Details

Autor(en) / Beteiligte

• Chen, Haiwei
• Wang, Feng
• Yu, Rufan
• Zhou, Haonan
• Li, Juzhi

Titel

Dual Reciprocal Sensing Prediction Method of Underground 3D Displacement Based on RBF-MLP Neural Network Combination Model

Ist Teil von

• 2023 IEEE 3rd International Conference on Power, Electronics and Computer Applications (ICPECA), 2023, p.1730-1735

Ort / Verlag

IEEE

Erscheinungsjahr

2023

Quelle

IEEE Electronic Library Online

Beschreibungen/Notizen

• Landslides are a common natural disaster, often causing severe damage to people's lives and property. It is, therefore, necessary to monitor the three-dimensional underground displacement in geological hazard monitoring. The device uses the dual reciprocal voltage method to infer three-dimensional subsurface displacements based on the principle of electromagnetic mutual inductance. In this dissertation, the horizontal and vertical displacements are studied, and the reciprocal inductance voltage one, reciprocal inductance voltage two, and tilt angle are used as input parameters. The radial basis neural network is chosen to approximate the non-linear function by interpolation. Considering that RBF has only a single hidden layer, it may not perform well for highly non-linear problems. Therefore, a new method is proposed for predicting subsurface 3D displacement based on a combined RBF-MLP neural network model by augmenting the number of nodes in the RBF output layer and connecting the MLP neural network. The training data and model structure are essential for predicting subsurface 3D displacements, so the intervals of horizontal and vertical displacements are 0-50 mm, sampled at 1 mm intervals, and the relative axes are 0°-80°, sampled at 5° intervals. In the model evaluation, the average absolute error of horizontal and vertical displacements is within 0.4mm, which meets the accuracy requirements of landslide subsurface displacement monitoring.