Dadu River Runoff Forecasting via Seq2Seq

Abstract

Precise runoff forecasting is playing a very important role in flood control and economics dispatch control for hydroplant. This paper investigates the accuracy of standard long short-term memory neural network and sequence to sequence(seq2seq) in prediction of hourly, daily runoff. This paper is divided into five sections; the main contents are as follows: The first section mainly introduces the research status of some machine learning algorithms in runoff forecasting. Section 2 describes the basic principles of Long short-term machine. In the third section, the establishment process of runoff forecasting model is introduced. The basic principle of Sequence to Sequence and the design of seq2seq model in this paper are mainly introduced. The fourth section introduces the actual forecasting effect of applying Sequence to Sequence to Danba hydrological station in the upper reaches of Dadu River. Compared with the ordinary forecasting model, it proves that Sequence to Sequence has better forecasting accuracy and has certain practical application value. The fifth section gives a conclusion of this paper and puts forward the next work plan.