[포스터] Harmful Algal Blooms Prediction By Extreme Learning Machine in the Nakdong River, Korea |
---|
학술지명 제8차 IWA-ASPIRE
저자 이혜숙,정선아,박연정,이희숙,최광순,이승윤,김호준
발표일 2019-10-31
|
Recently, harmful algal blooms among environmental degradation issues refer to the explosive increase and high concentration of phytoplankton in ecosystem in recent years. This has become a challenge facing human society today as the phenomenon of harmful algal blooms is becoming more prevalent throughout the world. Especially, harmful algal blooms would affect the health of public. However, the high complex nonlinearity of water and weather variables and their interactions makes it difficult in modeling harmful algal blooms. Recently, extreme learning machine (ELM) was suggested by Huang et al. (2006) and reported to have advantages of only requirement of a small amount of samples, high degree of prediction accuracy and long prediction period to solve the nonlinear problems. But studies have not focused on predicting harmful algal bloom using machine learning with water quality and climate data in regulated rivers. In this study, the ELM models for harmful algal blooms in 4 weirs were proposed to predict harmful algal blooms. The Nakdong river is the longest river in South Korea, and has 8 weirs which were built in sequence from 2012. Especially, 4 weirs of the mid-lower Nakdong river have harmful algal blooms every summer causing many problems for agricultural, residential and commercial water supply. We applied ELM which originally was proposed as a learning scheme for single-hidden-layer feed-forward neural networks (SLFNs). The water quality and weather data were collected weekly from 2013 to 2016 by the Ministry of Environment, Korea Meteorological Administration. Parameters in the dataset include daily air temperature, rainfall, solar radiation, and weekly T-N, T-P, N/P ratio and harmful algal cells as ELM model input. The 50% of dataset were applied for training and another 50% of dataset were applied for testing. The prediction powers of 4 weirs had R2 of 0.70, 0.67, 0.73, 0.72 of training and 0.57, 0.81, 0.69, 0.75 of testing. The prediction powers had RMSE of 0.501, 0.567, 0.441, 0.514 for training and 0.625, 0.430, 0.543, 0.440 for testing data sets. The results indicated that the ELM was successful in the prediction and forecast harmful algal blooms in the Nakdong river. |