应用全光谱测量水体化学需氧量(chemical oxygen demand,COD)、硝酸盐氮(NO3-N)浓度等水环境质量指标容易受水质环境影响,检测模型与特征波长一直是全光谱检测推广关注重点.该文提出一种基于遗传算法-径向基神经网络(genetic algorithm-radial basis function neural network,GA-RBFNN)全光谱水体COD与NO3-N浓度检测方法,鉴于GA搜索能力强、随机性高的特点,对预处理后全光谱吸收数据应用GA进行特征波长选取,以RBFNN神经网络留K法训练过程中平均决定系数作为适应度函数,输出最优特征波长与RBFNN神经网络参数进行部署,从而实现水体COD、NO3-N浓度准确测量.最后,开展GA-RBFNN、偏最小二乘(partial least squares,PLS)、GA-PLS、RBFNN四种模型对 160组水样的COD、NO3-N浓度检测实验,结果表明GA-RBFNN模型对COD、NO3-N检测平均决定系数、最大误差分别为0.9964、0.9950和3.9%、4.9%,均优于其他模型,方法具有重要推广价值.
Research on GA-RBFNN full spectrum water body COD and NO3-N detection methods
The paper proposes a method for detecting the concentrations of chemical oxygen demand(COD)and nitrate nitrogen(NO3-N)in water bodies based on genetic algorithm-radial basis function neural network(GA-RBFNN)using full-spectrum measurements.Given the strong search capability and high randomness of GA,it selects feature wavelengths from preprocessed full-spectrum absorption data.The RBFNN trained through the K-fold cross-validation method,utilizes the average coefficient of determination as the fitness function.This approach optimizes the feature wavelengths and RBFNN parameters,enabling accurate measurement of COD and NO3-N concentrations in water bodies.Comparing GA-RBFNN,partial least squares(PLS),GA-PLS,and RBFNN models in experiments with 160 sets of water samples for COD and NO3-N concentration detection,the results indicate that the GA-RBFNN model exhibits average determination coefficients and maximum errors of 0.9964,0.9950,3.9%,and 4.9%,respectively,outperforming other models,thus demonstrating significant potential for widespread application.
NETL
NSTL
万方数据
