PSO优化BP神经网络模型在医疗设备故障识别中的应用

Application of PSO Optimized BP Neural Network Model in Medical Equipment Fault Identification

目的提出一种粒子群算法优化后的BP神经网络模型,为呼吸机监测数据和呼吸机故障建立潜在映射关系,从而为呼吸机维修和预防性维护提供参考。方法介绍了BP神经网络模型、粒子群优化算法以及粒子群算法结合BP神经网络结模型的建立过程,使用2017年1月1日至2020年12月31日我院采集的某型号呼吸机运行数据作为研究对象,按照6∶4的比例随机将故障数据集划分为训练集(246条)和测试集(164条),分别使用训练集和测试集对BP神经网络模型和粒子群算法优化后的BP神经网络模型进行训练和测试,并使用准确率、AUC值、灵敏度、特异性作为模型评判指标。结果训练后的粒子群优化的BP神经网络模型对测试集故障数据模式识别的准确率、AUC值、灵敏度、特异性分别为0.921、0.811、0.923、0.942;相对于K-NN、NBC、SVM以及BP模型,PSO-BP神经网络模型准确率分别提高了10.4%、11.0%、5.2%和9.7%,提高效果显著(P<0.05);AUC值、灵敏度和特异性在一定程度上得到了提高。结论本文提出的粒子群算法优化后的BP神经网络模型对故障预测效果良好,可为呼吸机故障诊断和预防性维护提供新的思路。

Objective To propose a BP neural network model optimized by particle swarm optimization,to establish a potential mapping relationship between ventilator monitoring data and ventilator failure,thereby providing a reference for ventilator repair and preventive maintenance.Methods This paper introduced the establishment process of BP neural network model,particle swarm optimization algorithm and particle swarm optimization combined with BP neural network model.The operation data of a signal ventilator collected in our hospital from January 1,2017 to December 31,2020 was selected as the research object.We randomly divided the fault data set into training set(n=246)and test set(n=164)according to the ratio of 6∶4.The BP neural network model and the BP neural network model optimized by particle swarm optimization algorithm were respectively trained and tested using the training set and test set.The accuracy,AUC value,sensitivity,and specificity were used as model evaluation indicators.Results The accuracy,AUC value,sensitivity and specificity of BP neural network model optimized by particle swarm optimization were 0.921,0.811,0.923 and 0.942,respectively.Compared with K-NN,NBC,SVM and BP models,the accuracy of PSO-BP neural network model increased by 10.4%,11.0%,5.2%and 9.7%,respectively,and the effect was significantly improved(P<0.05).The AUC value,sensitivity and specificity were improved to some extent.Conclusion The BP neural network model optimized by swarm algorithm proposed in this paper has a good effect on fault prediction,and can provide new ideas for ventilator fault diagnosis and preventive maintenance.