Flow rate level prediction method based on convolutional neural network deep learning

Abstract

The invention provides a flow rate level prediction method based on convolutional neural network deep learning. The method comprises the following steps that: selecting an impact factor which is potentially related to the inflow flow rate of a reservoir as an input set; carrying out sample set classification and original input dataset construction; carrying out standardized processing on the original input dataset in the sample set; building a multilayer convolutional neural network; taking mean square error minimization as a loss function to determine prediction accuracy; carrying out networkparameter training; carrying out network performance testing; checking prediction accuracy; carrying out the rolling learning training on model parameters; automatically saving a learning training achievement, and automatically updating the knowledge record of a real-time library; and through a network model, carrying out calculation to obtain a final flow rate level prediction result. By use ofthe method, low-layer characteristics are combined to form high-layer characteristic fusion, so that an objective can be subjected to advanced abstract description, a data input pattern and a spatialand temporal distribution rule can be found through automatic learning, and therefore, the method can be effectively applied to the field of drainage basin water regimen forecasting.

Description

technical field [0001] The invention relates to the technical field of basin water regime prediction, in particular to a flow level prediction method based on convolutional neural network deep learning. Background technique [0002] Basin water regime prediction is an extremely important non-engineering measure for watershed management. It mainly analyzes the water cycle of the watershed based on relevant information such as the hydrometeorological characteristics of the watershed, the conditions of the underlying surface, the distribution of the main and tributary water systems, the layout of the telemetry station network, and the measured data of rainfall and runoff. Mechanism and rainfall runoff law, and then prepare a water regime forecast plan for the key control sections to predict the runoff process that may occur within the foreseeable period of each section, and provide a basis for the development of river runoff deduction, flood control situation analysis, water inf...

AI Technical Summary

Problems solved by technology

[0015] (3) The hydrological model is very dependent on rainfall runoff data, and it is necessary to have complete multi-year continuous sequence data to give full play to the role of the model and achieve high prediction accuracy, otherwise it will be powerless

[0016] (4) Hydrological model parameters are obtained through historical data sample calibration, which has a certain timeliness. As time goes by and data accumulation, its representativeness may be affected, and the latest data must be included in the sample through manual intervention on a regular basis for modelling. Recalibration, review and update of parameters

[0017] (5) Due to the large difference in the hydrological cycle mechanism of the watershed in the wet season and the dry season, the same watershed usually needs to build different hydrological models for parameter calibration and prediction calculations in different periods, which increases the application of inflow flow forecasting to a certain extent. Difficulty and Complexity

Images