Uso de neuro-fuzzy na avaliação da suscetibilidade de escorregamento de taludes.

Abstract

This paper is an application of Computational Intelligence in the Geotechnical Engineering, with the use of Neuro-Fuzzy Technique to indicate the susceptibility of landslide slopes in the city of Rio de Janeiro, from visual inspection. In this work, the susceptibility corresponds to the possibility of slipping without considering the damage related to the event. Adopted as the output variable Forecast Slip (PE Previsão de Escorregamento) with three adjectives that correspond to susceptibilities High, Medium and Low. The methodology used was to initially build a database with preliminary information for stability analysis, indicating the slip constraints related to geomorphology, rainfall, drainage capacity, vegetation, and occupation with their respective degrees of susceptibilities slip obtained on a set of the Reports of Condition Geo Rio. The database was implemented in a Neuro-Fuzzy algorithm. Several tests were conducted with the changes of model parameters Neuro-Fuzzy for a combination of conditioning factors of slipping and refinement of the database. The tests showed a decrease of the error provided by the program with increasing types of constraints used in the training, which allows us to infer that the slip occurs by a complex relationship between various conditioning factors. The database used in the tests discontinuities present in relations between the different conditions, ie, for the same range of height values of the slope, it is not possible to obtain a relationship for all the tracks of another condition, and even for all tracks from Forecast Slip. The PEs obtained in the validation of the model values were close to the desired only in sets of variables used for training. The model was not able to present values of susceptibilities in the range of values used in training for the combination of variables with small noise, which presents the need to expand the database so quantitatively and qualitatively in order to cover the discontinuities presented in relations between the variables.