Abstract: In order to avoid the random and uncertainty of the assessment method of reservoir bank stability , the artificial neural network(ANN) method with the function of disposing the nonlinear relation was applied to judge the grade of stability. A three-layer BP network model was established with 15 input nodes, 31 nodes in hided layer and 4 output nodes . In this network model, the BP elastic algorithm（RPROP）was adopted and the initialized weight value and valve were optimized and this model has realized the nonlinear reflection of the network and has the fast speed of convergence. This established method of the three layers of BP network is an effective method worthy of popularizing. Case study in the upstream banks of reservoir of Three Gorges shows that the judgment of stability grade of reservoir bank with this BP network mentioned above was similar to the results of conventional calculation methods.

Key words: artificial neural network, the BP elastic algorithm, the stability of reservoir bank