Particle filter hybrid prediction model

Combining a physics-based model with LSTM through particle filtering, it achieves high-accuracy real-time prediction of full-field riser dynamics using limited sensor data, with exceptional noise resistance.

Hu, Pengji · Yue, Xiao · Fanpeng, Li · Xiaoyu, Hu · Liu, Zhaowei · Liu, Xiuquan · 77318102 · Chen, Guoming · Wentuo, Li · Mingyuan, Sun

Ocean Engineering 2026

Advantages

Higher accuracy

Particle filtering fuses the physics-based model with LSTM by weighting state and observation vectors, yielding more accurate dynamic response prediction than standalone LSTM.

Noise-resistant

The particle filter framework provides exceptional resistance to measurement noise, enabling stable full-field response prediction even with noisy sensor data.

Applications

• Riser vibration monitoring：Predicts full-field riser dynamics in real time from sparse sensor data, replacing interpolation between measurement points.