Method for constructing diversified search strategy model based on deep reinforcement learning network

Abstract

The invention relates to a method for constructing a diversified search strategy model based on a deep reinforcement learning network, and the method is based on setting of the weight of a virtual reward, can enable different agents to access different states, and enables the intelligent agents to access different states. Once a certain intelligent agent is caught in a misleading reward, other intelligent agents access a series of states guiding the misleading reward again, the weight is a negative value, so that the signals of the virtual rewards obtained by the other agents are negative, and the intelligent agents are forced not to access the series of states guiding the misleading reward any more. In this way, it is guaranteed that different intelligent agents access different state sets, an updated search strategy model can find a second target position corresponding to global optimum after being trained; and the technical problem that in the prior art, the global optimum cannot be searched due to the fact that misleading rewards are caught when high-dimensional data are searched is effectively solved, and the probability that the agent falls into a local solution due to misleading rewards can be reduced.

Description

technical field [0001] The present disclosure relates to the field of deep reinforcement learning and the field of image processing technology, and in particular to a method for constructing a model of diverse search strategies based on a deep reinforcement learning network. Background technique [0002] With the development of artificial intelligence technology, when making decisions in the face of complex scenarios, a method of deep reinforcement learning has been proposed. Deep learning (DL, Deep Learning) is a method of representation learning for data in machine learning. Reinforcement learning (RL, Reinforcement Learning) is to build an environment model and learn an optimal strategy by exploring the unknown environment while exploring. Deep Reinforcement Learning (DRL, Deep Reinforcement Learning) combines the perception ability of deep learning and the decision-making ability of reinforcement learning, and can be directly controlled according to the input informatio...

AI Technical Summary

Problems solved by technology

[0004] However, traditional deep reinforcement learning exploration methods are difficult to deal with the problem of misleading rewards in scenarios where the input is high-dimensional data (such as the environment with images and high-dimensional vectors as states), and these misleading rewards will prevent agents from obtaining long-term Looking at higher rewards, this eventually makes the agent stuck in local solutions

Images