Assessment of liquefaction potential using neural networks

Abstract This paper presents the development of a computer model for the prediction or classification of soil liquefaction potential. The model is built by encapsulating data sets from past events using the neural networks technology, and is useful to the extent that future events conform to trends observed in the past. The data sets used for building the present model are compiled from the mammoth 1976 Tangshan Earthquake. The modified Mercalli Intensity (MMI) range of the collected data vary from VII to IX; hence, the prediction model can be used for predicting liquefaction potential for the ground shaking MMI level in the range of VII–IX. Accuracy of the model in recalling the foundation data sets and in predicting simulated events has been quantified. Limited comparison of forecasts made by the prediction model and conventional methods demonstrates that improved accuracy can be achieved.