Intelligent power generation control method for artificial deep emotional game reinforcement learning

Abstract

The invention provides an intelligent power generation control method for artificial deep emotional game reinforcement learning. The combines an artificial emotional reinforcement learning algorithm with a deep reinforcement learning algorithm and is used for intelligent power generation control. The method combines artificial emotion in artificial intelligence, reinforcement learning, a deep neural network algorithm and a game theory. The artificial deep emotional reinforcement learning algorithm integrates the deep neural network and the artificial emotion into a reinforcement learning framework. The deep neural network is configured to improve the probability update strategy of traditional reinforcement learning. The artificial emotion quantifies the power generation system informationby an emotional quantization function to update the learning rate, the discount factor, and the output action value of the traditional reinforcement learning. The method takes a frequency deviation and an area control error as inputs, and takes a power command of a generator set as an output. The intelligent power generation control method for artificial deep emotional game reinforcement learningis effective, feasible, robust and rapid.

Description

technical field [0001] The invention belongs to the field of power generation control of large-scale interconnected systems in electric power systems, and belongs to the field of automatic power generation control. Background technique [0002] With the rapid development of renewable energy and the continuous application of new technologies, modern power systems are facing some key challenges, such as the stability of active power and frequency. Therefore, automatic generation control has started to be widely implemented to solve these problems. The purpose of automatic generation control is to minimize the regional control error and system frequency deviation in each region. In the past few decades, people have been working on the control methods of automatic power generation control, such as proportional-integral-derivative algorithm, sliding mode control algorithm and fractional order proportional-integral-derivative algorithm. But these control algorithms with inherentl...

AI Technical Summary

Problems solved by technology

Although reinforcement learning has powerful online learning and updating capabilities, the control accuracy of the system may be low due to discrete actions, and the calculation time is too long; because reinforcement learning has a fixed reward function and a fixed learning rate, the learning efficiency of the system is low ; At the same time, reinforcement learning only has a single logical part for learning, and the degree of intelligence is low, so artificial emotional learning is introduced on the basis of reinforcement learning to optimize the performance of reinforcement learning

Images