A predictive analytics framework for sensor data using time series and deep learning techniques

Neural Computing and Applications, Journal Year: 2024, Volume and Issue: 36(11), P. 6119 - 6132

Published: Jan. 18, 2024

Abstract IoT devices convert billions of objects into data-generating entities, enabling them to report status and interact with their surroundings. This data comes in various formats, like structured, semi-structured, or unstructured. In addition, it can be collected batches real time. The problem now is how benefit from all this gathered by sensing monitoring changes temperature, light, position. paper, we propose a predictive analytics framework constructed on top open-source technologies such as Apache Spark Kafka. focuses forecasting temperature time series using traditional deep learning methods. analysis prediction tasks were performed Autoregressive Integrated Moving Average (ARIMA), Seasonal (SARIMA), Long Short-Term Memory (LSTM), novel hybrid model based Convolution Neural Network (CNN) LSTM. purpose paper determine whether recently developed learning-based models outperform algorithms the data. empirical studies conducted reported demonstrate that models, specifically LSTM CNN-LSTM, exhibit superior performance compared traditional-based algorithms, ARIMA SARIMA. More specifically, average reduction error rates obtained CNN-LSTM substantial when other indicating superiority learning. Moreover, CNN-LSTM-based exhibits higher degree closeness actual values LSTM-based model.

Language: Английский

---

A Novel Approach for State of Health Estimation and Remaining Useful Life Prediction of Supercapacitors Using an Improved Honey Badger Algorithm Assisted Hybrid Neural Network

Protection and Control of Modern Power Systems, Journal Year: 2024, Volume and Issue: 9(6), P. 1 - 18

Published: Nov. 1, 2024

Supercapacitors (SCs) are widely recognized as excellent clean energy storage devices. Accurate state of health (SOH) estimation and remaining useful life (RUL) prediction essential for ensuring their safe reliable operation. This paper introduces a novel method SOH RUL prediction, based on hybrid neural network optimized by an improved honey badger algorithm (HBA). The combines the advantages convolutional (CNN) bidirectional long-short-term memory (BiLSTM) network. HBA optimizes hyperparameters CNN automatically extracts deep features from time series data reduces dimensionality, which then used input BiLSTM. Additionally, recurrent dropout is introduced in layer to reduce overfitting facilitate learning process. approach not only improves accuracy estimates forecasts but also significantly processing time. SCs under different working conditions validate proposed method. results show that model effectively features, enriches local details, enhances global perception capabilities. outperforms single models, reducing root mean square error below 1%, offers higher robustness compared other methods.

Language: Английский

---

Comparative Analysis of Convolutional Neural Network-Long Short-Term Memory, Sparrow Search Algorithm-Backpropagation Neural Network, and Particle Swarm Optimization-Extreme Learning Machine Models for the Water Discharge of the Buzău River, Romania

Water, Journal Year: 2024, Volume and Issue: 16(2), P. 289 - 289

Published: Jan. 15, 2024

Modeling and forecasting the river flow is essential for management of water resources. In this study, we conduct a comprehensive comparative analysis different models built monthly discharge Buzău River (Romania), measured in upper part river’s basin from January 1955 to December 2010. They employ convolutional neural networks (CNNs) coupled with long short-term memory (LSTM) networks, named CNN-LSTM, sparrow search algorithm backpropagation (SSA-BP), particle swarm optimization extreme learning machines (PSO-ELM). These are evaluated based on various criteria, including computational efficiency, predictive accuracy, adaptability training sets. The obtained applying CNN-LSTM stand out as top performers, demonstrating superior efficiency high especially when set containing data series 1984 (putting Siriu Dam operation) September 2006 (Model type S2). This research provides valuable guidance selecting assessing prediction models, offering practical insights scientific community real-world applications. findings suggest that Model S2 preferred choice forecast predictions due its speed accuracy. S (considering recorded 2006) recommended secondary option. S1 (with period 1955–December 1983) suitable other unavailable. study advances field by presenting precise these their respective strengths

Language: Английский

---

Harnessing a Hybrid CNN-LSTM Model for Portfolio Performance: A Case Study on Stock Selection and Optimization

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 104000 - 104015

Published: Jan. 1, 2023

Language: Английский

---

Predictive evaluation of solar energy variables for a large-scale solar power plant based on triple deep learning forecast models

Alexandria Engineering Journal, Journal Year: 2023, Volume and Issue: 76, P. 51 - 73

Published: June 17, 2023

The advanced development of large-scale solar power plants (LSSPs) has made it necessary to improve accurate forecasting models for the output energy. Solar energy is still hampered by lack predictability in its output, which remains a major hurdle industry. This paper focuses on triple deep learning (DL) techniques such as Artificial Neural Network (ANN), Recurrent (RNN) and Convolutional Network- Long-Short Term Memory CNN-LSTM address this problem. These are utilized variables (SEVs) generation (MWh), soiling loss (%), performance ratio (PR %) determine optimal forecast model. novelty research that first time important system parameters PR have been studied predict feasible model using different DL scheme. SEVs real-time dataset procured from largest plant Pakistan, titled "Quaid-e-Azam Park" (QASP). main significance study ANN, RNN, CNN-LSTM-based were developed process through feature generation, data scaling, training, testing steps prediction values compared with plant's actual over last 7 years, then comparison was future trend next 20 years. aim goal develop three investigate results time-series dataset, well evaluate measure errors appropriate Based forecasting/prediction graphic error results, demonstrated hybrid more capable predictor ANN RNN models. However, slightly better performed predicting soling value than CNN-SLTM RNN. Thus, an SEVs, can guarantee variety LSSPs similar nature following investigations shows key findings importance industrial issue.

Language: Английский

---

An Improved CNN-BILSTM Model for Power Load Prediction in Uncertain Power Systems

Energies, Journal Year: 2024, Volume and Issue: 17(10), P. 2312 - 2312

Published: May 10, 2024

Power load prediction is fundamental for ensuring the reliability of power grid operation and accuracy demand forecasting. However, uncertainties stemming from generation, such as wind speed water flow, along with variations in electricity demand, present new challenges to existing methods. In this paper, we propose an improved Convolutional Neural Network–Bidirectional Long Short-Term Memory (CNN-BILSTM) model analyzing systems affected by uncertain conditions. Initially, delineate uncertainty characteristics inherent real-world establish a data-driven based on fluctuations source loads. Building upon foundation, design CNN-BILSTM model, which comprises convolutional neural network (CNN) module extracting features data, forward (LSTM) reverse LSTM module. The two modules account factors influencing timings entire thus enhancing performance data utilization efficiency. We further conduct comparative experiments evaluate effectiveness proposed model. experimental results demonstrate that can effectively more accurately predict loads within characterized generation demand. Consequently, it exhibits promising prospects industrial applications.

Language: Английский

---

Time-Series Hourly Sea Surface Temperature Prediction Using Deep Neural Network Models

Journal of Marine Science and Engineering, Journal Year: 2023, Volume and Issue: 11(6), P. 1136 - 1136

Published: May 29, 2023

Sea surface temperature (SST) is crucial in ocean research and marine activities. It makes predicting SST of paramount importance. While highly affected by different oceanic, atmospheric, climatic parameters, few papers have investigated time-series prediction based on multiple features. This paper utilized multi features air pressure, water temperature, wind direction, speed for hourly using deep neural networks convolutional network (CNN), long short-term memory (LSTM), CNN–LSTM. Models were trained validated epochs, feature importance was evaluated the leave-one-feature-out method. Air pressure significantly more important than direction speed. Accordingly, selection an essential step prediction. Findings also revealed that all models performed well with low errors, increasing epochs did not necessarily improve modeling. similarly practical, CNN considered most suitable as its training several times faster other two models. With this, variance data helped make accurate predictions, proposed method may higher errors while working variant

Language: Английский

---

Load Forecasting for the Laser Metal Processing Industry Using VMD and Hybrid Deep Learning Models

Energies, Journal Year: 2023, Volume and Issue: 16(14), P. 5381 - 5381

Published: July 14, 2023

Electric load forecasting is crucial for the metallurgy industry because it enables effective resource allocation, production scheduling, and optimized energy management. To achieve an accurate forecasting, essential to develop efficient approach. In this study, we considered time factor of univariate time-series data implement various deep learning models predicting one hour ahead under different conditions (seasonal daily variations). The goal was identify most suitable model each specific condition. two hybrid were proposed. first combines variational mode decomposition (VMD) with a convolutional neural network (CNN) gated recurrent unit (GRU). second incorporates VMD CNN long short-term memory (LSTM). proposed outperformed baseline models. VMD–CNN–LSTM performed well seasonal conditions, average RMSE 12.215 kW, MAE 9.543 MAPE 0.095%. Meanwhile, VMD–CNN–GRU variations, value 11.595 9.092 0.079%. findings support practical application electrical in diverse scenarios, especially concerning variations.

Language: Английский

---

Numerical analysis and deep learning algorithm for photovoltaic-thermal systems using various nanofluids and volume fractions at Riyadh, Saudi Arabia

Case Studies in Thermal Engineering, Journal Year: 2024, Volume and Issue: 54, P. 103974 - 103974

Published: Jan. 4, 2024

Conventional photovoltaic (PV) systems have elevated temperatures in the hot climate of Riyadh, resulting reduced electrical power generation. Therefore, nanofluids are employed photovoltaic-thermal (PV-T) to absorb self-generated heat that limits efficient operation. In addition, developing deep learning models would help improve optimization and control proposed system. This study's primary goal is numerically examine a PV-T system under influence using various with varying volume fractions Riyadh develop time-series algorithms based on findings this examination predict system's potential. A mathematical model investigate panel performance concentrations proposed. The generated data from different deployed train model, which convolutional neural networks integrated two layers long short-term memory (CNN-LSTM), order temperature. According investigation's findings, best coolant CuO nanofluid at 4 % concentration. Utilizing aforementioned can deliver an enhancement average daytime temperature, electrical, thermal, total exergy efficiency by 34.5 °C, 16.7 %, 79.2 18.07 respectively. developed were evaluated mean absolute error (MAE) coefficient determination (R2) scored 0.18–0.35 97.5–98.75

Language: Английский

---

Novel Deep Learning Method in Hip Osteoarthritis Investigation Before and After Total Hip Arthroplasty

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(2), P. 872 - 872

Published: Jan. 17, 2025

The application of gait analysis on patients with Hip Osteoarthritis (HOA) before and after Total Arthroplasty (THA) surgery can provide accurate diagnostics, reliable treatment decision making, proper rehabilitation efforts. Acquired kinematic trajectories discriminating features that be used to determine the patterns healthy subjects effects surgical operation. However, there is still a lack consensus best kinematics achieve this. Our investigation aims utilize Deep Learning (DL) methodologies improve classification results for parameters healthy, HOA, 6 months post-THA cycles. Kinematic angles from lower limb are directly as one-dimensional inputs into DL model. Based human cycle’s features, hybrid Long Short-Term Memory–Convolutional Neural Network (HLSTM-CNN) designed healthy/HOA/THA gaits. It was found, results, sagittal hip knee, front FPA most accuracy above 94% between HOA Interestingly, when using knee analyze THA gaits, common have same misclassifications. This crucial information provides glimpse in determination success or failure THA.

Language: Английский

---