| 초록 |
This study developed a data-based soft-sensor to predict indoor PM2.5 from easy-to-measure indoor air quality variables. The method consists of liquid time-constant networks, a subclass of continuous recurrent neural networks represented by an ordinary differential equation system. Two types of wiring were considered: 1) a fully connected (LTC-FC) and 2) neural circuit policies (LTC-NCP). This last includes four sensory layers loosely inspired in the neural system of the nematode C. elegans. The performance metrics indicate that the LTC-NCP yielded the most accurate predictive performance accounting for an improvement compared to other linear and neural methods of 44% using RMSE, 29 – 38% using MAPE, and 50 – 80% using R2. The LTC-NCP outperformed the LTC-FC model, comparing the critical success index and false alarm rate values of 0.7132 and 0.1849.This research was supported by a grant from the National Research Foundation of Korea (NRF) funded by the Korean government (MSIT) (No. 2017R1E1A1A03070713), and from the Subway Fine Dust Reduction Technology Project of the Ministry of Land Infrastructure and Transport from the Republic of Korea (20QPPW-B152306-02). |
