LSTM-based medium and small river short-term flood forecasting method

Abstract

The invention discloses an LSTM-based medium and small river short-term flood forecasting method, which comprises the following steps of: constructing an intelligent flood forecasting model by using historical hydrological data, and excavating medium and small river basin rainfall; the hidden information of the runoff is used for forecasting the short-term river outlet flow in the future on the basis of known or unknown future rainfall. The method comprises the following steps: firstly, preprocessing hydrological historical data, including data missing completion, normalization and the like; secondly, constructing an LSTM model, training the model through the selected training set, and adjusting parameters to improve the model precision; and finally, evaluating the performance of the modelthrough the performance of the model in the test set. The method has the advantages that the forecasting precision of the flood forecasting model based on the LSTM is superior to that of a traditional support vector machine model, and particularly the peak appearance time and peak forecasting precision of the model in the flood peak stage are greatly improved.

Description

technical field [0001] The invention relates to the technical field of data-driven water flow forecasting, in particular to an LSTM-based short-term flood forecasting method for medium and small rivers. Background technique [0002] Traditional flood forecasting models generally have clear physical relationships, but there are certain problems: the physical model is relatively complex, and it is difficult to collect all the detailed data distributed in time and space required to build the model. The hydrological process is a nonlinear process, which is difficult to accurately simulate through physical models, especially in small and medium-sized watersheds, which have complex hydrological characteristics, boundary conditions, and active human activities. It is generally difficult for models to fully consider various complex situations. Therefore, data mining technology is introduced into flood forecasting. Data mining technology starts with historical hydrological data, ado...

AI Technical Summary

Problems solved by technology

[0002] Traditional flood forecasting models generally have clear physical relationships, but there are certain problems: the physical model is relatively complex, and it is difficult to collect all the detailed data distributed in time and space required to build the model

However, due to the difficulty of learning long-distance information, RNN may encounter gradient disappearance / explosion during the training process, so the LSTM (Long Short-Term Memory) method is proposed to solve the above problems

Images