基于q学习的生物序列比对方法-计算机软件与理论专业论文.docxVIP

山东师范大学硕士学位论文的SAQL较单纯的SAQL有改善。 山东师范大学硕士学位论文 的SAQL较单纯的SAQL有改善。 关键词Q学习，序列比对，Agent，先验知识 山东师范大学硕士学位论文Abstract 山东师范大学硕士学位论文 Abstract Sequence Alignment is one of the common problems and also one of the most important tools in bioinformatics research．As the base of bioinformatics msearch such as gene identification， molecule evolution and life origin，it compares the similarity of biology data to infer their functions,structures and evolution informations． Reinforcement learning is壮unsupervised learning technology,by which the agent can find optimal behavior sequence and perform on—line learning．So reinforcement learning is recognized as an ideal technology to construct intelligent agent．Q-learning algorithm is the most popular reinforcement—learning algorithm,but the algorithm exist some problems．Firstly,Watkins Q-learning algorithm USeS greed strategy to select the action．This strategy is a kind of shortsighted strategy．Secondly,when the state space and continUOnS action space are very large，the learning speed of Q-learning algorithm is very slow．In this thesis，we improve and e】【tend Q—learning algorithm by k step and prior knowledge system．We prove a new a approach sequence alignment base on叫earning．Give the formula approval of the time complexity and the space complexity, alsoprovethetimecomplexityandthe spacecomplexity reduceto(0㈣)by experiment． The main work is as follows： 1． We proved a new approach sequence alignment base on Q—caming(SAQL)． Look on finding the optimal Sequence Alignment compare resdt aS the progress of the Agent finding the optimal strategy．The state set is used as the expression of the nucleotide(AⅡlino Acids) and the gaps inset into the alignment aimed t0 get the optimal Sequence Alignment．Give a score for every action of the Agent as the immediately rewards by gap penalties and score matrix，count all the immediately rewards of every strategy as the expected profits，select the strategy which has the largest expected profits to direct next action．