Application of an artificial neural network formulated to predict the outbreak of musty odor and control it

饮用水水塘爆发霉味，造成自来水霉味等严重供水问题。霉味控制是水库水质管理的重要任务之一。本文描述了人工神经网络(ANN)模型在预测日本东京地区饮用水源Watarase水库霉味化合物2-甲基异龙脑(2-MIB)暴发的新应用。利用人工神经网络模型，建立了Watarase淡水水库中2-MIB浓度绝对值的预测模型。人工神经网络模型以1992~1997年的气象条件、水质、营养盐、浮游植物和Watarase水库的运行状况等数据作为输入层数据。将ANN模型计算的2-MIB浓度绝对值与实测值进行比较，需要注意的是，ANN模型能够很好地预测2-MIB的暴发时间。对-2-MIB后3天的绝对值进行预测，相关系数为0.65，表明用ANN模型预测2-MIB浓度具有良好的可行性。

Outbreak of musty odor in reservoirs for drinking water caused serious problems with regard to water-supply such as musty taste of tap water. Control of musty odor is one of the most important tasks in water quality management of reservoirs. This paper describes the novel application of an artificial neural network (ANN) model based on the back-propagation method formulated to predict outbreak of a musty-odorous compound, 2-methylisoborneol (2-MIB), in Watarase Reservoir, which is one of drinking water resources for Tokyo area in Japan. By using ANN model, we constructed the model predicting absolute values of the 2-MIB concentrations in Watarase freshwater reservoir. As the input layer data, ANN model used various data obtained from 1992 to 1997, i.e. meteorological conditions, water qualities, nutrients, phytoplanktons, and operational conditions of Watarase Reservoir. Comparing the absolute values of 2-MIB concentrations calculated by the ANN model with those observed, it should be noted that the timing of outbreak of 2-MIB was well-predicted by the ANN model. Prediction of the absolute values of 3 days after- 2-MIB concentrations resulted in 0.65 of the correlation coefficient, thereby indicating good feasibility of predicting 2-MIB concentrations by the ANN model.