摘要
目的公共卫生事件、自然灾害等突发事件会导致对某些医用耗材的需求急剧增加且难以提前预测。因此,在医用耗材需求预测中,需要分析较长时间段内的历史数据来捕捉需求趋势和周期性特征。双向记忆门控循环单元(gated recurrent unit,GRU)网络作为一种改进GRU网络,能够捕捉医用耗材需求相关历史时间序列数据中的双向依赖关系,保证需求预测精度及精细化管理效果。为此,针对医用耗材需求预测中存在的长周期非线性时序依赖捕捉困难和突发需求响应不足等问题,提出一种基于改进GRU网络的医用耗材需求预测方法。方法通过分析医用耗材需求影响变量与需求量的相关性,筛选关键影响变量,将其输入至双向记忆GRU网络,构建医用耗材需求预测模型;由模型中多尺度特征卷积层提取影响变量的多尺度时间序列特征,结合注意力机制层由其中筛选出关键时空特征,输入至双向记忆GRU层,得到影响变量与医用耗材需求间的双向时序依赖关系,解码后经由全连接层输出双向时序依赖关系相匹配的医用耗材需求预测值。结果该方法应用下,口罩5月峰值8841个,针头2月峰值30102个,可实现各类医用耗材量管理。结论该方法可依据历史相关数据精准预测出各类医用耗材的月度需求量,并结合所得预测结果实现各类医用耗材量的精细化管理,保障医用耗材的及时供应。
Objective Public health events,natural disasters and other emergencies can lead to a sharp increase in the demand for certain medical consumables,which is difficult to predict in advance.Therefore,in the demand forecasting of medical consumables,it is necessary to analyze historical data over a relatively long period of time to capture the demand trends and cyclical characteristics.As an improved gated recurrent unit(GRU)network,the bidirectional memory GRU network can capture the bidirectional dependencies in the historical time series data related to the demand of medical consumables,ensuring the accuracy of demand prediction and the effect of refined management.Therefore,aiming at the problems such as the difficulty in capturing long-period nonlinear temporal dependencies and the insufficient response to sudden demands in the demand prediction of medical consumables,a demand prediction method for medical consumables based on the improved GRU network is proposed.Methods By analyzing the correlation between the influencing variables of medical consumables demand and the demand quantity,screening the key influencing variables and inputting them into the bidirectional memory GRU network,a demand prediction model for medical consumables was constructed.The multi-scale time series features of the influencing variables were extracted from the multi-scale feature convolutional layer in the model.Combined with the attention mechanism layer,the key spatiotemporal features were screened out from them and input into the bidirectional memory GRU layer to obtain the bidirectional temporal dependence relationship between the influencing variables and the demand for medical consumables.After decoding,the predicted value of the demand for medical consumables that matched the bidirectional temporal dependence relationship was output through the fully connected layer.Results When this method was applied,the peak number of masks in May was 8841 and the peak number of needles in February was 30102,which could achieve the quantity management of various medical consumables.Conclusions This method can accurately predict the monthly demand for various medical consumables based on historical relevant data,and combine the obtained prediction results to achieve refined management of the quantity of various medical consumables,ensuring the timely supply of medical consumables.
作者
王莹洁
WANG Yingjie(Beijing Chao-Yang Hospital,Capital Medical University,Beijing 100020)
出处
《北京生物医学工程》
2025年第4期386-392,共7页
Beijing Biomedical Engineering
基金
北京市科技计划课题(Z231100000425176)资助。
关键词
改进门控循环单元网络
医用耗材
需求预测
双向记忆
双向时序依赖
improving gated recurrent unit network
medical consumables
demand forecasting
bidirectional memory
bidirectional temporal dependence
作者简介
通信作者:王莹洁。E-mail:zeofpy2@163.com。
