Aug 2026· International Conference on Machine Vision and Deep Learning· Vol 14326, pp. 143262J - 143262J-9· 0 citations· 12 references
Engineering
TL;DR
The proposed residual prediction method effectively solves the problem of insufficient utilization of environmental field features in short-term typhoon prediction and provides efficient and accurate deep learning technical support for typhoon disaster early warning.
Abstract
Aiming at the limitations that deep learning methods rarely integrate large-scale environmental field information in typhoon track and intensity prediction, and existing models have overfitting risks in absolute coordinate prediction, this paper proposes a residual prediction method for typhoon multi-task prediction. First, based on three architectures including Long Short-Term Memory (LSTM), Transformer and Multilayer Perceptron (MLP), a residual prediction strategy is introduced to convert absolute coordinate prediction into displacement increment prediction. Second, a multi-source dataset is constructed using the International Best Track Archive for Climate Stewardship (IBTrACS) and ERA5 reanalysis data, and 6-hour short-term prediction experiments are designed. With Mean Distance Error (MDE) and classification accuracy as the core evaluation indicators, performance comparison with baseline models is conducted. Experimental results show that the proposed method performs excellently in all models. Among them, the MLP Baseline achieves the optimal track prediction performance (MDE=46.47 km) and the optimal intensity classification performance (accuracy=91.62%); the LSTM+ERA5 fusion model reduces the track MDE by 4.65% compared with the LSTM Baseline, verifying the effectiveness of environmental field feature fusion. The proposed method effectively solves the problem of insufficient utilization of environmental field features in short-term typhoon prediction and provides efficient and accurate deep learning technical support for typhoon disaster early warning.
To address the strong dependence of space object orbit prediction on physical models and initial conditions, as well as the difficulty of completely eliminating prediction errors, this study proposes a satellite orbit prediction correction method that integrates an attention mechanism with a long short-term memory (LSTM) network. Taking the LAGEOS satellite as the research object, the proposed method uses position error, velocity, and acceleration features extracted from historical orbital data to train a deep learning model for predicting one-day-ahead orbital errors and correcting the SGP4 orbit prediction results. The experimental results show that the ATLSTM model outperforms the LSTM, support vector machine (SVM), back propagation neural network (BP), and bidirectional long short-term memory (BiLSTM) models in both orbital error prediction and correction. The residual ratios of ATLSTM in the X, Y, and Z axes are reduced to 3.68%, 4.77%, and 2.37%, respectively, effectively improving the accuracy of satellite orbital error prediction. Further analysis indicates that a reasonable setting of the number of neurons helps improve model performance, while the prediction difficulty increases with the extension of the prediction duration, suggesting that the ATLSTM model is more suitable for short-term orbital error prediction and correction. In addition, validation results for satellites at different orbital altitudes demonstrate that the proposed model has certain generalization capability. In summary, combining deep learning methods with physical orbital models can effectively improve the accuracy of space object orbit prediction and provides an effective approach for orbital error prediction, space situational awareness, and collision warning.
Qingshan Luo, Jiahao Ji, Tao Yang et al.· PLoS ONE· 0 citations
To address the problems of low model prediction accuracy caused by time-varying nonlinearity and susceptibility to random noise in water quality data, a water quality prediction model based on IGJO-TCN-BiGRU-MHSA has been proposed. First, the adaptive unscented Kalman filter is adopted to filter out random noise in the data, thereby improving data smoothness. Second, the bidirectional gated recurrent unit (BiGRU) is utilized as the benchmark model, capturing both forward and backward temporal features of the water quality data simultaneously. A temporal convolutional network (TCN) is adopted to extract local features from the data, enhancing the ability of the model to extract features at different time scales. Multi-head self-attention (MHSA) is adopted to capture key information in the data in parallel, thereby enhancing the multi-scale feature fusion ability of the model. Finally, the improved golden jackal optimization (IGJO) algorithm is utilized to optimize the L2 of the TCN, the learning rate and the neuron number of the BiGRU, and the key dimension of the MHSA, improving the accuracy of water quality prediction. Historical water quality data from the Huaihe River Basin were used as experimental samples. Compared with classical algorithms such as SVR, LSTM, GRU, BiGRU, and TCN-BiGRU-MHSA, the IGJO-TCN-BiGRU-MHSA prediction algorithm proposed in this study reduced the mean absolute error (MAE) by 39.18%, 34.15%, 31.15%, 27.85%, and 1.94%, reduced the root mean square error (RMSE) by 35.05%, 27.60%, 24.46%, 20.53%, and 3.06%, and improved the coefficient of determination (R2) by 5.34%, 3.39%, 2.89%, 2.21%, and 0.18%, respectively. The experimental results demonstrated that the prediction algorithm based on IGJO-TCN-BiGRU-MHSA significantly improved the water quality prediction accuracy.
Zhengwei Li, Jing Chen· Engineering Research Express· 0 citations
Typhoon track forecasting is of great significance for disaster prevention and mitigation in coastal areas. In recent decades, due to advances in methods like numerical weather prediction, ensemble forecasting, or data-driven modeling. In this paper, an overview will be provided on the current state-of-the-art typhoon track forecasting methods such as traditional NWP system, novel DL or PDE-based hybrid methods. In particular, the review gives an overview on basic concepts, key publications and performance in applications for both direct methods and for ensemble methods. The benefits and drawbacks of data-driven deep learning models will also be reviewed and compared, such as single-modal, multimodal, and physics-informed structures, in terms of prediction accuracy, uncertainty expression, physical consistency, and commercialization relevance. Finally, key challenges and outlooks are presented. These challenges include probabilistic calibration, model generalizability, physical model interpretability and incorporating AI into operation forecasting system. It is hoped that this review will provide an organized guide for readers who are interested in advancing current methods on typhoon tracking forecasts.
Junyan Wu· Mathematical Modeling and Al...· 0 citations
To address the need for high-precision prediction of mine water inflow under complex geological conditions, this study proposes a hybrid deep learning framework for accurate water inrush forecasting. Using daily water inrush records and nine meteorological and hydrological variables collected from the Maoping mining area in northeastern Yunnan Province, China, in 2023, five deep learning models, namely, CNN, LSTM, Transformer, CNN-LSTM, and LSTM-Transformer, were systematically developed and comparatively evaluated. The results demonstrate that the LSTM-Transformer hybrid model achieved the best predictive performance, with an MAE of 0.196, an RMSE of 0.243, and an R2 of 0.882 outperforming both the individual deep learning models and other hybrid architectures. By combining the temporal memory capability of LSTM with the global attention mechanism of Transformer, the proposed model effectively captures the nonlinear, non-stationary, and multi-scale temporal dependencies embedded in mine water inflow series, thereby substantially enhancing prediction accuracy. These findings indicate that hybrid deep learning architectures integrating local temporal memory with global attention mechanisms provide a robust and effective approach for complex engineering time-series forecasting, offering valuable support for high-precision early warning and intelligent prevention of mine water inflow hazards.
Lu-Jun Chai, Zhuo-Lin Li, Fan-Jun Wang et al.· Water· 0 citations
The proposed framework integrates attention-enhanced recurrent modelling with physically informed heuristic constraints with physically informed heuristic constraints, allowing the model to capture complex nonlinear atmospheric dynamics while maintaining meteorological consistency.
Kalyan Chatterjee, Mudassir Khan, Bhoomeshwar Bala et al.· Scientific Reports· 0 citations
Accurate dew point temperature predictions are vital for weather forecasting, agriculture, energy management and environmental monitoring, especially in climate sensitive areas. The traditional statistical method and shallow learning method have some difficulties in the modeling of complicated nonlinear temporal correlation relationship of multi-variate meteorological information. In this paper, this work introduces an attention-based deep learning approach to forecast dew point temperature in the Kaggle Weather Dataset. The proposed approach involves feature engineering, application of SHAP scores for feature selection, Min-Max normalization, and temporal data processing, to increase the predictive accuracy of the model and its interpretability. The model thoroughly evaluates multiple deep learning architectures (LSTM, CNN, CNN-LSTM, attention based models, BiLSTM, GRU, TCN, and hybrid models) based on RMSE, MAE, R2, Adjusted R2, test loss and validation loss. The experimental results confirm that the suggested BiLSTM–BiGRU hybrid model has the best predictive performance with an RMSE of 0.0163 and a R2 score of 0.9938 compared to the other examined models. SHAP, LIME, and ELI5 enhance the transparency of the models, such as global and local descriptions of feature contributions. Additionally, the trained model is integrated into a Flask-based web application that predicts the dew point in real-time and visualizes the results, demonstrating its practical application for decision support systems in climate conditions.
Arrolla Upendar, Dr. K. R. Shailaja, D. R. Kumar et al.· 2026 6th International Confe...· 0 citations