Abstract: Slope displacement monitoring plays a vital role in the field of geology and civil engineering,where changes in slope displacement provide a key basis for predicting landslides and slumps.Traditional neural networks usually use matrices or tensors to process data in slope displacement prediction,ignoring the spatial correlation between monitoring point.Graph neural networks have been gradually applied to the field of slope displacement prediction because they can effectively capture the spa tial relationships between nodes in graph-structured data.However,traditional graph neural networks are often designed with only one graph structure when applied to slope displacement prediction,which cannot capture the dynamic changes of spatio temporal relationships between monitoring points in different time periods.To address these problems,a slope displacement pre diction model based on balanced graph structure is proposed,which combines graph structure learning and time series predic tion,and is able to capture the potential spatio-temporal relationships between displacement sequences of multiple monitoring points to improve the model's ability to model complex relationships.In the prediction model,the multi-graph generation net work (MGN) and the graph selection module are introduced.The MGN can adapt to the dynamic changes of the spatio-tempo ral relationships of the monitoring points in different time periods to make the model more flexible,and the graph selection module selects the optimal graphs from the set of the graphs generated by the MGN to improve the efficiency of the model.In order to reduce the computational cost,"Smooth Sparse Unit" is introduced to sparsify the graph structure.Taking the mine slope displacement monitoring data of a cement plant in Shijiazhuang as an example,the results show that the proposed predic tion model can greatly reduce the calculation cost while ensuring the prediction accuracy,and meet the needs of slope monito ring and early warning.

Key words: slope,displacement prediction,balanced graph,spatio-temporal prediction