Dynamic landslide displacement prediction method based on long short term memory network (LSTM)

A technology of dynamic prediction and displacement, applied in neural learning methods, measuring devices, biological neural network models, etc., can solve problems such as gradient explosion and gradient disappearance, and achieve the effects of avoiding large deviations, increasing timeliness, and improving prediction accuracy

CN110470259AInactive Publication Date: 2019-11-19XI'AN POLYTECHNIC UNIVERSITY
0 Cites 10 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2019-11-19
Estimated Expiration
Not applicable · inactive patent

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a dynamic landslide displacement prediction method based on a long short term memory network (LSTM). The method comprises the steps of firstly building an online landslide displacement monitoring system, monitoring in real time to acquire complete displacement data within a period, removing abnormal values of the collected displacement data via a 3[omega] algorithm, and normalizing; then, building and training a landslide displacement prediction model of LSTM; and at last, using the acquired normalized data as an input of the model to be input into the landslide displacement prediction model, processing the input data via the prediction model, and thus achieving prediction on landslide displacement in a future period. According to the dynamic landslide displacementprediction method based on LSTM disclosed by the invention, the phenomena of gradient explosion and gradient vanishing that may appear when a recurrent neural network (RNN) network model is training are avoided, and thus the landslide displacement prediction accuracy of the training model is further improved.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention belongs to the technical field of geological disaster monitoring and forecasting, and in particular relates to a dynamic prediction method of landslide displacement based on a long short term memory network (Long Short Term Memory Network, LSTM). Background technique

[0002] Landslide disasters are common geological disasters in China. Because of their wide distribution, frequent occurrence, and rapid movement, the economic losses caused by landslides can reach 1 billion US dollars every year. Not only that, but the secondary disasters caused by landslides are immeasurable. Therefore, it is of great economic value and social significance to take necessary measures to monitor it, and then to predict and forecast landslide disasters scientifically and effectively.

[0003] In the previous studies on landslide displacement prediction, intelligent algorithms such as RBF (Radial Basis Function), BP (Back Propagation), and k-means were mainly...

Examples

Embodiment

[0103] Using Matlab to construct LSTM dynamic prediction model of landslide displacement:

[0104] Data loading and normalization processing:

[0105] [train_data, test_data] = LSTM_data_process();

[0106] data_length=size(train_data,1);

[0107] data_num=size(train_data,2);

[0108] Node number setting after grid search method:

[0109] input_num=12;

[0110] cell_num=18;

[0111] output_num=4;

[0112] The bias of the gates in the network:

[0113] bias_input_gate = rand(1, cell_num);

[0114] bias_forget_gate = rand(1, cell_num);

[0115] bias_output_gate = rand(1, cell_num);

[0116] Network layer weight initialization:

[0117] ab=15.5;

[0118] weight_input_x = rand(input_num, cell_num) / ab;

[0119] weight_input_h=rand(output_num, cell_num) / ab;

[0120] weight_inputgate_x = rand(input_num, cell_num) / ab;

[0121] weight_inputgate_c = rand(cell_num, cell_num) / ab;

[0122] weight_forgetgate_x = rand(input_num, cell_num) / ab;

[0123] weight_forgetgate_c = ra...