서낙동강 수질 모니터링 데이터를 활용한 관정의 센서 상태 진단 모델

Abstract

In this paper, we present a pattern-based model predicting and diagnosing the salinity of ground water around the West-Nakdong River that runs through the south-eastern area of Korean peninsula. This area is significantly recognized as the possible danger of sea water intrusion while the tide is high. To build our model, we considered electric conductivity, water level and water temperature data gathered from the USN system that was deployed to 7 wells around that area. Since those wells are widely used to supply water for agricultural and domestic use, it is very important to predict each well’s salinity level in real-time before its utilization for crops and people. Though the level of salinity can be indirectly measured by electric conductivity somewhat easily, rapid change of electric conductivity caused by daily variation and seasonal effects makes it difficult to apply conventional statistical methods. Instead of using statistical methods with time-series analysis, we introduce pattern-based approach to predict sensor values and diagnose the behavior of sensor unit. With the introduction of pattern-based approach, we could better predict the salinity of well-water than conventional adaptive exponential smoothing method.