We examine a prediction method of groundwater table fluctuations in landslide area by neural networks, through the data obtained from the Kuchisakamoto landslide, Shizuoka prefecture, for the purpose of making to be sure the properties of these methods. Obtained results are as follows.
1) We clarify that the groundwater table on the day before prediction day must be considered, through studies of some parameters needed in prediction of groundwater table fluctuations using statistical model.
2) We confirme that neural networks are efficient methods to grasp the character of non-liner functions, that “neural-net” is efficient method in case of enough teaching-data, and that “fuzzy-inference” is available method in case of not enough teaching-data, through the examination of numerical experimentation.
3) We show that considering of relation pattern between rainfall and groundwater table taking into the calculation is more efficient.
View full abstract