System Change Detection Method Using Recurrent Neural Networks

作者：Hayashida Tomohiro^*; Yamamoto Toru; Kinoshita Takuya; Nishizaki Ichiro; Sekizaki Shinya; Hiratsuka Naoto

来源：Electronics and Communications in Japan, 2018, 101(1): 39-46.

DOI：10.1002/ecj.12020

• time-series forecasting
• recurrent neural networks
• failure detection

A single or multiple kinds of internal or external environmental variations of the system often cause the property variation of any system under control, and the readjustment of controller parameters is required. To maintain high performance of controlling and minimize the total cost for readjustments of the controller parameters, determination of the appropriate timing for readjustment of the controller parameters is important. This paper proposes new procedure to determine the appropriate timing for the readjustments based on the time-series data using the recurrent neural networks (RNNs). A well-coordinated RNN with proper structure has high performance on the time-series data forecasting with the assistance of its internal signal feedback structure. This paper conducts some numerical experiments to verify the availability of the proposed method to some systems. The experimental result indicates that the proposed method has higher performance than other existing method with the same aim.

• 出版日期2018-1