土壤环境质量评价的径向基函数神经网络的模型设计与应用

Design and Application of a Radial Basis Function(RBF) Neural Network in Evaluating Soil Environmental Quality

针对土壤环境质量评价缺乏精确客观的评价数值作为目标集的难点,研究了构建起对应升序排列的训练、检验样本集。在此基础上,定义以检验集输出评价结果的排序误排率作为判定各参数组合下RBF网络性能优劣的指标,以此性能指标为依据,在MATLAB6.5环境下,用编程语句实现了应用于该质评工作的径向基函数网络的大量参数组合的全面快速寻优,克服了目前用试凑法进行参数组合优选的随机性,确保所选参数组合为使整体性能最优的组合。利用确立的网络对象,以某地土壤重金属监测值为实例,对监测点进行质评,评价结果表明:该设计好的网络的评价结果科学、稳定、表达精度高。

As there are no objective and precise quality-evaluation values in work of soil environmental quality assessment,we constructed an ascendingly-sorted training set and proving set.With the foundation of such design,we defined the ratio of wrongly-sorted samples(err) as indicating variable of the quality of a networks with given parameter values.To search optimized parameter values,commonly approach is an attempting method,which attempts several different parameters,and choose better ones as the optimized parameter values.As this method can only attempt fairly small amounts of different parameter values,its ability of parameter optimization is very limited.In this study,by using MATLAB 6.5 neural network toolbox,we are able to deal with a large amount of parameter values quickly,with the target of getting the smallest value of ratio of wrongly-sorted samples,thus optimizing searching course can be automatically performed by computer.Thinks to the present method,we are able to get the best parameter values.Finally,we use the network with the optimized parameter values we have designed to assess the soil environmental quality according to some monitoring data,the assessing results are scientifically stable and precise.