To provide a new optimization procedure and device excellent in adaptability for optimization of core design and core control, in which the state of an environment is autonomously learnt, in integrated optimization of layout and control of core fuel in a specified reactor, without using foresight information in order to perform the optimization considering the performance in operation.

This system comprises a process performing an optimization procedure comprising giving an agent itself a state structure, and dispersively solving problems while mutually interfering a plurality of state transitions; a reinforcement learning process performed so that the action main body of the agent autonomously learns the state of the environment in the above process; and a process for integrating either one or both of experimental reinforcement type reinforcement learning method and an environment identification type reinforcement learning method in the reinforcement learning process, and optimizing a means to be applied to the environment of the object by use of an external input of reward. According to this, the core design and the core control are optimized considering the performance in operation.