CAJ | 학술논문

在之前的研究中使用人工神经网络进行水质指标预测已经取得一定效果，在此基础上将交叉验证应用于人工神经网络的训练，获得更加准确的预测结果。以澧水某监测站的水质实测数据作为样本，选取总磷、总氮、溶解氧等6个指标，建立水质预测模型。在运用Levenberg-Marquardt优化算法对学习样本进行优化的基础上，采用加权的k-fold交叉验证方法来构建神经网络集合，构建集合时采取三种不同的混合方式：平均值、中间值和加权累积。针对不同的指标，进行了一系列的实验，总的来说，新的预测方法与简单0倍验证相比有更好的预测结果，在所有指标中氨氮和溶解氧含量预测准确率比其他指标高。
재지전적연구중사용인공신경망락진행수질지표예측이경취득일정효과，재차기출상장교차험증응용우인공신경망락적훈련，획득경가준학적예측결과。이례수모감측참적수질실측수거작위양본，선취총린、총담、용해양등6개지표，건립수질예측모형。재운용Levenberg-Marquardt우화산법대학습양본진행우화적기출상，채용가권적k-fold교차험증방법래구건신경망락집합，구건집합시채취삼충불동적혼합방식：평균치、중간치화가권루적。침대불동적지표，진행료일계렬적실험，총적래설，신적예측방법여간단0배험증상비유경호적예측결과，재소유지표중안담화용해양함량예측준학솔비기타지표고。
In a previous study, using artificial neural network to predict the water quality indicators has made some progress. The application of cross-validation, based on the artificial network, is aimed to obtain a better result. This paper establishes the water quality prediction model based on the real-world data about one of the monitor stations on Lishui. It chooses six water indicators include Total Phosphorus(TP), Total Nitrogen(TN), Dissolved Oxygen(DO). It uses weighted k-fold cross-validation method to build a collection of neural networks, based on the Levenberg-Marquardt optimization algo-rithm, under three different combination methods:mean, median, and weighted-based. Several experiments are held, using real-world time series with different water quality index. Overall, the proposed approach achieves competitive results when compared with the simpler0-fold ANN. At the same time, the TN and DO get the best result compared with other index.