论文部分内容阅读
针对神经网络中模型可靠性问题,提出了趋势检查法的思路,采用评价指标中评价等级的影响趋势对模型进行检查,基本过程为不断调整模型参数、训练、趋势检查,直到获得最优模型。趋势检查法为一种通用方法,可用于任何基于先知经验方法的模型可靠性检查,为模型可靠性检查提供了一种新思路。对于神经网络学习样本贡献度不同的问题,采用样本加权的方法,对样本进行预处理,并将样本权值应用于神经网络的目标函数中,由此建立了加权神经网络目标函数。最后引入遗传算法来优化神经网络参数,建立了基于趋势检查法的遗传神经网络模型,并应用于实际工程中的围岩分类问题,结果表明该模型泛化能力强,具有较高的分类精度。
In order to solve the problem of model reliability in neural network, the idea of trend check method is put forward. The trend of evaluation grade is used to check the model. The basic process is to continuously adjust the model parameters, training and trend check until the optimal model is obtained. The trend check method is a universal method that can be used for any model reliability check based on the prophetic experience method, which provides a new idea for the model reliability check. For the problems of different contributions of neural network learning samples, the method of sample weighting is used to preprocess the samples and the sample weights are applied to the objective function of neural network, so the weighted objective function of neural network is established. Finally, the genetic algorithm is introduced to optimize the neural network parameters. A genetic neural network model based on the trend check method is established and applied to the classification of surrounding rock in practical engineering. The results show that the model has a good generalization ability and a high classification accuracy.