Ozone monitoring in laboratory environments is essential for ensuring personnel safety and maintaining stable experimental conditions, particularly in enclosed facilities where ozone may accumulate during high-energy radiation operations. This study investigates the short-term prediction of ozone concentration using data obtained from a sensor-based ozone monitoring system deployed at the National Synchrotron Radiation Research Center (NSRRC). Ozone concentration measurements were collected using a UV absorption-based ozone analyzer and analyzed as a time-series dataset under controlled experimental conditions. Three forecasting models—Autoregressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), and linear regression—were evaluated for short-term ozone concentration prediction. Experimental results indicate that the ARIMA model provides superior predictive performance for the small-sample dataset used in this study. In the Right direction, ARIMA achieved R2 values of 89.5%, 86.3%, and 81.1% at distances of 5 cm, 10 cm, and 15 cm, respectively, while also demonstrating stable performance in the Up direction. The results highlight the effectiveness of classical time-series models for sensor data analysis in environments with limited sensing data. The proposed framework demonstrates the potential of integrating sensing devices with predictive data analytics to support real-time environmental monitoring and safety management in laboratory facilities.
Wen et al. (Mon,) studied this question.