Prediction method of reservoir inflow flow based on deep learning

Abstract

The invention relates to the technical field of water regime prediction in a river basin, and discloses a method for predicting the inflow flow of a reservoir based on deep learning. The corresponding relationship between the inflow flows of the reservoir, and then the predicted reservoir inflow flow in the absence of rain, the predicted reservoir inflow flow in the rainy situation and the difference delta in the rainy situation are obtained. Through LSTM training and learning, obtain The difference of the predicted inflow flow under the condition of rain is used to obtain the final predicted inflow flow of the reservoir. The present invention is based on the deep learning-based reservoir inflow prediction method, which integrates the deep belief network and the long short-term memory network algorithm and applies it to the inflow prediction, which improves the prediction accuracy of the inflow and improves the reliability of the model. and scalability.

Description

technical field [0001] The invention relates to the technical field of water regime prediction in a river basin, in particular to a method for predicting the inflow flow of a reservoir based on deep learning. Background technique [0002] At present, the methods for predicting the inflow of reservoirs are mainly divided into two categories: methods based on traditional hydrology research methods and methods based on traditional machine learning such as artificial neural networks. In traditional hydrology research methods, various physical quantities such as soil water content, rainfall, water storage and discharge conditions in upstream and downstream reservoirs are generally calculated, and water regimes are predicted by simulating the runoff and runoff mechanism. Traditional reservoir inflow prediction models generally have clear physical relationships, but relatively speaking, there are certain problems: the physical model is relatively complex, and it is difficult to col...

AI Technical Summary

Problems solved by technology

Traditional reservoir inflow flow prediction models generally have clear physical relationships, but relatively speaking, 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, such as terrain data, river channel characteristic data, watershed soil characteristic distribution, and rainfall runoff data; in addition, as a nonlinear process, runoff production and confluence are too complex to be accurately simulated by physical models; moreover, the conditions of different watersheds vary widely, and the hydrological models used are generally different. It is difficult to fully consider all kinds of complex situations

And most of them are learning by using shallow models, which are easy to fall into local optimum, long calculation time, etc., and cannot simulate large-scale and complex mathematical calculations, so it is difficult to achieve ideal results in inbound traffic forecasting

In addition, these network models are relatively single, with a relatively simple structure, lack of analysis and design for specific problems, and poor scalability

Images