[4F2-OS-30b] OS-30

[4F2-OS-30b-05] Non-invasive Learning Data Ectraction of Step Response during Facility Ordinary Operation

Application of DNA Analysis Method to Time-series Data Detection

〇Chuzo Ninagawa^1,2, Keita Suzuki² (1. N Laboratory, Inc., 2. Interlink, Inc.)

Keywords:contorol modeling, time-series data, DNA Analysis

This paper shows an example of practical learning data mining method for step response neural network modeling during ordinary operation of the building air-conditioner facility. In order to avoid invasive step input control command, we applied a DNA analysis technique, i.e., reakpoint method, to extract step-response-like data from the ordinary operation data. Then, our modified SMOTE method was applied to strengthen the under sampling patterns of the raw data. Our method not only avoided intervention to the ordinary operation but also reduced work time by approximately half of that of step resoponse tests.