Parallel learning automaton optimization method based on partial mean value fusion

Abstract

The invention discloses a parallel learning automaton optimization method based on partial mean value fusion. A learning automaton firstly independently and dispersedly interacts with environment, a probability obtained a mean value fusion machine and an original probability are subjected to weighted average, and a weighted average value is assigned to all learning automations to carry out next circulation, the mean value is not directly assigned to a probability vector, and instead, the mean value is assigned to the probability vector by a certain probability. By use of the method, in an initial learning stage, an own learning direction is kept, and meanwhile, continuous correction is carried out in a learning process.

Description

technical field [0001] The invention relates to a technology in the field of information processing, in particular to a parallel learning automaton optimization method based on partial mean fusion. Background technique [0002] Learning automata is an autonomous system that adjusts its own decision-making behavior through interaction with the environment, and belongs to the field of reinforcement learning. As the space for improving the speed of a single learning automaton becomes smaller, theoretical research needs to consider the problem of improving the speed of learning automata from other perspectives. Structured learning automata has achieved certain results in speeding up learning automata by combining simple learning automata according to a certain structure, the most typical of which is the hierarchical and parallel processing methods. The hierarchical learning machine model divides a large-scale action set into small-scale action sets, which are processed by a hie...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the defects that the prior art cannot be applied in a noisy environment, and the robustness is poor; only the performance of a single automaton is considered, and it is difficult to expand, etc., the present invention proposes a parallel learning automata optimization method based on partial mean fusion. The stage retains its own learning direction, and can be continuously revised during the learning process

Images