基于长短时记忆网络的水质预测模型研究

Water quality forecast and prediction model based on long-and short-term memory network

在当前水质数据急剧增加的背景下,为了挖掘水质时间序列中的更多信息,提升水质预测的精度,构建了基于缺失值填补算法和长短时记忆网络(LSTM)相结合的水质预测模型。通过缺失值填补算法进行水质数据的缺失值处理,利用LSTM网络分别构建不同水质参数的预测模型,以太湖水质监测数据为样本,对模型进行检验。结果表明,基于缺失值填补算法-LSTM的水质预测模型适应性强,相较传统SVM、BP神经网络、RNN、LSTM模型预测精度更高,对水环境保护具有重要意义。

This paper takes its aim to propose a water quality prediction model based on the missing data input algorithm and a long short-term memory(LSTM)neural network to solve such problems.For,as is known,water quality prediction and forecast can be of great significance for the water resource management.Most of the currently existing water quality prediction methods have limited ability to deal with the large water quality time series data,and also lack of corresponding means to deal with the missing water quality data.And,to deal with the problem,the data imputation algorithm has been proposed in this paper to fill the missing data in the water quality parameters control sequence.Then,the parameters of LSTM network have been chosen and trained to establish the water quality prediction models respectively for p H,COD,DO and other parameters.And,finally,2 experiments have been carried out to verify the performance of this model with the online monitoring data set and daily report data set of Taihu Lake as a case study sample.The first experiment built two kind of water quality prediction models.The online monitoring data set was used to establish the shortterm water quality prediction model for the next 2 h and the daily report data set was used to construct the long-term water quality prediction model for the next 3 d.The experiment results have shown that the model built in this paper could achieve accurate predictions in both prediction models,with its adaptability being strong.The last experiment is used to compare the models proposed in this paper with the traditional ones,including SVM,BPNN,RNN and LSTM.The results demonstrate that the RMSE and MAE of the models proposed in this paper are equal to0.067 and 0.03,respectively,which are the lowest among all the models,so that the prediction accuracy of the proposed model turn to be higher than others.To sum up,the water quality prediction model proposed in this paper based on the missing data imputation algorithm and a long short-term memory(LSTM)neural network enjoys the advantages of high precision and strong adaptability,which enjoys its great superiority in the water resources management to the water supply departments.