The Improvement And Application Of Reinforcement Learning Algorithm Research

Reinforcement learning in the 1990 s with the integration of operations research, control theory, made some breakthrough in terms of control theory and algorithm research, laid the theoretical basis of reinforcement learning, and rules in intelligent control, robot system and analysis of prediction and other sequential decision making in the success of applications. Reinforcement learning as a kind of no mentor machine learning method, through trial and error with environment constantly learning, improved their cognitive ability to the environment. Although the reinforcement learning algorithm in a lot of great success, but still faces in the problem solving exploration and utilization of dilemma, dimension disaster, the problem such as slow convergence, there are still many problems worth studying. This topic selection of reinforcement learning algorithm of classic Q_learning algorithm, the original algorithm was improved, and put forward their models and insights, introducing heuristic reward function, the improved algorithm is applied to path planning, and the effectiveness of the algorithm is verified by simulation experiment.This paper introduces the development history of reinforcement learning, and research status at home and abroad to introduce, points out that the reinforcement learning at this stage, the main problems in simple framework introduces the main content of this article and chapter.Secondly, in view of the reinforcement learning algorithm used in this paper introduces the related theory, technology and relevant models, and some of the commonly used reinforcement learning algorithm is made a preliminary introduction and process.Third, to use Q_learning learning algorithm introduced in this paper and gave detailed proof, Q_learning algorithm learning time is too long, pointed out the defects of slow convergence speed. Is proposed to extract features from the environment, with the aid of man’s experience and the problem of background can be a good design inspiration function and into reinforcement learning, improve the learning efficiency of the algorithm, speed up the convergence of the algorithm and improve the agent’s ability to learn the knowledge from the environment, and the simulation test and verify.Finally, to summarize the work of this paper, in combination with related literature and books, and gives the development direction of next step.