针对毁伤预测中神经网络超参数设置及调试过程较为复杂的问题,提出一种基于特征选择结合群体智能(feature selection and swarm intelligence algorithm,FS-SIA)的超参数优化方法,用于在毁伤预测中对神经网络进行超参数的搜索和优化.首先,通过多种特征排序方法确定毁伤特征的重要性,选取公共的特征偏序子集用于模型训练.其次,针对具体的神经网络模型,分别采用多种群体智能算法进行超参数的搜索和优化.最后,得出特征集性能最优的超参数训练模型.实验结果表明,相较于未经特征排序而单纯采用群体智能算法的其他超参数优化模型,所提方法在毁伤预测中具有更快的收敛速度和更高的准确率.
A Hyperparameter Optimization Method for Damage Prediction Neural Network Based on FS-SIA
To address the complexity of neural network hyperparameter setting and tuning process in dam-age prediction,a hyperparameter optimization method was proposed based on feature selection and swarm intelligence algorithm (FS-SIA). The method was utilized for search and optimization of neural network hyperparameters in damage prediction. Firstly,the importance of damage features was determined through various feature ranking methods,and a common partial subset of features was selected for model training. Secondly,for the specific neural network model,various swarm intelligence algorithms were employed separately to conduct hyperparameter search and optimization. Finally,the hyperparameter trained model was obtained which performed optimally for the feature set. The experimental results showed that the pro-posed method had faster convergence speed and higher accuracy in damage prediction than other hyperpa-rameter optimization models using swarm intelligence methods alone without feature ordering.
万方数据
维普
