[2Q1-J-2] Machine learning: models for prediction

[2Q1-J-2-03] Study on the relationship between data non-linearity and dam inflow discharge prediction accuracy

〇Masazumi Amakata¹, Takuto Yasuno¹, Junichiro Fujii¹, Yuri Shimamoto¹, Junichi Okubo¹ (1. Yachiyo Engineering Co.,Ltd.)

Keywords:dam inflow discharge, LSTM, data non-linearity

We aim at improving the predictive precision of the dam inflow discharge using the correlation between the upper stream data and the lower stream data. When we handle the correlation between the upper and lower stream data, we focus on the non-linearity the data have and we have to change the way of modeling according to the degree of the non-linearity. In this paper, we gave the sequential data generated by the distributed flow analysis model the deviation in order to express the non-linearity of the data. After that, we selected LSTM which is a kind of deep learning network and made the predictive model of the dam inflow discharge learning the non-linearity data. As the result, we knew that the deviation which the normal observation values have doesn’t influence the complexity of deep learning model.