Modeling strategy searching learning method based on condition generation confrontation network

Abstract

The invention relates to a modeling strategy searching learning method based on a condition generation confrontation network. The method is characterized by comprising following steps of collecting areal state transferring sample of an environment; constructing a condition generation confrontation network model, wherein the condition generation confrontation network comprises a generator and a discriminator; by use of the real state transferring sample, training the condition generation confrontation network model until convergence so that the generator obtained through the training is an environment state transferring prediction model; generating enough quantity of path samples; and by use of a path sample updating strategy, searching parameters of the strategy model in the reinforcementlearning algorithm until the parameters of the strategy model are updated and converged. According to the invention, the method is proper in design; when the generator of the environment model is obtained, there is no need to require the additional expense to carry out the sampling of the samples during learning of the strategies; and a complex decision problem in a large-scale environment can besystematically and effectively solved.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and relates to a reinforcement learning algorithm, in particular to a modeling strategy search learning method based on a conditional generation confrontation network. Background technique [0002] Reinforcement learning (RL) is an important learning method in the field of machine learning. It mainly studies how the agent makes better decisions according to the environment at that time. one of the research areas of interest. [0003] Reinforcement learning describes the process of continuous decision-making and control of agents to achieve tasks. It does not require prior knowledge like supervised learning, nor does it require experts to give accurate reference standards, but acquires knowledge by interacting with the environment. , make action selection autonomously, and finally find an optimal action selection strategy suitable for the current state, and obtain the maximum cumulative ...

AI Technical Summary

Problems solved by technology

[0008] To sum up, although there have been some research results on the modeled policy search learning method, it still cannot effectively solve the practical problems in practical applications.

Images