Multi-step daily runoff forecasting method based on meteorological information and deep learning algorithm

Abstract

The invention discloses a multi-step daily runoff forecasting method based on meteorological information and a deep learning algorithm, and the method comprises the steps: carrying out the feature selection of a meteorological data set through employing a maximum mutual information coefficient (MIC) method, and taking the selected feature as a prediction factor candidate input of a model; adopting a cross-correlation function (CCF) and a partial autocorrelation function (PACF) to select historical lag data of observed runoff and rainfall as predictive factor candidate input of the model; in the second stage, firstly performing data scale scaling on the selected predictive factor, and then dividing the data set into a training set, a verification set and a test set; in the third stage, calibrating GBRT model parameters through a grid search algorithm, and then carrying out forecasting on a test set through optimized parameters. Tests show that the GBRT-MIC can well perform runoff prediction in a prediction period, and research results have important significance in assisting a power plant in making a power generation plan in advance, reducing hydropower abandoned water, increasing hydropower electric quantity and improving the hydropower scientific scheduling level.

Description

technical field [0001] The invention relates to the field of hydrological forecasting, in particular to a multi-step daily runoff forecasting method based on meteorological information and deep learning algorithms. Background technique [0002] Runoff forecasting is the basis of reservoir optimal dispatching and plays a vital role in reservoir management and operation. With the large-scale commissioning of cascade hydropower stations on large rivers in my country, the huge hydropower system is facing very complex dispatching problems. Improper dispatching can easily lead to water abandonment. Therefore, the accuracy of reservoir runoff forecasting is required in the dispatching process. However, the current daily runoff forecast accuracy is obviously insufficient, especially for southern China. Due to the impact of strong convective weather such as typhoons, heavy rainfall in southern my country is usually concentrated within a few days. Low-precision runoff forecasts can e...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved in the present invention is to propose a multi-step daily runoff forecasting method based on meteorological information and deep learning algorithms for the problem of insufficient runoff forecasting accuracy in the optimal operation of reservoirs, from input selection, forecasting model construction and forecasting to A three-step assessment of the results proceeded to construct a daily runoff forecasting framework

Images