A Kind Of Sequential Decision Method Based On MCTS With Its Application In The Acupoints' Ranking Schema

The traditional sequential decision method aimed at modeling the decision process and decision steps to obtain the optimal decision sequences.However,the traditional sequential decision modeling process required high certainty and accuracy of the objective function and the decision nodes,and the algorithm of sequence search mostly focused on blind search and heuristic search.Most of the search algorithms were global or local search under the constraint condition and the objective function unchanged,where the random characteristics in the search process were rarely considered.In recent years,Monte Carlo tree search algorithm(MCTS),a kind of dynamic search algorithm for stochastic game environment,provided a powerful way for the solution of this topic.MCTS is a framework of reinforcement learning algorithms,which is suitable for solving random sequential decision and nodes' searching problems in dynamic environments.However,the algorithm was only applied to the game-type search process and other "zero-sum problems".There were few sequential decision search problems studied the knowledge constraints under the participation of experts.In addition,traditional MCTS algorithms were often suffering the large search ranges and untimely convergence due to their randomness and tree scalability.In order to solve the problem,firstly,we proposed an MCTS sequential decision algorithm based on the improved tree search strategy combining the optimizing the search performance of traditional MCTS,as well,gave a detailed solution flow and discussion.Then,a single-machine scheduling problem was used to verify the advantages of the proposed method;secondly,based on the evaluation module of MCTS,a hybrid evaluation module was proposed.The module had two parts:the first part was the hybrid knowledge of fusion group decision experience knowledge and partial deterministic decision sequential segment as static constraint,the second part was the evaluation model based on the actual objective data is used as the dynamic constraint;Finally,the proposed MCTS algorithm based on hybrid evaluation module was applied to the problem of acupoints' ranking schema for a type of post-stroke dysphagia.Compared with other traditional sequential decision searching algorithms,we verified the feasibility and effectiveness of the proposed method.This work provided a reference for the acupuncture diagnosis of expert physicians and addressed a theoretical basis for the standardized training work of acupuncture programs for young physicians,as well,established a methodological support for the intelligent acupuncturists.