Study On Slope Displacement Prediction Based On Conv-LSTM

With the development of big data and artificial intelligence technology,the stability of slopes can be predicted more accurately,which is of great significance for protecting the people's property safety.The displacement of a slope is an external feature that reflects its internal stability,and it is affected by many factors such as temperature,humidity,and precipitation.Using the slope displacement monitoring system to collect slope displacement data,and establish a deep learning model to analyze and predict the data,it can effectively grasp the law of slope displacement changes and predict its displacement value.Deep learning methods have outstanding advantages in analyzing complex nonlinear changing systems and have been widely used in many fields.Based on the convolutional neural network(CNN)and long short-term memory network(LSTM),this paper analyzes the slope displacement systematically,uses the CNN part to extract the features of the slope displacement multi-factor data set,and uses the LSTM part to predict the displacement.The main research contents are as follows:(1)The development status of slope displacement prediction and deep learning is introduced.Since the deformation process of the slope is a non-linear change process affected by multiple factors,the traditional prediction method cannot fully extract the characteristics of the multi-factor data,so the deep learning method is used to predict the displacement.At the same time,the collection process of the slope displacement multi-factor data set is also introduced.The GNSS receiver can accurately monitor the slope displacement,providing reliable data support for subsequent experiments.(2)Introduced the relevant theoretical basis and adaptive training methods of CNN and LSTM,and established an adaptive slope displacement prediction model based on CNN-LSTM.By analyzing the number of convolutional layers,the size of the convolution kernel,the number of LSTM memory units,and the choice of optimizer and other hyperparameter settings,the prediction model with the best prediction effect is determined.The prediction results show that,compared with a single prediction model,the average absolute error is 3.176%,and the root mean square error is 0.197.All prediction indicators are better than single models such as LSTM and BP.(3)Introduced the Conv-LSTM prediction model that introduced the Dropout mechanism.In view of the possible overfitting of the model,the Dropout mechanism was added during the training process,and the main role played by the convolutional layer and the pooling layer in the CNN was taken into account,and the Conv-LSTM prediction model optimized by Dropout was established.The results show that the average absolute error of the Conv-LSTM model is about 1.6% lower than that of the CNN-LSTM model.Compared with the parallel LSTM prediction model,its running time is shorter,which saves time and cost.(4)Introduce the Conv-LSTM prediction model optimized by the self-attention mechanism.Aiming at the problem that in the process of extracting features of the convolutional layer,the internal feature extraction between the data may be insufficient,the self-attention mechanism Conv-LSTM prediction model is established,and the self-attention mechanism is used to fully extract the slope displacement data The relationship between the characteristics.The results show that the average absolute error of the self-attention mechanism Conv-LSTM slope displacement prediction model is only 0.441%,which is 1.1% lower than the Conv-LSTM model and 2.735% lower than the CNN-LSTM model.