【摘要】 传统神经网络在进行特征提取的过程中，会丢失部分浅层特征信息，导致模型评估准确度降低。为了充分利用各层特征信息，本文在传统神经网络中嵌入特征融合模块，提出一种图像特征筛选与融合网络。该网络首先将各层特征经过线性及非线性变化进行融合，提高有效特征的表达，然后基于Pearson相关性量化特征融合权重，并根据该权重对各层特征进行加权融合。在IEEE-39节点和IEEE-145节点系统中的仿真结果表明，本文所提网络比传统神经网络具有更优的评估性能。更多还原

【Abstract】 During the feature extraction process of traditional neural networks, some shallow information may be lost, which leads to a decrease in the accuracy of model evaluation. In order to fully utilize the feature information of each layer, in this paper, the feature fusion module is embedded in traditional neural networks. An image feature filtering and fusion network is proposed. The network first fuses the features of each layer through linear and nonlinear changes to improve the expression of effective features. Then, the feature fusion weights are quantified based on Pearson correlation. Finally,weighted fusion is performed on the features of each layer based on the feature fusion weights. The simulation experiments are carried out in IEEE 39-bus system and IEEE 145-bus system. The results show that the proposed network has better evaluation performance compared to traditional neural networks.更多还原