深度复合模型下的土壤重金属含量预测

Prediction of soil heavy metal content based on deep composite model

为提高土壤重金属含量预测的准确性,提出一种深度复合模型(DCM)。以径向基神经网络(RBFNN)为基础,将基于双曲正切函数变换的概率调整和容忍准则引入遗传算法,用于RBFNN输出层参数的生成,将随梯度正负值变化而自适应调整的学习率运用到均方根反向传播(RMSProp)算法上,用于RBFNN监督学习过程中参数的优化。结合武汉市6个新城区的农田土壤重金属含量数据进行仿真预测,验证了该模型较RBFNN等几个对比模型具有更高的预测准确性。

To improve the accuracy of soil heavy metal content prediction,a deep composite model(DCM)was proposed.The model was based on radial basis neural network(RBFNN),and the probability adjustment based on hyperbolic tangent function transformation and tolerance criterion was introduced into the genetic algorithm for the generation of RBFNN output layer parameters,the learning rate which was adaptively adjusted with the change of the positive and negative gradients was applied to the root mean square back-propagation algorithm(RMSProp)to optimize the parameters of the RBFNN supervised learning process.The model was applied to the prediction of soil heavy metal content in six new urban areas of Wuhan.Experimental results verify that the proposed model has higher prediction accuracy than several comparative models.