Android Studio Development Platform

This paper developed an artificial intelligence home environment monitoring system by using the Android Studio development platform. A database was constructed within a server to store sensor data. The proposed system comprises multiple sensors, a message queueing telemetry transport (MQTT) communication protocol, cloud data storage and computation, and end device control. A mobile application was developed using MongoDB software, which is a file-oriented NoSQL database management system developed using C++. This system represents a new database for processing big sensor data. The k-nearest neighbor (KNN) algorithm was used to impute missing data. Node-RED development software was used within the server as a data-receiving, storage, and computing environment that is convenient to manage and maintain. Data on indoor temperature, humidity, and carbon dioxide concentrations are transmitted to a mobile phone application through the MQTT communication protocol for real-time display and monitoring. The system can control a fan or warning light through the mobile application to maintain ambient temperature inside the house and to warn users of emergencies. A long short-term memory (LSTM) model and a convolutional neural network (CNN) model were used to predict indoor temperature, humidity, and carbon dioxide concentrations. Average relative errors in the predicted values of humidity and carbon dioxide concentration were approximately 0.0415% and 0.134%, respectively, for data storage using the KNN algorithm. For indoor temperature prediction, the LSTM model had a mean absolute percentage error of 0.180% and a root-mean-squared error of 0.042 °C. The CNN–LSTM model had a mean absolute percentage error of 1.370% and a root-mean-squared error of 0.117 °C.

Active noise cancellation using hybrid CNN-LSTM classification and genetic algorithm-based community detection extraction with LMS noise filtering process

Unwanted signals in information-bearing signal referred to as noise could degrade the strength of signals in terms of intelligibility and quality. Over the decade, various researchers developed algorithms to enhance speech signal quality and noise reduction. To address the issue, the study propounded the active noise cancellation method by using a hybridized convolutional neural network–long short-term memory (CNN-LSTM) approach and genetic algorithm (GA)-based community detection feature extraction enhanced with least mean square (LMS) noise filtering process. The quality filtered signals were extracted with feature correlation data and precise relevant features using genetic algorithm-based community detection. The selective parameters aid the classification performance, facilitated by hyperparameter fine-tuning of GA-based community detection. The results of incorporating LSTM will eliminate unnecessary memory content by correlating past information outcomes to classify new feature values. In this method, abnormal and normal signals are classified by our LSTM layers output. These classified outcomes aid in the denoise of active voice signals with a fast convergence rate. The efficiency of the proposed method was assessed in terms of its performance through comparative active noise cancellation (ANC) methods. The higher accuracy rate and low NMSE in the classification of audio signals evidenced the efficacy of the proposed ANC model.

Building energy management model integrating rule-based control algorithm and genetic algorithm

Energy is a crucial material foundation for the development of human society. Building energy consumption accounts for a significant proportion of global energy consumption. Optimizing building energy management is of great significance for achieving sustainable development. A building energy management model that integrates rule-based control algorithm and genetic algorithm is proposed, aiming to optimize building energy utilization and reduce operating costs. Mathematical models for different devices in the building energy system are established, and the rule-based control algorithm is used to provide system decision support. Then, the genetic algorithm is integrated to address the complexity and uncertainty of energy optimization problems. The comparative test results showed that the proposed fusion algorithm had higher fitness values and faster convergence speed. The root mean square errors of the algorithm in the training and testing sets were 43.6544 and 43.6844, with the lowest error and highest accuracy among the four algorithms. The simulation experiment results showed that the building energy management model integrating rule-based control algorithm and genetic algorithm had energy expenditures of 788.3 yuan and 967.6 yuan for two types of buildings, respectively. Taking Building 1 as an example, compared with Supervisory Control and Data Acquisition (SCADA), Beetle Antennae Search and Particle Swarm Optimization (BAS-PSO) algorithm, and Long Short-Term Memory-Convolutional Neural Network (LSTM-CNN) algorithm, the proposed model reduced the cost of energy consumption optimization by 39.30%, 28.32%, and 20.20%, respectively. Overall, the proposed building energy management model effectively reduces operating costs, utilizes building energy, and contributes to daily building energy management and decision support.

A CNN‐LSTM Method Based on Voltage Deviation for Predicting the State of Health of Lithium‐Ion Batteries

ABSTRACTEnsuring the accurate estimation of the state of health (SOH) of lithium‐ion batteries (LIBs) is essential for the reliability and safe operation of battery management systems. The prediction of SOH has witnessed significant advancements recently, largely propelled by the powerful nonlinear modeling capabilities of deep learning. Despite these advancements, the intricate nature of the battery degradation process poses a challenge in accurately simulating it using measurement data. In this paper, we introduce a novel approach by focusing on the charging voltage deviation, which is defined as the discrepancy between the charging voltage and its average value over each charge/discharge cycle. This deviation is rooted in the electrochemical reactions that lead to capacity decay and voltage fluctuations. We propose a convolutional neural network‐long short‐term memory (CNN‐LSTM) hybrid framework aimed at estimating the SOH of the battery. For each charge/discharge cycle, a conventional CNN is employed to extract key capacity features from sequential charging data, encompassing voltage deviation, current, and charging duration. Following this, an LSTM network is leveraged to build the long‐term dependencies of battery capacities, facilitating the SOH prediction process. The experimental results indicate that our model not only simplifies the computational complexity but also significantly enhances the precision of SOH predictions. This innovative approach holds promise for the advancement of battery management systems, ensuring their continued reliability and safety.

An Intelligent Approach for Early and Accurate Predication of Cardiac Disease Using Hybrid Artificial Intelligence Techniques.

This study proposes a new hybrid machine learning (ML) model for the early and accurate diagnosis of heart disease. The proposed model is a combination of two powerful ensemble ML models, namely ExtraTreeClassifier (ETC) and XGBoost (XGB), resulting in a hybrid model named ETCXGB. At first, all the features of the utilized heart disease dataset were given as input to the ETC model, which processed it by extracting the predicted probabilities and produced an output. The output of the ETC model was then added to the original feature space by producing an enriched feature matrix, which is then used as input for the XGB model. The new feature matrix is used for training the XGB model, which produces the final result that whether a person has cardiac disease or not, resulting in a high diagnosis accuracy for cardiac disease. In addition to the proposed model, three other hybrid DL models, such as convolutional neural network + recurrent neural network (CNN-RNN), convolutional neural network + long short-term memory (CNN-LSTM), and convolutional neural network + bidirectional long short-term memory (CNN-BLSTM), were also investigated. The proposed ETCXGB model improved the prediction accuracy by 3.91%, while CNN-RNN, CNN-LSTM, and CNN-BLSTM enhanced the prediction accuracy by 1.95%, 2.44%, and 2.45%, respectively, for the diagnosis of cardiac disease. The simulation outcomes illustrate that the proposed ETCXGB hybrid ML outperformed the classical ML and DL models in terms of all performance measures. Therefore, using the proposed hybrid ML model for the diagnosis of cardiac disease will help the medical practitioner make an accurate diagnosis of the disease and will help the healthcare society decrease the mortality rate caused by cardiac disease.

Integrated CNN‐LSTM for Photovoltaic Power Prediction based on Spatio‐Temporal Feature Fusion

ABSTRACTThe accurate prediction of the output power of each power plant is crucial for effective resource deployment. This paper proposes a convolutional neural network‐long short‐term memory (CNN‐LSTM) network integration model based on spatio‐temporal feature fusion. Firstly, the temporal correlation of the PV features of the target power plant and the spatial correlation between the PV power of the target power plant and the PV power of the neighboring power plants are computed. The features are then fused according to the strength of the correlation, which allows the features to be combined with spatial and temporal attributes, which promotes faster and more effective training of the model. Subsequently, an integrated network architecture comprising three individual models, CNN, LSTM, and CNN‐LSTM, is designed. The SENet attention mechanism is utilized to add non‐linear integration weights to the outputs of the individual models. Due to the variability of different neural networks, the prediction results of the integrated model are often higher than the best‐performing individual model. Additionally, we designed different case studies to compare model performance under sunny, rainy, and cloudy conditions. Extensive simulation experiments demonstrate the effectiveness of our proposed integrated approach. When the prediction interval is set to 5 min, the RMSE loss of the integrated model on the test set is reduced by 13.5%, 6.9%, and 5.1% compared to the CNN, LSTM, and CNN‐LSTM models included in the ensemble, respectively.

DeepONet-Inspired Architecture for Efficient Financial Time Series Prediction

Financial time series prediction is a fundamental problem in investment and risk management. Deep learning models, such as multilayer perceptrons, Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM), have been widely used in modeling time series data by incorporating historical information. Among them, LSTM has shown excellent performance in capturing long-term temporal dependencies in time-series data, owing to its enhanced internal memory mechanism. In spite of the success of these models, it is observed that in the presence of sharp changing points, these models fail to perform. To address this problem, we propose, in this article, an innovative financial time series prediction method inspired by the Deep Operator Network (DeepONet) architecture, which uses a combination of transformer architecture and a one-dimensional CNN network for processing feature-based information, followed by an LSTM based network for processing temporal information. It is therefore named the CNN–LSTM–Transformer (CLT) model. It not only incorporates external information to identify latent patterns within the financial data but also excels in capturing their temporal dynamics. The CLT model adapts to evolving market conditions by leveraging diverse deep-learning techniques. This dynamic adaptation of the CLT model plays a pivotal role in navigating abrupt changes in the financial markets. Furthermore, the CLT model improves the long-term prediction accuracy and stability compared with state-of-the-art existing deep learning models and also mitigates adverse effects of market volatility. The experimental results show the feasibility and superiority of the proposed CLT model in terms of prediction accuracy and robustness as compared to existing prediction models. Moreover, we posit that the innovation encapsulated in the proposed DeepONet-inspired CLT model also holds promise for applications beyond the confines of finance, such as remote sensing, data mining, natural language processing, and so on.

Semantically-Enhanced Feature Extraction with CLIP and Transformer Networks for Driver Fatigue Detection.

Drowsy driving is a leading cause of commercial vehicle traffic crashes. The trend is to train fatigue detection models using deep neural networks on driver video data, but challenges remain in coarse and incomplete high-level feature extraction and network architecture optimization. This paper pioneers the use of the CLIP (Contrastive Language-Image Pre-training) model for fatigue detection. And by harnessing the power of a Transformer architecture, sophisticated and long-term temporal features are adeptly extracted from video sequences, paving the way for more nuanced and accurate fatigue analysis. The proposed CT-Net (CLIP-Transformer Network) achieves an AUC (Area Under the Curve) of 0.892, a 36% accuracy improvement over the prevalent CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) end-to-end model, reaching state-of-the-art performance. Experiments show that the CLIP pre-trained model more accurately extracts facial and behavioral features from driver video frames, improving the model's AUC by 7% over the ImageNet-based pre-trained model. Moreover, compared with LSTM, the Transformer more flexibly captures long-term dependencies among temporal features, further enhancing the model's AUC by 4%.

Flexible Polymer-Based Electrodes for Detecting Depression-Related Theta Oscillations in the Medial Prefrontal Cortex

This study investigates neural activity changes in the medial prefrontal cortex (mPFC) of a lipopolysaccharide (LPS)-induced acute depression mouse model using flexible polymer multichannel electrodes, local field potential (LFP) analysis, and a convolutional neural network-long short-term memory (CNN-LSTM) classification model. LPS treatment effectively induced depressive-like behaviors, including increased immobility in the tail suspension and forced swim tests, as well as reduced sucrose preference. These behavioral outcomes validate the LPS-induced depressive phenotype, providing a foundation for neurophysiological analysis. Flexible polymer-based electrodes enabled the long-term recording of high-quality LFP and spike signals from the mPFC. Time-frequency and power spectral density (PSD) analyses revealed a significant increase in theta band (3–8 Hz) amplitude under depressive conditions. Using theta waveform features extracted via empirical mode decomposition (EMD), we classified depressive states with a CNN-LSTM model, achieving high accuracy in both training and validation sets. This study presents a novel approach for depression state recognition using flexible polymer electrodes, EMD, and CNN-LSTM modeling, suggesting that heightened theta oscillations in the mPFC may serve as a neural marker for depression. Future studies may explore theta coupling across brain regions to further elucidate neural network disruptions associated with depression.

Quantitative Modeling and Predictive Analysis of Chemical Oxygen Demand in Wastewater Treatment Systems Utilizing Long Short-Term Memory Neural Network

In the realm of water resource management, meticulous monitoring and control methodologies are quintessential to the refinement of wastewater treatment processes. This research elucidates an avant-garde methodology for forecasting the Chemical Oxygen Demand (COD), an instrumental indicator of water quality, by harnessing the capabilities of long short-term memory (LSTM) neural networks in conjunction with Internet of Things (IoT) paradigms. The efficacy of the LSTM model is juxtaposed with that of an advanced Deep Belief Network (DBN), as well as contemporary models like a Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) hybrid model and a Transformer-based model, employing data sourced from a wastewater treatment facility located in Changsha. The empirical findings show that notwithstanding the comparable training durations used, the LSTM model exhibits a preeminent error rate of merely 7%, thus surpassing the DBN model (which has an error rate of 35%), the CNN-LSTM model (registering a 22% error rate), and the Transformer-based model (with a 17% error rate) in its predictive precision. This research underscores the potential of integrating an astute wastewater control system with IoT and LSTM models, thereby hinting at prospective enhancements in the sustainability and operational efficacy of wastewater treatment installations.

Analysis of athletes’ technical action based on deep learning

In order to raise athletes’ technical proficiency and overall performance, this paper uses deep learning technology to precisely assess table tennis technical activities. The paper builds a hybrid neural network model for technical action analysis of table tennis players, based on Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) in deep learning. First, CNN extracts the spatial characteristics from the video frames. After that, an LSTM processes these data in a time series to accurately recognize and classify the movements of athletes. Lastly, through trials, the model is trained and evaluated using the Table Tennis Tracking Network (TTNET) dataset. The experimental findings demonstrate that: (1) In table tennis technical action analysis, the suggested deep learning model performs better than other comparative models. The efficacy of combining convolution with sequence learning is fully demonstrated by the CNN-LSTM hybrid model, which performs best in all indicators when compared to the CNN, LSTM, Multi-Objective Function (MOF), and Convolutional Neural Network—Long Short-Term Memory (CNN-LSTM) combined models. Its accuracy rate, precision rate, recall rate, and F1 score are 0.923, 0.918, 0.925, and 0.921, respectively. In contrast, the performances of LSTM and CNN are also excellent, but the performance of MOF model is relatively low; (2) In the classification of technical actions, the model has the highest classification accuracy of service, reaching 0.930, and the classification accuracy of other technical actions such as forehand stroke, backhand stroke and spike is also above 0.9, and the time sequence consistency index also maintains a high level, indicating that the model can effectively identify and analyze table tennis technical actions. In addition, the performance evaluation of real-time feedback shows that the model can achieve low processing time and feedback delay in video data processing with different lengths, which ensures the real-time and reliability in practical applications. These results show that the proposed model can not only provide accurate technical action recognition, but also provide timely and effective feedback in practical application, which has high practical value. The results of this paper prove the potential of deep learning technology in the analysis of athletes’ technical actions, and provide scientific basis and effective tools for the technical training and optimization of table tennis players.

(Invited) Comparison of Tree-Based Model with Deep Learning Model for Predicting Effluent pH and Concentration in an Electrochemical Water Desalination Process

Capacitive deionization (CDI) is an emerging water treatment technology that offers potential advantages, including reduced environmental impact, simpler operation, and lower energy consumption in certain applications. However, for optimizing the process and ensuring high-quality treated water, accurate prediction of effluent pH and concentration in CDI systems is crucial. Various numerical models developed for simulating CDI performance often struggle to capture the complexities of real-world operating conditions. Recently, artificial intelligence (AI) techniques have been widely employed to overcome these limitations by their capability to model complex nonlinear relationships and adaptively respond to changing conditions. Therefore, this study presents a comprehensive comparison of tree-based models and a deep learning model in predicting the effluent pH and concentration in a CDI system. The study developed and evaluated the performance of a steady random forest (SRF) model, a dynamic random forest (DRF) model, and a convolutional neural network-long short-term memory (CNN-LSTM) model using the same dataset. The models were trained and tested on tabular data from constant current (CC) and constant voltage (CV) MCDI operations.The results showed that the tree-based models and the CNN-LSTM model achieved similar prediction accuracy when used on the operation mode they were trained on, with R2 values exceeding 0.98 for both effluent pH and concentration. However, in generalization tests across different operation modes, the performance of the SRF model without binary features significantly deteriorated (R2 < 0.115). To address this, binary features were incorporated to distinguish between the CC and CV operating modes, which substantially improved the SRF model's generalization ability, with R2 values (model R2 > 0.5) comparable to the CNN-LSTM model.Feature importance analysis revealed that voltage, current, and influent pH were the most influential factors in predicting both effluent pH and concentration, aligning with the characteristics of the CDI process. Notably, the tree-based models demonstrated a significant advantage in computational efficiency, requiring over 100 times less computation time per iteration than the CNN-LSTM model.These findings suggest that tree-based models, particularly the SRF model with binary features, can be a viable and computationally efficient alternative to deep learning models for modeling and optimizing CDI processes. The approach of incorporating binary features to enhance the generalization ability of tree-based models is a practical and cost-effective solution that can be applied to other complex engineering problems. Further research in this area could help reduce modeling complexity and computational costs, offering more straightforward computational modeling methods for CDI processes.

Inline Detection of Citrus Rind Micro-wounds Using Contrast-enhanced X- ray Imaging: a Feasibility Study

Decay management is crucial in the citrus industry due to the rapid spread of infections through wounds. Despite the urgency, effective methodologies for screening citrus rind micro-wounds are lacking. This study presents a preliminary investigation into real-time detection of citrus rind micro-wounds using contrast-enhanced X-ray imaging. This method highlights and magnifies rind wounds in X-ray images. The process involves immersing fruit in a contrast solution, capturing three sequential X-ray images, and then washing off residual contrast.Satsuma mandarins were used, with potassium iodide (KI) as the contrast agent duo to its distinct contrast properties on rind wounds coupled with high safety levels. A Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model with multi-head attention mechanisms was developed, achieving a detection accuracy of 97.19 %. Post-radiography assessments showed minimal effects on the fruit’s external appearance and internal quality, though a slight weight loss was observed. These results demonstrate the proposed method’s effectiveness in detecting citrus rind micro-wounds, offering a promising approach for enhancing decay management in the citrus industry.

AI-Driven Prediction of Symptom Trajectories in Cancer Care: A Deep Learning Approach for Chemotherapy Management.

This study presents an advanced method for predicting symptom escalation in chemotherapy patients using Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs). The accurate prediction of symptom escalation is critical in cancer care to enable timely interventions and improve symptom management to enhance patients' quality of life during treatment. The analytical dataset consists of daily self-reported symptom logs from chemotherapy patients, including a wide range of symptoms, such as nausea, fatigue, and pain. The original dataset was highly imbalanced, with approximately 84% of the data containing no symptom escalation. The data were resampled into varying interval lengths to address this imbalance and improve the model's ability to detect symptom escalation (n = 3 to n = 7 days). This allowed the model to predict significant changes in symptom severity across these intervals. The results indicate that shorter intervals (n = 3 days) yielded the highest overall performance, with the CNN model achieving an accuracy of 81%, precision of 87%, recall of 80%, and an F1 score of 83%. This was an improvement over the LSTM model, which had an accuracy of 79%, precision of 85%, recall of 79%, and an F1 score of 82%. The model's accuracy and recall declined as the interval length increased, though precision remained relatively stable. The findings demonstrate that both CNN's temporospatial feature extraction and LSTM's temporal modeling effectively capture escalation patterns in symptom progression. By integrating these predictive models into digital health systems, healthcare providers can offer more personalized and proactive care, enabling earlier interventions that may reduce symptom burden and improve treatment adherence. Ultimately, this approach has the potential to significantly enhance the overall quality of life for chemotherapy patients by providing real-time insights into symptom trajectories and guiding clinical decision making.

Predicting Monthly Runoff: A BO-CNN-LSTM Approach for Enhanced Water Resource Management

Monthly runoff prediction is critical for sustainable water resource management, flood control, and drought preparedness. Accurate forecasts assist decision-making and management of water resources, which are critical for human and environmental demands. Previous prediction algorithms frequently lack accuracy due to human-defined hyperparameters. Autonomous hyperparameter tuning is required for improved predictions. Objectives: To improve the accuracy of monthly runoff predictions, this research proposes a BO-CNN-LSTM model that integrates Bayesian optimization (BO) with a convolutional neural network-long short-term memory neural network (CNN-LSTM). Methods/Analysis: The BO algorithm is utilized to enhance the CNN-LSTM model's hyperparameters, which helps to overcome the limits of human-set parameters. Model training and validation were carried out using monthly runoff data from 1956 to 2019 from Lanxi Station in the Hulan River Basin. The model's prediction performance was assessed using root mean square error (RMSE), mean absolute error (MAE), fitting coefficient (R²), and mean relative error (MAPE), and compared with other previous models. Findings: The results show that the BO-CNN-LSTM model outperforms conventional models in prediction accuracy and reduction of errors. The BO method efficiently enhances hyperparameters, resulting in higher performance metrics. This model gives more precise and dependable monthly runoff forecasts, which are critical for handling the Hulan River Basin's water resources. Novelty/Improvement: By combining BO and CNN-LSTM, the proposed model solves the constraints of previous models, improving prediction accuracy and reducing errors. This innovative technique presents a promising novel approach for monthly runoff prediction, with possible applications in larger hydrological prediction.

How Sophisticated Are Neural Networks Needed to Predict Long-Term Nonadiabatic Dynamics?

Nonadiabatic dynamics is key for understanding solar energy conversion and photochemical processes in condensed phases. This often involves the non-Markovian dynamics of the reduced density matrix in open quantum systems, where knowledge of the system's prior states is necessary to predict its future behavior. In this study, we explore time-series machine learning methods for predicting long-time nonadiabatic dynamics based on short-time input data, comparing these methods with the physics-based transfer tensor method (TTM). To understand the impact of memory time on these approaches, we demonstrate that non-Markovian dynamics can be represented as a linear map within the Nakajima-Zwanzig generalized quantum master equation framework. We further propose a practical method to estimate the effective memory time, within a given tolerance, for reduced density matrix propagation. Our predictive models are applied to various physical systems, including spin-boson models, multistate harmonic (MSH) models with Ohmic spectral densities and for a realistic organic photovoltaic system composed of a carotenoid-porphyrin-fullerene triad dissolved in tetrahydrofuran. Results indicate that the simple linear-mapping fully connected neural network (FCN) outperforms the more complicated nonlinear-mapping networks including the gated recurrent unit (GRU) and the convolutional neural network/long short-term memory (CNN-LSTM) in systems with short memory times, such as spin-boson and MSH models. Conversely, the nonlinear CNN-LSTM and GRU models yield higher accuracy in the triad MSH systems characterized by long memory times. These findings offer valuable insights into the role of effective memory time in non-Markovian quantum dynamics, providing practical guidance for the application of time-series machine learning models to complex chemical systems.

Spatio-Temporal Feature Extraction for Pipeline Leak Detection in Smart Cities Using Acoustic Emission Signals: A One-Dimensional Hybrid Convolutional Neural Network–Long Short-Term Memory Approach

Pipeline leakage represents a critical challenge in smart cities and various industries, leading to severe economic, environmental, and safety consequences. Early detection of leaks is essential for overcoming these risks and ensuring the safe operation of pipeline systems. In this study, a hybrid convolutional neural network–long short-term memory (CNN-LSTM) model for pipeline leak detection that uses acoustic emission signals was designed. In this model, acoustic emission signals are initially preprocessed using a Savitzky–Golay filter to reduce noise. The filtered signals are input into the hybrid model, where spatial features are extracted using a CNN. The features are then passed to an LSTM network, which extracts temporal features from the signals. Based on these features, the presence or absence of a leakage is determined. The performance of the proposed model was compared with two alternative approaches: a method that employs combined features from the time domain and LSTM and a bidirectional gated recurrent unit model. The proposed approach demonstrated superior performance, as evidenced by lower validation loss, higher validation accuracy, enhanced confusion matrices, and improved t-distributed stochastic neighbor embedding plots compared to the other models when tested on industrial data. The findings indicate that the proposed model is more effective in accurately detecting pipeline leaks, offering a promising solution for enhancing smart cities and industrial safety.

A Convolutional Neural Network–Long Short-Term Memory–Attention Solar Photovoltaic Power Prediction–Correction Model Based on the Division of Twenty-Four Solar Terms

The prevalence of extreme weather events gives rise to a significant degree of prediction bias in the forecasting of photovoltaic (PV) power. In order to enhance the precision of forecasting outcomes, this study examines the interrelationships between China’s 24 conventional solar terms and extreme meteorological events. Additionally, it proposes a methodology for estimating the short-term generation of PV power based on the division of solar term time series. Firstly, given that the meteorological data from the same festival is more representative of the climate state at the current prediction moment, the sample data are grouped according to the 24 festival time nodes. Secondly, a convolutional neural network–long short-term memory (CNN-LSTM) PV power prediction model based on an Attention mechanism is proposed. This model extracts temporal change information from nonlinear sample data through LSTM, and a CNN link is added at the front end of LSTM to address the issue of LSTM being unable to obtain the spatial linkage of multiple features. Additionally, an Attention mechanism is incorporated at the back end of the CNN to obtain the feature information of crucial time steps, further reducing the multi-step prediction error. Concurrently, a PV power error prediction model is constructed to rectify the outcomes of the aforementioned prediction model. The examination of the measured data from PV power stations and the comparison and analysis with other prediction models demonstrate that the model presented in this paper can effectively enhance the accuracy of PV power predictions.

Oil well fault diagnosis method based on markov transition field and convolutional neural network-long short-term memory

Abstract The analysis of electrical parameters and fault diagnosis in oil wells is crucial in the petroleum industry. Traditional methods often struggle to accurately capture the characteristics of time-series data, resulting in inadequate diagnostic precision. This study investigates a novel approach to improving feature extraction and fault diagnosis of electrical parameter data from oil wells. Specifically, the paper proposes encoding oil well electrical parameter data into two-dimensional images using the Markov Transition Field (MTF) technique to generate an oil well fault dataset. By incorporating the Squeeze-and-Excitation (SE) attention mechanism, the enhanced Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model demonstrates more efficient feature extraction capabilities in oil well fault classification. Experimental results show that this method achieves a fault diagnosis accuracy of 96.7%, significantly surpassing the traditional method’s 92.7%.This research provides an effective solution for improving the precision of fault diagnosis in pumping units, thereby enhancing the safety and economic efficiency of oil well operations.

A novel method of long-term aging prediction for proton exchange membrane fuel cell under the dynamic load cycling condition

Estimating the remaining useful life (RUL) of proton exchange membrane fuel cell (PEMFC) is beneficial for deploying control strategies. Long-term aging prediction forms the foundation for RUL estimation, but currently long-term aging prediction and RUL estimation for dynamic load cycling conditions have not been achieved, and uncertainty during the prediction and estimation processes has not been considered. Therefore, in this study, a novel hybrid prediction framework based on Bayesian gated recurrent unit, model uncertainty and state estimation (MUSE), and convolutional neural network-long short-term memory (CNN-LSTM) is proposed. First, MUSE is used to simultaneously quantify the state of health (SOH) and predict aging trend to generate a guiding sequence Next, based on the sequence and historical experimental data, CNN-LSTM is trained and multi-step prediction is performed. Subsequently, Bayesian gated recurrent unit is employed to fuse the MUSE-based prediction sequence with the CNN-LSTM-based prediction, and confidence interval is provided. This results in a novel model and data-driven long-term aging prediction structure. Finally, SOH during the prediction stage is evaluated by MUSE and RUL is estimated. The proposed method is validated using experimental data under a dynamic load cycling condition, demonstrating accurate long-term aging prediction and RUL estimation while considering uncertainty.