Short-Term Prediction of Vertical Deformation in Tidal Flat Terrains Based on PSO-VMD-LSTM

• Published in: IEEE transactions on instrumentation and measurement Vol. 73; pp. 1 - 14
• Main Authors: Zhou, Qixiao, Ge, Yongqiang, Zhou, Peng, Ge, Han, Wang, Yuhong, Chen, Jiawang, Mei, Deqing
• Format: Journal Article
• Language: English
• Published: New York IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Correlation; Correlation coefficients; Data acquisition; Data models; Decomposition; Deformation; Deformation effects; Emergency warning programs; Fast Fourier transformations; Feature recognition; Fourier transforms; Machine learning; Microelectromechanical systems; Monitoring; Neural networks; Noise prediction; Particle swarm optimization; prediction; Predictive models; Root-mean-square errors; Sensor arrays; terrain deformation; Tidal flats; Tides; variational mode decomposition (VMD); vertical deformation (VD)
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2024.3413180

The accurate prediction of vertical deformation (VD) in tidal flat terrains is crucial for their rational development and disaster warning. However, the high-precision and strong robustness prediction method of VD data in tidal flat terrain is still a great challenge due to the complex environmental conditions. In this study, VD data was collected over seven consecutive days using microelectromechanical system (MEMS) sensor arrays deployed in a tidal flat. Particle swarm optimization (PSO)-optimized variational mode decomposition (VMD) is applied to decompose the acquired VD data for noise removal. The fast Fourier transform (FFT) and Spearman correlation coefficient analysis were employed to identify the inducing factors by comparing them with the decomposed VD. Four machine learning algorithms were evaluated to predict the trend and periodic components of VD, and the results showed that long short-term memory (LSTM) neural network outperformed the other three, achieving a root mean square error (RMSE) less than 0.15 and an R-squared (<inline-formula> <tex-math notation="LaTeX">R^{2} </tex-math></inline-formula>) greater than 0.90. The proposed combined PSO-VMD-LSTM model effectively addresses the challenges of feature recognition, noise removal, and deformation prediction of VD data in tidal flat terrains.