Abstract

Abstract

Abstract

The accurate forecasting of wind power plays a veritable part in integrating renewable energy from wind turbines into power grids. Wind power, being a highly volatile mode of energy generation owing to temporal variations and complex weather patterns, renders reliable predictions essential for energy management and grid stability. In order to tackle this, we propose a hybrid Multi-Head Attention and 1D-Convolutional Neural Network (MHA-CNN) architecture that combines attention mechanisms and convolutional layers to capture both long-term dependencies and localized features in time-series data from a Supervisory Control and Data Acquisition (SCADA) system. The model effectively improves forecasting performance by attaining an R2score of 99.42 for hour-ahead and 96.52 for day-ahead predictions on a 50,540-sample, 10-min SCADA dataset using 5-fold chronological cross-validation, outperforming traditional methods without any manual feature engineering. The proposed method is also evaluated across multiple scenarios to assess the robustness of the proposed approach.OPEN ACCESS Received: 01/10/2025 Accepted: 10/11/2025

Abstract

The accurate forecasting of wind power plays a veritable part in integrating renewable energy from wind turbines into power grids. Wind power, being a highly volatile mode of energy generation owing to temporal variations and complex weather patterns, renders reliable predictions essential for energy management and grid stability. In order to tackle this, we propose a hybrid Multi-Head Attention and 1D-Convolutional Neural Network (MHA-CNN) architecture that combines attention mechanisms and convolutional layers to capture both long-term dependencies and localized features in time-series data from a Supervisory Control and Data Acquisition (SCADA) system. The model effectively improves forecasting performance by attaining an R2score of 99.42 for hour-ahead and 96.52 for day-ahead predictions on a 50,540-sample, 10-min SCADA dataset using 5-fold chronological cross-validation, outperforming traditional methods without any manual feature engineering. The proposed method is also evaluated across multiple scenarios to assess the robustness of the proposed approach.OPEN ACCESS Received: 01/10/2025 Accepted: 10/11/2025 Published: 23/01/2026

Abstract

Abstract

Abstract

In response to the low accuracy and poor performance of traditional machine learning methods in identifying debris flow fans. This paper proposes an optimized SimAm attention mechanism named Spatial Coordinate Attention Module. It combines with convolutional neural networks to achieve precise segmentation of debris flow fans. Firstly, the energy function of the SimAm is improved to retain the spatial coordinate information of features. Secondly, the closed-form solution of the module is obtained through optimization theory to ensure lightweightness, resulting in the Spatial Coordinate Attention Module. Finally, the Spatial Coordinate Attention Module is embedded into classic segmentation network models to compare with mainstream attention mechanisms. Experimental results demonstrate that the proposed method outperforms mainstream attention mechanisms in various classic models, yielding more complete segmentation results. This approach effectively enhances the segmentation performance of the network models in the task of debris flow fans segmentation.

Abstract

In response to the low accuracy and poor performance of traditional machine learning methods in identifying debris flow fans. This paper proposes an optimized Simple, Parameter-Free Attention Module (SimAM) attention mechanism named Spatial Coordinate Attention Module. It combines with convolutional neural networks to achieve precise segmentation of debris flow fans. Firstly, the energy function of the SimAm is improved to retain the spatial coordinate information of features. Secondly, the closed-form solution of the module is obtained through optimization theory to ensure lightweightness, resulting in the Spatial Coordinate Attention Module. Finally, the Spatial Coordinate Attention Module is embedded into classic segmentation network models to compare with mainstream attention mechanisms. Experimental results demonstrate that the proposed method outperforms mainstream attention mechanisms in various classic models, yielding more complete segmentation results. This approach effectively enhances the segmentation performance of the network models in the task of debris flow fans segmentation

Abstract

Machine learning (ML) techniques, especially Deep Learning (DL) techniques, have been applied for flood risk analysis and prediction on spatial historical data to minimize the risk of the loss of lives and properties associated with floods. In recent years, various studies have established DL as an effective approach for building potential flood prediction models. However, a thorough, systematic integration of recent advancements in DL for hydrological forecasting, their reported performance, and challenges is essential for guiding future work effectively. This paper presents a systematic survey of various DL models applied to flood prediction. This systematic review integrates studies published between 2018 and 2025 that used DL models in hydrological forecasting, such as flood prediction, streamflow forecasting, and runoff modeling. A systematic search of major electronic databases was performed using pre-defined inclusion and exclusion criteria. The overall search provided 647 records, which were narrowed down to a final collection of 45 studies after screening and full-text examination based on factors like study design, hydrological forecasting relevance, and application of DL. The review quantitatively summarizes the reported performances of various DL models, including RNN variants (LSTM, GRU), CNN, GAN, and hybrid architectures, across different hydrological forecasting tasks and datasets. Findings indicate that DL models consistently achieve high performance metrics such as NSE of (0.99), RMSE of (24.61) and MAPE of (1.73) for certain applications. Despite these advancements, significant research gaps remain, particularly concerning the scarcity of high-quality, publicly available datasets with detailed spatial information, the need for more robust real-time prediction systems with minimized false alarms, and the development of more generalized models applicable across diverse geographical regions. this review highlights the significant potential of DL in hydrological prediction as well as clearly stating the fundamental challenges that need to be overcome in order to achieve more robust and generalizable flood prediction systems.OPEN ACCESS Received: 25/04/2025 Accepted: 22/07/2025 Published: 27/10/2025

Abstract

(1) Background: Accurately decoding motor imagery (MI) tasks is a prerequisite for creating a MI-based brain-computer interface (BCI). However, low signal-to-noise ratio and non-stationarity of EEG signals present a huge challenge for the classification of MI-EEG signals, restricting the extensive development of the BCI industry.''' '''(2) Methods: In this paper, we propose a novel deep learning model CTANet that integrates both channel and temporal attention mechanisms into a convolutional neural network to improve the classification accuracy of the MI-BCI systems. The model is constituted first by three serially connected temporal, spatial, and temporal convolution layers to extract features from the brain signals, with an efficient channel attention module inserted between the second and the third convolutional layers to highlight useful feature channels. Subsequently, the time segment for task decoding is partitioned into several time windows, and a variance layer is employed for computing the logarithmic variance of each window. Next, a multi-head attention mechanism is adopted to extract temporal dependency of features from different windows. Finally, a fully connected layer is used for classifying MI-EEG signals. (3) Results: The performance of the proposed model was evaluated on two publicly available BCI datasets and compared with the state-of-the-art methods. The experimental results show that for dataset BCIC-IV2a, our network achieved classification accuracies of 81.17% and 84.33% in inter-session and intra-session scenarios respectively, whereas for dataset OpenBMI, our network achieved classification accuracy of 73.06% and 77.59% in inter-session and intra-session scenarios respectively. (4) Conclusions: These results outperform state-of-the-art networks, indicating significant potential of the proposed model CTANet in MI decoding.