AIを用いた異常予兆検知システムの開発 （2）２段階オートエンコーダの発電所データでの評価
概要
In power plants, a large number of sensors are installed and the measured values are used to control and operate the plants. Although operators monitor the measured values for daily operation and maintenance, the number that can be visually confirmed is limited. So far, we have proposed two-stage autoencoder for anomaly detection. To evaluate its effectiveness, it and simple autoencoder as a comparison method were applied to the operational data of a power plant. As a result, deterioration of a thermometer in the boiler combustion chamber was detected by both methods. The false positive ratio of the two-stage autoencoder was about 70% of that of the simple autoencoder. These results showed practicality of the two-stage autoencoder.
