Creating interpretable synthetic time series for enhancing the design and implementation of Internet of Things (IoT) solutions

Internet of Things, Год журнала: 2025, Номер unknown, С. 101500 - 101500

Опубликована: Янв. 1, 2025

Язык: Английский

---

Time Series Data Augmentation for Energy Consumption Data Based on Improved TimeGAN

Sensors, Год журнала: 2025, Номер 25(2), С. 493 - 493

Опубликована: Янв. 16, 2025

Predicting the time series energy consumption data of manufacturing processes can optimize management efficiency and reduce maintenance costs for enterprises. Using deep learning algorithms to establish prediction models sensor is an effective approach; however, performance these significantly influenced by quantity quality training data. In real production environments, amount that be collected during process limited, which lead a decline in model performance. this paper, we use improved TimeGAN augmentation data, incorporates multi-head self-attention mechanism layer into recovery enhance accuracy. A hybrid CNN-GRU used predict from operational equipment. After augmentation, exhibits significant reductions RMSE MAE along with increase R2 value. The accuracy maximized when generated synthetic approximately twice original

Язык: Английский

---

A Framework for the Development of Soft Sensors for Process Plants

Chemie Ingenieur Technik, Год журнала: 2025, Номер unknown

Опубликована: Май 13, 2025

Abstract By monitoring sensors, problematic system states in process engineering plants and production lines can be detected corrected before significant damage occurs. Soft sensors as a combination of data analysis digitalization are tailored to specific application. During their development, large amounts time series from therefore analyzed, interpreted consultation with the plant operators, applied basis for design processing methods using digital models, which form core soft sensors. The framework presented here is employed development processes make research, integration into operational technical more effective efficient.

Язык: Английский

---

TSDSystem: a framework to collect, archive and share time series data at volcanological observatories

Bulletin of Volcanology, Год журнала: 2024, Номер 86(8)

Опубликована: Июль 9, 2024

Abstract This paper presents a framework designed to collect, archive, and share time series data coming from sensor networks at Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Etneo (Italy), which we have developed called Time Series Database management System (TSDSystem). The proposes flexible database model for the standardization of implements an optimized technology storage retrieval acquired data. It is implementation multiparametric databases then suitable development in volcanological observatories worldwide. proposed provides web service perform writing reading via standard communication protocol, easily enables interaction with other instruments or automatic systems. All results provided by TSDSystem are represented using common formats context online services. In particular, station metadata representation follows schema inspired International Federation Digital Seismograph Networks, widely known seismology. A GUI (graphical user interface) test document service. Additionally, basic built-in applications supplied joint synchronized visualization as well stations on geographical map. also offers administration tools access policy management, creation monitoring dashboards publication through pages. authorization system that can be used restrict both operations. useful tool engineering surveillance implementing code available open source license public repository together manual.

Язык: Английский

---

Boosting field data using synthetic SCADA datasets for wind turbine condition monitoring

Journal of Physics Conference Series, Год журнала: 2024, Номер 2767(3), С. 032033 - 032033

Опубликована: Июнь 1, 2024

Abstract State-of-the-art Deep Learning (DL) methods based on Supervisory Control and Data Acquisition (SCADA) system data for the detection prognosis of wind turbine faults require large amounts failure successful training generalisation, which are generally not available. This limitation prevents benefiting from superior performance these methods, especially in SCADA-based prognosis. augmentation approaches have been proposed literature generating instances within a SCADA sequence to reduce imbalance between healthy faulty state points, is relevant fault tasks. However, implementation DL-based requires availability multiple run-to-failure sequences. paper proposes data-driven method synthetic sequences with custom operational environmental conditions progression degradation. An Artificial Neural Network (ANN) trained signals that represent factors reconstruct signals. Then, it used generate datasets available experienced gearbox failure. Synthetic sets generated evaluated basis similarity their signal distributions, temporal dynamics each signal, among different those similar field datasets. The results show consistent counterparts, comparatively lower diversity dynamic behaviour time.

Язык: Английский

---

IoT-Based Energy Consumption Prediction Using Transformers

Gazi University Journal of Science Part A Engineering and Innovation, Год журнала: 2024, Номер 11(2), С. 304 - 323

Опубликована: Июнь 5, 2024

With the advancement of various IoT-based systems, amount data is steadily increasing. The increase on a daily basis essential for decision-makers to assess current situations and formulate future policies. Among types data, time-series presents challenging relationship between dependencies. Time-series prediction aims forecast values target variables by leveraging insights gained from past points. Recent advancements in deep learning-based algorithms have surpassed traditional machine IoT systems. In this study, we employ Enc & Dec Transformer, latest neural networks problems. obtained results were compared with Encoder-only Decoder-only Transformer blocks as well well-known recurrent based algorithms, including 1D-CNN, RNN, LSTM, GRU. To validate our approach, utilize three different univariate datasets collected an hourly basis, focusing energy consumption within Our demonstrate that proposed model outperforms its counterparts, achieving minimum Mean Squared Error (MSE) 0.020 small, 0.008 medium, 0.006 large-sized datasets.

Язык: Английский

---

Flexible system architecture used to collect and store signals acquired by IoT devices

2022 14th International Conference on Electronics, Computers and Artificial Intelligence (ECAI), Год журнала: 2024, Номер unknown, С. 1 - 5

Опубликована: Июнь 27, 2024

Язык: Английский

---

Challenges in Design and Validation of MEMS Based Smart Sensor Systems

Опубликована: Июнь 2, 2024

Язык: Английский

---

TriCache: Providing three-tier caching for time series data in serverless healthcare services

Опубликована: Сен. 30, 2024

Healthcare services and IoT, as highlighted by Hu et al. [9], generate enormous volumes of time series data. Using caching in serverless functions can significantly reduce latency improve performance when storing frequently accessed data memory. Although several approaches offer improvements, such the use in-memory caching, prediction, distributed systems, none them fully addresses need for a robust efficient system healthcare, leaving gap necessary availability optimization. The TriCache model proposes three-tier to optimize storage access healthcare functions, using combination memory function, cache, disk storage, addition predictive intelligence. main contribution is significant reduction improvement hit rate efficiently predicting allocating across different cache layers. Experiments demonstrated notable response time, with 110 millisecond decrease 99th percentile. Additionally, performed significantly, achieving 93% rate, compared 78% observed traditional model.

Язык: Английский

---

Lindorm-UWC: An Ultra-Wide-Column Database for Internet of Vehicles

Proceedings of the VLDB Endowment, Год журнала: 2024, Номер 17(12), С. 4117 - 4129

Опубликована: Авг. 1, 2024

In the Internet of Vehicle (IoV) systems, intelligent vehicles generate huge amounts data that supports diverse services and applications. practice, database systems are deployed in cloud to manage uploaded from vehicle side provide real-time query capacities. However, existing ill-suited because IoV contains a large number metrics is written at an extremely high throughput. To better understand corresponding challenges underlying we conduct first extensive empirical study real-world workloads. According our findings study, design Lindorm-UWC as superior for systems. It implements distributed architecture cold/hot separation mechanism accommodate massive data. each partition, it deploys ultra-wide-column storage engine efficiently handle ingestion multi-metric We evaluate under different scales various types query. Our experimental results show can always achieve higher write throughput (over 79% increase) competitive performance compared alternative solutions. has been serving enterprise customers on Alibaba Cloud since 2019, managing tens petabytes

Язык: Английский

---