基于双向长短期记忆网络组合模型的水质预测

Water quality prediction based on a composite model of bidirectional long short-term memory networks

水质是生态环境保护和水资源管理的重要指标之一,准确预测水质对于保障水环境安全和可持续利用至关重要.本研究提出了一种基于双向长短期记忆网络(Bi-directional Long Short-Term Memory,BiLSTM)的组合水质预测方法.该方法首先采用鱼鹰优化算法(Osprey optimization algorithm,OOA)改进的变分模态分解(Varational Mode Decomposition,VMD)方法将水质时间序列数据分解为多个本征模态函数(Intrinsic Mode Functions,IMF),以便捕捉不同时间尺度的波动特征;其次通过麻雀搜索算法(Sparrow Search Algorithm,SSA)优化后的双向长短期记忆网络进行单个IMF的预测;最后将各模态预测结果叠加融合得到最终预测结果.结果表明,该模型在黑河张家桥监测站点数据集上溶解氧、氨氮、总磷、总氮的均方根误差分别为0.162、0.002、0.003、0.063,校正系数分别为0.991、0.979、0.977、0.994;与BiLSTM和OOA-VMDBiLSTM相比,均方根误差分别降低了0.393、0.005、0.004、0.376和0.236、0.004、0.003、0.148,校正系数分别提高了0.090、0.131、0.046、0.271和0.043、0.090、0.040、0.059.在其他检测站点的实验中,该方法也取得了良好的效果,进一步验证了其准确性和可行性.该方法作为水环境管理和决策的一种有效手段,能够帮助提高水质监测和水资源管理的效率.

Water quality serves as a pivotal indicator in the realm of ecological conservation and water resource governance.Accurate prediction of water quality is essential for ensuring water environment safety and sustainable utilization.This study proposes a combined water quality prediction method based on bidirectional long short-term memory(BiLSTM)networks.Firstly,the water quality time series data is decomposed into multiple intrinsic mode functions(IMFs)using the variational mode decomposition(VMD)method improved by the osprey optimization algorithm(OOA)to capture fluctuation characteristics at different time scales.Secondly,the bidirectional long short-term memory network optimized by the sparrow search algorithm(SSA)is used for the prediction of individual IMFs.Finally,the predicted results of each mode are aggregated and fused to obtain the final prediction result.Experimental results on the dataset from the Zhangjiaqiao monitoring station in the Heihe River show that the root mean square errors for dissolved oxygen,ammonia nitrogen,total phosphorus,and total nitrogen are 0.162,0.002,0.003,and 0.063,respectively,with correlation coefficients of 0.991,0.979,0.977,and 0.994,respectively.Compared with BiLSTM and OOA-VMD-BiLSTM,the root mean square errors are reduced by 0.393,0.005,0.004,and 0.376,and 0.236,0.004,0.003,and 0.148,respectively,while the correlation coefficients are improved by 0.090、0.131、0.046、0.271和0.043、0.090、0.040、0.059,respectively.This method has also achieved good results in experiments at other monitoring stations,further verifying its accuracy and feasibility.As an effective tool for water environment management and decision-making,this method can help improve the efficiency of water quality monitoring and water resources management.