Flow prediction of underground water supply network integrating multi-modal characteristics
The underground water supply system of coal mine is the lifeline of safe production in coal mines,and the prediction of water flow in the water supply pipe network is the basis for the optimal scheduling of the water supply system,and the precision of prediction has an important impact on water supply dispatching.We proposed a flow prediction method for coal mine underground water supply pipe network by integrating multi-modal data features,which was different from previous methods,and fused various data modal features such as spatial topology,historical time dependence,actual underground production conditions,and cycle correlation of underground pipe network through the method of graph deep learning.Specifically,we used a graph convolutional neural network with added spatial attention mechanism to obtain the spatial topology relationship of monitoring points in the underground pipeline network,and used the gated recurrent unit in the recurrent neural network to obtain the time dependence of monitoring points,then combined the coal mine production law with the flow data of different cycles to form the final prediction results.According to the test data of a mine in Shaanxi,the proposed prediction method was able to predict the future trend of underground flow more accurately than SVM,LSTM,STGCN methods,reducing the prediction deviation by 9.3%,6.84%and 3.65%,respectively.
NETL
NSTL
万方数据
