CLformer: Locally grouped auto-correlation and convolutional transformer for long-term multivariate time series forecasting

Engineering Applications of Artificial Intelligence, Год журнала: 2023, Номер 121, С. 106042 - 106042

Опубликована: Фев. 26, 2023

Язык: Английский

---

An Ensemble Framework for Short-Term Load Forecasting Based on TimesNet and TCN

Energies, Год журнала: 2023, Номер 16(14), С. 5330 - 5330

Опубликована: Июль 12, 2023

Accurate and efficient short-term power load forecasting is crucial for ensuring the stable operation of systems rational planning electricity resources. However, data are often characterized by nonlinearity instability due to external factors such as meteorological conditions day types, making accurate challenging. While some hybrid models can effectively capture spatiotemporal features data, they overlook multi-periodicity leading suboptimal feature extraction efficiency. In this paper, a novel framework based on TimesNet temporal convolutional network (TCN) proposed. Firstly, original preprocessed reconstruct matrix. Secondly, transforms one-dimensional time series into set two-dimensional tensors multiple periods, capturing dependencies within different scales relationships between in data. Then, employed further extract long-term enabling more global pattern be obtained information. Finally, results achieved from fully connected layer extracted features. To verify effectiveness generalization proposed model, experiments have been conducted ISO-NE Southern China datasets. Experimental show that model greatly outperforms long memory (LSTM), TCN, TimesNet, TCN-LSTM, TimesNet-LSTM models. The reduces mean absolute percentage error 20% 43% dataset 10% 31% dataset, respectively.

Язык: Английский

---

ProtInteract: A deep learning framework for predicting protein–protein interactions

Computational and Structural Biotechnology Journal, Год журнала: 2023, Номер 21, С. 1324 - 1348

Опубликована: Янв. 1, 2023

Proteins mainly perform their functions by interacting with other proteins. Protein–protein interactions underpin various biological activities such as metabolic cycles, signal transduction, and immune response. However, due to the sheer number of proteins, experimental methods for finding non-interacting protein pairs are time-consuming costly. We therefore developed ProtInteract framework predict protein–protein interaction. comprises two components: first, a novel autoencoder architecture that encodes each protein's primary structure lower-dimensional vector while preserving its underlying sequence attributes. This leads faster training second network, deep convolutional neural network (CNN) receives encoded proteins predicts interaction under three different scenarios. In scenario, CNN class given pair. Each indicates ranges confidence scores corresponding probability whether predicted occurs or not. The proposed features significantly low computational complexity relatively fast contributions this work twofold. First, assimilates into pseudo-time series. Therefore, we leverage nature time series physicochemical properties encode amino acid space. approach enables extracting highly informative attributes reducing complexity. Second, utilises information identify based on configuration. Our results suggest performs high accuracy efficiency in predicting protein-protein interactions.

Язык: Английский

---

Application of robust deep learning models to predict mine water inflow: Implication for groundwater environment management

The Science of The Total Environment, Год журнала: 2023, Номер 871, С. 162056 - 162056

Опубликована: Фев. 8, 2023

Язык: Английский

---

TDCN: A novel temporal depthwise convolutional network for short-term load forecasting

International Journal of Electrical Power & Energy Systems, Год журнала: 2025, Номер 165, С. 110512 - 110512

Опубликована: Фев. 5, 2025

Язык: Английский

---