.png){kind=icon}
.png){kind=icon}
Return
A Flood Prediction Method Using Improved Diffusion and Transformer Models
.png){kind=icon}
OA
AI
DOI：10.3390/w18141691.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Flood forecasting is crucial for predicting floods and facilitating timely evacuations. Machine learning and deep learning algorithms, such as artificial neural networks (ANN), recurrent neural networks (RNN), and Transformer models, have shown significant success in time series prediction tasks. Recently, Diffusion models, which add noise to training data and then reverse the process to recover the data, have gained popularity in data generation. In flood prediction, however, the limited availability of hydrological data from reservoirs often hinders the training of deep learning models. This study proposes a novel approach by combining the Diffusion model with the Transformer model to address the issue of insufficient data. The Diffusion model is used for data augmentation, while the Transformer model is employed for flood flow prediction. For multivariate input datasets, we introduce a Convolutional Block Attention Module (CBAM) into the Transformer. It can adaptively learn the importance weights of different input variables. Additionally, considering the rich combinations of real-world time series—such as temporal trends, periodicity, and local specificities—which are often disrupted by the gradual addition of noise during the Diffusion process, we incorporate a time trend module into the Diffusion model. This enhancement allows the Diffusion model to better extract the temporal characteristics of the original data during generation, producing new data that retains more of the original information and improves the effectiveness of data augmentation. By combining the improved Diffusion and Transformer models, leveraging the powerful generative capability of the enhanced Diffusion model and the better predictive ability of the enhanced Transformer model, the prediction performance is significantly enhanced. Experimental results demonstrate that, in most cases, our proposed model achieves better prediction performance compared to ANN, LSTM, Transformer models and the other flood prediction methods.
Keywords:
flood forecasting
time series data
Transformer model
diffusion model
Journal
W
IF:
3
Papers:
3.2W
Citations:
7.4W