基于多尺度卷积神经网络的绩效数据特征提取方法

Performance data feature extraction method based on multiscale convolutional neural networks

针对传统医疗机构绩效评估算法存在的主观性强、数据特征提取能力差的缺点,文中基于多尺度卷积神经网络提出一种绩效数据特征提取模型。该模型对传统卷积神经网络进行改进,使用空间化可提升效率的方法构建了胶囊网络,并使用多种尺寸不同的卷积核对数据进行训练,从而保证了特征提取的全面性。在数据训练过程中,使用熵权法对各参数指标进行权重确定,并用麻雀搜索算法进行模型参数优化。在实验测试中,参数优化后的模型预测准确率更高,在所有对比算法中,所提算法的MAE、MAPE、RMSE等误差指标最低,迭代次数也仅为7次,表明模型具有最优性能的同时训练速度也较快。

Aiming at the shortcomings of traditional medical institution performance evaluation algorithms such as strong subjectivity and poor data feature extraction ability,this paper proposes a performance data feature extraction model based on multi⁃scale convolution neural network.This model improves the traditional convolutional neural network,constructs a capsule network using a spatialization method that can improve efficiency,and uses multiple convolutional check data of different sizes for training,thereby ensuring the comprehensiveness of feature extraction.During the data training process,entropy weight method is used to confirm the weight of each parameter index,and sparrow search algorithm is used to optimize the model parameters.In experimental tests,the prediction accuracy of the model after parameter optimization is higher.Among all the comparison algorithms,the proposed algorithm has the lowest error indicators such as MAE、MAPE and RMSE,and the number of iterations is only 7,indicating that the model has optimal performance while training faster.