時系列データを活用したプラント動特性解析コード代替システムと異常同定システムの改良

関 暁之; 吉川 雅紀; 沖田 将一朗; 高屋 茂; Yan, X.

no journal, , 

By using deep learning techniques, a surrogate system for a conventional plant dynamics analysis code and an anomaly identification system had been developed. An attempt is made to improve performance of both systems by learning on time-series data obtained from the analysis code. As a result, the surrogate system has yielded improved accuracy of estimating the state of the plant over an extended period of time, whereas the anomaly identification system can now estimate not only the state of the disturbance, but also the timing of the anomaly occurrence.