Attention Mechanism In Randomized Time Warping

The widely distributed “Step-type” landslides in the Three Gorges Reservoir (TGR) area have caused serious casualties and heavy economic losses. The prediction research of landslide displacement will be beneficial to the establishment of local geological hazard early warning systems for the realization of scientific disaster prevention and mitigation. However, the number of observed data like landslide displacement, rainfall, and reservoir water level in this area is very small, which results in difficulties for the training of advanced deep learning model to obtain more accurate prediction results. To solve the above problems, a Two-stage Combined Deep Learning Dynamic Prediction Model (TC-DLDPM) for predicting the typical “Step-type” landslides in the TGR area under the condition of small samples is proposed. The establishment process of this method is as follows: (1) the Dynamic Time warping (DTW) method is used to enhance the small samples of cumulative displacement data obtained by the Global Positioning System (GPS); (2) A Difference Decomposition Method (DDM) based on sequence difference is proposed, which decomposes the cumulative displacement into trend displacement and periodic displacement, and then the cubic polynomial fitting method is used to predict the trend displacement; (3) the periodic displacement component is predicted by the proposed TC-DLDPM model combined with external environmental factors such as rainfall and reservoir water level. The TC-DLDPM model combines the advantages of Convolutional Neural Network (CNN), Attention mechanism, and Long Short-term Memory network (LSTM) to carry out two-stage learning and parameter transfer, which can effectively realize the construction of a deep learning model for high-precision under the condition of small samples. A variety of advanced prediction models are compared with the TC-DLDPM model, and it is verified that the proposed method can accurately predict landslide displacement, especially in the case of drastic changes in external factors. The TC-DLDPM model can capture the spatio-temporal characteristics and dynamic evolution characteristics of landslide displacement, reduce the complexity of the model, and the number of model training calculations. Therefore, it provides a better solution and exploration idea for the prediction of landslide displacement under the condition of small samples.

Bearing is a key component in rotary machines. Their failures may cause the abrupt shutdown of these machines, which would result in substantial economic losses. Therefore, the prediction of the remaining useful life (RUL) of bearings is regarded as one of the critical approaches to avoid failure of bearings and their systems. In this article, an ensemble data-driven approach is proposed to predict the RUL of bearings. It uses feature extraction, an attention mechanism, and uncertainty analysis. First, the features embedded in the bearings' vibration signals are extracted. Second, a stacked gated recurrent unit (GRU) is constructed to predict the bearing RUL. A novel attention mechanism based on dynamic time warping (DTW) is developed to improve the performance of information extraction, and a Bayesian approach is employed to analyze the prediction uncertainty. Finally, the proposed approach is validated using two benchmark-bearing data sets. The results show that the proposed approach can predict the bearing RUL effectively, and the prediction uncertainty can also be evaluated.

The standardisation and accuracy of the starting movement will affect the athletes' performance and the results of the game, and the differences in different sports, athletes' physical qualities and coaches' teaching methods make it difficult for the traditional correction of the starting movement to reflect a better application effect. Therefore, the study is based on the Openpose human posture estimation and recognition algorithm, which performs frame-posture relationship correlation analysis, continuity movement design and joint bone point repair to improve the accuracy of recognition. At the same time, taking into account the differences in the content of different assignments of posture data information, the study introduces a lightweight attention mechanism that focuses on temporal patterns on the basis of the recognition algorithm. The experimental results show that the recognition accuracy of the improved algorithm for the starting posture is above 0.85, and its recognition accuracy for the three phases of the starting action is above 93% compared with the Dynamic Time Warping (DTW) model and the High Resolution Human Keypoint Detection Network (HRNet-w48), with an F1 value of 0.6857, which is better than that of the DTW model (0.6857), and the F1 value of 0.6857. better than the DTW model (0.6214) and HRNet-w48 model (0.6321). The detection and recognition time of the improved algorithm for the starting movement is less than 0.1s, and it can better provide athletes with movement correction suggestions and formulate personalised training plans.

Ensuring the safety of nuclear energy necessitates proactive measures to prevent the escalation of severe operational conditions. This article presents an efficient and interpretable framework for the swift identification of abnormal events in nuclear power plants (NPPs), equipping operators with timely insights for effective decision-making. A novel neural network architecture, combining Long Short-Term Memory (LSTM) and attention mechanisms, is proposed to address the challenge of signal coupling. The derivative dynamic time warping (DDTW) method enhances interpretability by comparing time series operating parameters during abnormal and normal states. Experimental validation demonstrates high real-time accuracy, underscoring the broader applicability of the approach across NPPs.

Oral translation in English serves as a critical conduit for international communication and cultural exchange. However, the prevalent variations in pronunciation and the rapid pace of spoken language currently impede the efficacy of synchronous translation methods. To improve the quality and efficiency of synchronous oral translation, this paper explores the integration of cross-modal semantic understanding and synchronous enhancement specifically for English oral translation. This exploration commences with the implementation of a cross-modal translation scenario. Subsequently, the text sequence derived from this process is amalgamated with the original speech features via Bidirectional Encoder Representations from Transformers (BERT). The cross-information between modalities is explored, and linear transformation optimization is performed on the self-attention mechanism in Transformer to achieve context-awareness and understanding of oral-transcribed text. In conclusion, the integration of dynamic time warping (DTW) enhances real-time synchronization between speech and text, thereby improving translation fluency. Experimental results reveal that, when compared to the existing bilingual attention neural machine translation (NMT) model and the context-aware NMT model, the model proposed in this study yields an average bilingual evaluation understudy (BLEU) score that is 9.3% and 26.9% higher, respectively. Furthermore, its synchronization speed surpasses that of the other two models by 17.9% and 16.8%, respectively. These findings suggest that the fusion model, which incorporates context-awareness and an attention mechanism in cross-modal translation, can significantly elevate the quality and efficiency of English oral translation, offering a novel approach to the synchronous translation of spoken English.

With the advancement of the Rural Revitalization Strategy and the “Dual Carbon” goals, rural energy systems are exhibiting pronounced multi-energy coupling, a high penetration of renewable energy, and strong load randomness, placing higher demands on the construction of source-load scenarios across multiple time scales. Addressing the limitations of traditional statistical models in generating high-quality short-term source-load scenarios and the tendency of deep learning methods to overlook medium- to long-term seasonal evolution patterns, this paper proposes a hybrid data- and model-driven method for constructing multi-energy source-load scenarios in rural systems. This method establishes a multi-time-scale generation framework encompassing daily scenarios, monthly scenarios, and a full-year 8760-h annual scenario. The proposed approach begins with data preprocessing to form a high-quality sample set. Subsequently, it utilizes a Wasserstein Generative Adversarial Network with Gradient Penalty model (WCGAN-GP) integrated with a temporal attention mechanism to achieve high-fidelity short-term generation of source-load profiles, including photovoltaic, small hydropower, electrical load, and thermal load. Furthermore, it characterizes seasonal evolution features by combining Dynamic Time Warping K-Medoids (DTW–K-Medoids) for typical day extraction and monthly Markov transition modeling. Finally, the annual scenario is continuously spliced and generated through a cross-month boundary cost function and dynamic programming path optimization. Results demonstrate that the short-term scenarios perform excellently across various evaluation metrics, while the monthly and annual scenarios accurately reflect seasonal variations and long-term trends, validating the accuracy and applicability of the proposed method.

GAOformer: An adaptive spatiotemporal feature fusion transformer utilizing GAT and optimizable graph matrixes for offshore wind speed prediction

Precise customer-transformer topological relationships are critical for the safe operation and management of a low-voltage distribution grid (LVDG). This paper proposes a customer-transformer topological relationship identification method for LVDG using self-adaptive physics-constrained graph attention network (SA-PCGAT). First, considering voltage data misalignment in time series, dynamic time warping (DTW) is used for preprocessing. This aligns the raw voltage data to precisely measure fluctuation similarities. Second, the DTW is used as a similarity measure to cluster customer voltage sequences, thereby preliminarily identifying the customer-transformer topological relationships. Then, a SA-PCGAT is introduced to leverage a multi-head attention mechanism to extract nodal electrical features. Its novel adaptive coordination mechanism dynamically balances various physical constraints, achieving refined precision in the topological relationships. Finally, case studies demonstrate that the customer-transformer topological map generated by the proposed method achieves an approximate 10% performance improvement over standard DTW-based clustering methods (with an accuracy of 80.5%). This result validates the precision of the proposed method, highlighting its practical value and potential for widespread application.