Abstract:

To control the balance between exploration and exploitation, a reinforcement learning algorithm based on information entropy is proposed. A new state importance measure is defined from information entropy and is applied to measure the interrelatedness between state and objectives. Based on this new measure, an exploration mechanism is designed for adjusting the balance between exploration and exploitation adaptively. In addition, an autonomic reduction method is obtained by setting the variable threshold of measure, the size of state space can gradually reduce to a small and adapt space, which will save computing resource and accelerate learning speed. Simulation results indicate the good learning performance of the presented reinforcement learning algorithm.