Abstract: The selection of soil parameters in finite element simulation of foundation pit engineering has an important influence on the accuracy of final results.It is of great significance to obtain soil parameters quickly and accurately and predict structural deformation.Based on this,a method of soil parameter inversion and underground structure deformation prediction for underground space development is proposed,which mainly includes the following aspects:(1) Algorithm improvement.On the basis of sparrow algorithm SSA,an improved sparrow algorithm ISSA is constructed by introducing Tent chaotic disturbance mapping function,Cubic chaotic mapping initialization population and elite reverse learning initialization population;The performance evaluation results of ISSA and other four algorithms show that ISSA algorithm has obvious advantages in global search and convergence speed.(2) Parameter selection.Based on the modified Cambridge model which can reflect the stress state of soil in foundation pit excavation,sensitivity analysis was conducted on the parameters with greater dispersion degree,and the correlation order of parameters was κ>e₀>M>λ>ν>K₀.(3) Model construction.50 groups of orthogonal tests were designed to carry out finite element simulation considering the construction process of foundation pit,and the horizontal displacement of the connecting wall was extracted as the learning sample;The ISA-ELM network model was constructed to invert soil parameters,and the precision of the network model was tested with small samples.(4) Engineering verification.Soil parameter inversion and deformation prediction of underground structure are conducted based on monitoring data.The results show that the predicted value of inversion analysis under subsequent working conditions is relatively closer to the monitoring value,and the update of data can make the inversion and prediction more accurate.