Abstract: Earthwork allocation is an important issue in the design and construction of water conservancy and hydropower projects, and traditional methods, such as linear programming, large-scale system decomposition and coordination, and dynamic programming, have some limitations in practice. This paper explores a new method of using discrete Q-learning in reinforcement learning to solve the problems of earthwork allocation. We discuss the construction and solution of a Q-learning model for earthwork allocation problems and verify its feasibility through the analysis of an engineering example. This work would lay a basis for further studies on the balance of dynamic earthwork allocation using reinforcement learning.

Key words: water conservancy and hydropower project, earthwork allocation, reinforcement learning, discrete Q-learning