Land subsidence prediction using recurrent neural networks

Abstract

In an environment, one of the natural geological hazards is land surface subsidence. There are several reasons for land subsidence among them are underground coal mining and coal fire in subsurface. The deformation is primarily measured in terms of change in ground elevation values (Z-dimension) at different time intervals at identified ground locations. All the conventional and exiting techniques have certain limitations in monitoring and predicting land surface subsidence. In this work, we predict the land subsidence for one year in the interval of twelve days on the datasets collected through a monitoring technique called Modified PSIn-SAR. The sample datasets contains 14 locations and 67 previous land subsidence value calculated from each location. We train and test predictive models and perform the prediction of the land subsidence using Vanilla and Stacked long short-term memories (LSTMs). Finally, we demonstrates the predicted deformation values of the 14 locations for one year.

Activity location recognition from mobile phone data using improved HAC and Bi‐LSTM

“…The performance of Bi‐LSTM is under the influence of two hyperparameters, which are (a) Window length and (b) Batch size [46]. We conducted experiments on these hyperparameters and plot the results in Figure 18.…”

“…Each LSTM network utilises an input gate, an output gate and a forget gate to control memory cells. Figure 7 shows the structure of the LSTM cell [46]. For time t , the input gate is represented by i t , the forget gate by f t and the output gate by o t .…”

“…For time t , the input gate is represented by i t , the forget gate by f t and the output gate by o t . Equations (8)–(10) are used to calculate the gates of a cell [46, 47]. $$\begin{equation}{i}_t = \sigma \left( {{{{\bf W}}}_t \cdot \left[ {{h}_{t - 1},{x}_t} \right] + {b}_i} \right),\end{equation}$$ $$\begin{equation}{f}_t = \sigma \left( {{{{\bf W}}}_f \cdot \left[ {{h}_{t - 1},{x}_t} \right] + {b}_f} \right),\end{equation}$$ $$\begin{equation}\left\{ { \def\eqcellsep{&}\begin{array}{@{}*{1}{l}@{}} {{o}_t = \sigma \left( {{{{\bf W}}}_o \cdot \left[ {{h}_{t - 1},{x}_t} \right] + {b}_o} \right)}\\[7pt] {{h}_t = {o}_t \times \tanh \left( {{C}_t} \right)} \end{array} } \right.,\end{equation}$$where x t is the input data at moment t .…”

“…For time t, the input gate is represented by i t , the forget gate by f t and the output gate by o t . 8)-( 10) are used to calculate the gates of a cell [46,47].…”

STA-TSN: Spatial-Temporal Attention Temporal Segment Network for action recognition in video

“…It is well known that LSTM [ 29 ] has an excellent ability to explore temporal dynamic features. However, most methods often use temporal dynamic features directly to predict the action.…”