Research And Applications Of Action Strategy In Deterministic Strategy Reinforcement Learning Algorithm

In this thesis,we weaken the shortages of random strategies such as ?-greedy strategies and Gaussian strategies in exploration and exploitation in reinforcement learning.We proposes two exploration methods based on value functions.In these two methods,we use the modified value functions to estimate the state.At last,we apply our method on the Pendulum problem.The numerical experiments show that our methods perform well.