Design and Performance Verification of Deep Learning-Based River Flood Prediction System Design and Digital Twin-Based Its Application

This paper presents a digital twin-based river management and flood prediction system designed for hydrological environments, including volcanic geology. To address the problems of rapid runoff and complex terrain, a deep learning-based hybrid model is proposed that integrates a Convolutional Neural Network (CNN) for spatial feature extraction and a Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) units for temporal sequence modeling. The performance evaluation results show that the proposed CNN-RNN hybrid model outperforms individual CNN and RNN baselines. The hybrid model achieves a macro-average precision of 0.97, a recall of 0.99, and an F1 score of 0.98, significantly outperforming existing methods. The system is also integrated with a 3D digital twin visualization platform to enable real-time monitoring and data-driven decision-making.