Research On Intelligent Decision Model Based On Deep Reinforcement Learning

Nowadays,artificial intelligence plays a role in all areas of our lives.The products of artificial intelligence,such as automatic driving,face recognition,intelligent robots,etc.,are all showing its strength.Machine learning is the foundation and core of artificial intelligence,and reinforcement learning is a promising direction in the field of machine learning research.Reinforcement learning solves the problem that other machine learning algorithms rely too much on the number of samples by making agents interact with the environment to generate samples.The reinforcement learning algorithm draws experience from the generated samples,continues the interaction process according to the experience,improves and updates the decision-making strategy through this self-learning way,and gradually achieves the optimal decision-making effect.Deep reinforcement learning combines the perception ability of deep learning on the basis of strengthening the excellent decision-making ability of learning,and extracts the state of agent by using neural network,which makes the algorithm more powerful.Since it was put forward,deep reinforcement learning has made remarkable achievements in theory and application,such as DQN algorithm,alpha go,etc.,which makes deep reinforcement learning algorithm further applied in various fields.Reinforcement learning can make the computer make intelligent decision without model and supervision.It has the advantages of wide application range,strong generality,independent of samples,and has broad application prospects.Therefore,the research on deep reinforcement learning has important theoretical value and engineering application value in the field of intelligent decision-making.Based on the deep reinforcement learning algorithm,this thesis studies the intelligent decision-making in path planning and target assignment.In the path planning problem,according to the complexity of application scenarios,this thesis formalizes four path planning problems with different environments.In view of the above four problems,Markov decision process modeling is carried out respectively,and then multiple reinforcement learning and deep reinforcement learning decision algorithms are proposed or implemented,which are verified by experiments.In order to explore the feasibility of deep reinforcement learning and reinforcement learning algorithm in path planning.According to the research route from simple to complex,this thesis trains the intelligent decision-making model of path planning with excellent decision-making effect in different environmentsIn the problem of target assignment,this thesis presents the problem as the problem of target assignment without considering the time factor and the problem of target assignment with considering the time factor,and solves the problem by genetic algorithm and deep reinforcement learning algorithm.Through the comparison of the results,the feasibility and effect of the deep reinforcement learning algorithm in the target assignment problem are explored.First of all,aiming at the problem of target allocation without considering the time factor,this thesis uses genetic algorithm and microbial genetic algorithm to model and solve.Among them,the microbial genetic algorithm can calculate the optimal solution of the target allocation problem without considering the time factor faster.Secondly,aiming at the problem of target allocation considering time factor,this thesis models Markov decision-making process,explores the decision-making model of target allocation based on deep reinforcement learning algorithm,and provides another solution to the problem of target allocation.In this thesis,the intelligent decision-making scheme of multiple path planning and target allocation based on deep reinforcement learning is implemented,and a large number of experiments are carried out to verify it.The experimental results show that the modeling methods and solutions proposed in this thesis can achieve the expected goals,and provide support for the next step in the application and deployment of complex scenes.