対策案提示システムの学習におけるハイパーパラメータ調査

Hyperparameter survey on learning of countermeasure proposal system

吉川 雅紀 ; 関 暁之   ; 沖田 将一朗  ; 高屋 茂  ; Yan, X. 

Yoshikawa, Masanori; Seki, Akiyuki; Okita, Shoichiro; Takaya, Shigeru; Yan, X.

Sufficient experience and skilled personnel are required in operation of NPPs, but these are lacking in advanced NPPs. Therefore, there is a need for support systems to help operators deal with anomalies in advanced NPPs. We are developing such a system, which we call Countermeasure Proposal System, by using reinforcement learning models. By adopting reinforcement learning models, it is expected to automatically and immediately propose countermeasures against various abnormalities in advanced NPPs. In this paper, to improve efficiency of Countermeasure Proposal System, we conducted grid searching for three hyperparameters in PPO. As the result of the survey, we found pairs of hyperparamters, which provides the most effective Countermeasure Proposal System.