Differential evolution variable parameter vector emotion deep reinforcement learning power generation control method

Abstract

The invention provides a differential evolution variable parameter vector emotion deep reinforcement learning power generation control method. The method is composed of two differential evolution variable parameter emotion deep reinforcement learning components, and each component is used for the power generation control of an integrated energy system by combining the differential evolution, artificial emotion and deep reinforcement learning. According to the method, firstly the differential evolution is used for jumping out a local optimal solution; secondly, the sentiment output value of the artificial sentiment is calculated by using a variable selected by differential evolution; then, the deep reinforcement learning updates a learning rate and an action value by utilizing the emotion output value; and finally, the state variable and the output active power of an energy storage device are respectively learned by two deep reinforcement learning modes, and the output of the state variable and the output active power is used as the result of the proposed method after vector operation. The method can solve the problem that deep reinforcement learning is liable to fall into a local optimal solution, can improve the adjustment precision of automatic power generation control by considering the energy storage, reduces the power generation cost and carbon emission, and accelerates the realization of a carbon neutralization target.

Description

technical field [0001] The invention belongs to the field of power generation control of an integrated energy system, relates to an artificial intelligence-based method, and is applicable to the power generation control of an integrated energy system. Background technique [0002] With the "dual carbon" goal of achieving carbon peaking by 2030 and carbon neutrality by 2060, traditional thermal power units are facing shutdowns and upgrades, and the proportion of thermal power will decrease year by year, while wind power, photovoltaic power generation and other new The proportion of energy power generation will increase. In addition, the development of energy storage technologies such as pumped storage power stations and battery energy storage systems will add another boost to the realization of the "double carbon" goal. However, the randomness and volatility of wind power and photovoltaic power generation, the operation of energy storage devices such as pumped storage power ...

AI Technical Summary

Problems solved by technology

However, the number of hidden layers and neurons in the deep neural network is difficult to determine in advance, and the training of the deep neural network also requires a large amount of data. A deep neural network with excellent performance needs to be trained for a long time

In addition, reinforcement learning needs to seek the optimal action through interaction with the environment, and there is a problem of long trial and error time; when the agent is divided into too many states, it will also produce a "dimensionality" disaster; reinforcement learning may also fall into local Optimal solution

Images