基于回声状态网络的水质预测方法

The Research on Water Quality Prediction Method based on Echo State Network

为了准确预测水质参数的变化趋势,基于回声状态网络(Echo state networks,ESN)对地表水质预测进行了试验。首先根据ESN的特点和训练步骤对ESN网络的重要参数进行分析;然后采用网格搜索(Grid Search,GS)算法对ESN的储备池规模、谱半径、泄漏率、正则化系数进行寻优;在此基础上,结合福建省某水库的真实监测数据建立GS-ESN水质预测模型,对该水库的溶解氧及高锰酸盐指标进行短期预测。结果表明,GS-ESN水质预测模型的准确性相对于经验调参方法有明显提高。

Surface water quality is predicted based on Echo State Networks(ESN)to accurately predict the variation trend of water quality parameters.The important parameters of ESN network are analyzed according to the characteristics and training steps of ESN.The Grid Search(GS)algorithm is used to optimize the reservoir scale,spectral radius,leaking rate and regularization coefficient of ESN.Based on the real monitoring data of a reservoir in Fujian Province,the GS-ESN water quality prediction model is established to predict the dissolved oxygen and permanganate index of the reservoir over a short period.Experimental results show that the accuracy of GS-ESN water quality prediction model is significantly improved comparing with the empirical adjustment method.