Q function adaptive learning method based on multilayer classification network

Abstract

The invention discloses a Q function adaptive learning method based on a multilayer classification network, comprising the following steps: S1, training a multilayer classification network; and S2, using the multilayer classification network for work. Through the establishment of a hierarchical classification network, adaptive partitioning of a problem space is realized. Compared with an unimproved fuzzy adaptive resonance classification network, the partitioning mode is more flexible and can be used to get a better Q value fitting function. The embodiment shows that the method has stronger performance in acquisition of an optimal action strategy.

Description

technical field [0001] The invention belongs to the field of intelligent decision-making, and in particular relates to a method for implementing self-adaptive learning of a Q function in a Q-Learning algorithm by using a multi-layer classification network. Background technique [0002] Q-learning is a kind of reinforcement learning technology, and it is the most widely used reinforcement learning technology. The goal of Q-learning is to find a Q utility function (hereinafter referred to as "Q function") for the problem space, and map the <state, action> pair to a specific utility value (hereinafter referred to as "Q value"). Once the Q-function is obtained, the optimal action strategy in any state can be determined, so this method is widely used as a solution framework for decision-making problems. However, for a decision problem with a continuous state space (hereinafter referred to as "continuous problem"), it is difficult to quickly obtain the Q function. [0003]...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that it is only suitable for problems with low dimensionality of the state space

But its problem is: the specific form of many fitting functions needs to be manually set by experienced designers

This setting depends on personal experience, and it is often difficult to form a good approximation to the "true" Q value (here refers to the theoretically accurate representation), which will reduce the quality of the final decision

Images