基于医疗大数据结合人工智能算法在呼吸机故障识别与预防性维护中的应用  

Application of Medical Big Data Combined with Artificial Intelligence Algorithm in Ventilator Fault Identification and Preventive Maintenance

作　　者：宫昕晨 温林[1] GONG Xinchen;WEN Lin(Department of Medical Equipment,Sichuan Academy of Medical Sciences&Sichuan Provincial People’s Hospital,Chengdu Sichuan 610072,China)

机构地区：[1]四川省医学科学院·四川省人民医院医学装备部,四川成都610072

出　　处：《中国医疗设备》2025年第3期41-48,共8页China Medical Devices

摘　　要：目的提出一种基于粒子群优化(Particle Swarm Optimization,PSO)算法和反向传播(Back Propagation,BP)神经网络模型的呼吸机故障识别与预防性维护策略,旨在提高呼吸机设备管理、维修水平,为呼吸机预防性维护提供参考。方法选取2017—2023年我院使用的呼吸机日常质量控制数据、临床使用数据、环境数据等多模态数据为研究对象,介绍PSO算法,建立粒子群优化-反向传播(PSO-BP)模型,同时引入K近邻(K-Nearest Neighbor Classification,KNN)模型、支持向量机(Support Vector Machine,SVM)模型以及极端梯度提升(eXtreme Gradient Boosting,XGBoost)模型作为对比模型,并选择准确度(Accuracy,ACC)、精准度(Precision,PRE)、召回率、F1得分以及曲线下面积(Area Under Curve,AUC)对模型进行评价。结果训练后的PSO-BP模型ACC、PRE、召回率、F1得分及AUC值分别为90.05%、91.00%、89.30%、0.90以及0.88;相对于KNN、SVM、XGBoost以及BP模型,PSO-BP模型识别ACC分别提高了6.64%、4.50%、3.32%、7.35%;召回率、F1得分及AUC值在一定程度上也得到了提高。模型最优阈值为0.6768,呼吸机安全区、稳定区、危险区以及高危区区间分别为[0,0.3384]、(0.3384,0.6768]、(0.6768,0.8384]、(0.8384,1.0000]。结论通过高通量医疗大数据建立的PSO-BP模型可有效识别呼吸机故障,并可使用定量数据为呼吸机预防性维护提供参考,具有一定的理论和实际应用意义。Objective To propose a ventilator fault identification and preventive maintenance strategy based on particle swarm optimization(PSO)algorithm and back propagation(BP)neural network model,aiming at improving the management and maintenance level of ventilator equipment,and providing reference for preventive maintenance of ventilator.Methods The daily quality control data,clinical use data,environmental data and other multi-modal data of ventilators used in our hospital from 2017 to 2023 were selected as the research objects to introduce the PSO algorithm and establish the PSO-optimized BP neural network model(PSO-BP).At the same time,K-nearest neighbor classification(KNN)model,support vector machine(SVM)model and eXtreme gradient boosting(XGBoost)were introduced and used as the comparison model,and accuracy(ACC),precision(PRE),recall rate,F1 score and area under curve(AUC)were selected to evaluate the model.Results ACC,PRE,recall rate,F1 score and AUC values of the trained PSO-BP model were 90.05%,91.00%,89.30%,0.90 and 0.88,respectively.Compared with KNN,SVM,XGBoost and BP models,the recognition ACC of PSO-BP model was improved by 6.64%,4.50%,3.32%and 7.35%,respectively.Recall rates,F1 scores and AUC values had also improved to some extent.The optimal threshold of the model was 0.6768,and the safe area,stable area,danger area and high-risk area were[0,0.3384],(0.3384,0.6768],(0.6768,0.8384]and(0.8384,1.0000],respectively.Conclusion The establishment of PSO-BP model through high-throughput medical big data can effectively identify ventilator faults,and can provide reference for preventive maintenance of ventilator with quantitative data,which has certain theoretical and practical significance.

关 键 词：PSO-BP模型 故障识别 预防性维护 K近邻模型 支持向量机 极端梯度提升 高通量数据 

分 类 号：R197.39[医药卫生—卫生事业管理] TP312[医药卫生—公共卫生与预防医学]