Abstract
A method for several-days-ahead prediction of inflow rate into a dam was investigated. We introduced a new type of variable sampling rate model which aggregated rainfall information with different time duration depending upon the characteristics of the inflow rate into the dam. As the rainfall information was appropriate-ly aggregated into the model, the proposed variable sampling rate model was able to express both short term characteristics and long term characteristics of the inflow rate into the dam. We constructed five different models depending upon the amount of the inflow rates and selectively used one model out of the five models in predicting the inflow rate into the dam. The proposed method was evaluated based on the actual inflow rates data and corresponding rainfall data during past several month at Midorikawa Dam. The proposed method gave satisfactory accuracy in the several-days-ahead prediction of the inflow rate into the dam. We further evaluated the proposed variable sampling rate model by comparing a conventional constant sampling rate model without data aggregation. The proposed model gave 20% reduction in the error of 8-days-ahead prediction of the inflow rates data. The variable sampling rate model was proved to be effective for the prediction problems in wide actual fields.
