A parameter adjustment method of gwlf model based on deep reinforcement learning

Abstract

The invention discloses a GWLF model parameter adjustment method based on deep reinforcement learning, comprising the following steps: the deep reinforcement learning model generates GWLF model parameter values ​​based on the local optimal NSE initialization state, and the GWLF model uses meteorological data sets and GWLF model parameter values ​​to calculate Generate NSE coefficients and pass them into the deep reinforcement learning model; among them: the state adjustment module selects and executes action a based on the current state based on the neural network, and then changes the state from s to s'; the calculation reward module calculates through the NSE coefficients corresponding to the previous state and the next state Action reward r; the step size adjustment module attenuates the action step size based on the accumulated rewards of each round; the memory pool stores the updated state s, s', action a and reward r at any time; the neural network module periodically updates the memory pool Sampling learning is performed to update neural network parameters to improve network decision-making ability; the invention improves the speed of adjusting GWLF model parameters, optimizes NSE coefficients, and improves the effect of GWLF model.

Description

technical field [0001] The invention relates to a method for improving the hydrological prediction ability of GWLF model parameters, in particular to a method for adjusting parameters of a GWLF model based on deep reinforcement learning. Background technique [0002] Reinforcement learning (Agent), the agent (Agent) receives the environment (Environment) state s, selects the corresponding action a according to the strategy and acts on the environment, the environment state transfers to the next state s', and at the same time gives the return reward r, intelligent Through continuous interaction with the environment, continuous trial and error, the body finally learns experience and strategies, and then guides subsequent actions. [0003] The process behaves as a transition to the next state not only with the previous state s t-1 related to, and also related to t-2 ,s t-3 ,...,s 0 related. Consider the simplification of the model, the current state s t only with the prev...

AI Technical Summary

Problems solved by technology

[0009] Aiming at the problems of the large number of parameters, large intervals, and difficult control of the precision of the GWLF model, this paper provides a parameter adjustment method of the GWLF model based on deep reinforcement learning, which can automatically adjust the parameters in the limited learning process and speed up the adjustment speed. Improved model accuracy

Images