气象污染因子对心脑血管疾病急诊量影响的预报模型研究

发布时间：2019-01-04 15:25

【摘要】：心脑血管疾病死亡居死因首位。流行病学统计与病理学研究均证明了气象、污染因子的变化与心脑血管疾病事件发生之间有密切的联系。基于成熟的气象、污染预报系统进行心脑血管疾病事件发生的预报可以对特定人群进行提前干预，减少疾病的发病率与病死率。本文使用北京市4年气象、污染与医院急诊记录，通过统计方法分析三者之间的影响关系，采用人工神经网络和支持向量机两种人工智能方法建立以气象、污染因子为输入变量预测心脑血管疾病急诊量的模型。收集北京市2008-2011三甲医院急诊就诊记录与同期的常规气象、污染监测资料,如气温、气压、湿度、风速、SO2浓度、NO2浓度、Pm10浓度等。对医疗资料中的诊断结果进行规范化，并提取ICD-10编码为I00-I99范围内的条目为本文急诊量资料。对数据进行初步统计分析，发现气象、污染因子之间存在复杂的中高度线性相关关系，不适合对输入变量个数敏感的建模方法；急诊量资料存在年份增长效应，引入年份哑变量作为输入变量加以控制；输入变量与输出变量的线性相关不高，因此对所有变量进行7天平滑处理。人工智能方法对于复杂的、非线性模型有着独特的优势。经过上述处理后的1455条数据按照建模过程需要随机分为训练集、测试集和独立样本集三组。分别使用带动量因子的BP神经网络与支持向量机回归模型进行拟合实验。训练数据的特征与模型参数的选择一同影响最终的建模效果，因此本文分别比较了两种建模方法中如隐含层单元数、惩罚因此等具有代表性意义的参数变化对测试集平均绝对误差的影响情况，，进行了模型参数优化。最终选择两种建模方法下的最适合于本文数据的模型。在独立样本集上的预测效果检验上，人工神经网络模型与支持向量机回归模型的预测值序列与原值序列均呈现高度线性相关，而后者所得结果的平均绝对误差更低,且对高峰、低谷值这类少数样本的预测效能比前者好。最终选择以径向基为核函数的支持向量机回归模型为本研究中气象、污染因子影响心脑血管疾病急诊量的最适模型。
[Abstract]:Cardiovascular and cerebrovascular diseases were the leading cause of death. Epidemiological statistics and pathological studies have proved that there is a close relationship between meteorology, pollution factors and the occurrence of cardiovascular and cerebrovascular diseases. Based on the mature meteorology, the pollution forecasting system can predict the occurrence of cardiovascular and cerebrovascular diseases in advance and reduce the morbidity and mortality of the disease. In this paper, four years of meteorological records, pollution and hospital emergency records in Beijing are used to analyze the relationship between them, and artificial neural network and support vector machine (SVM) are used to establish meteorology. The pollution factor is an input variable to predict the emergency volume of cardiovascular and cerebrovascular diseases. To collect the records of emergency visits and routine meteorological and pollution monitoring data such as air temperature, air pressure, humidity, wind speed, SO2 concentration, NO2 concentration, Pm10 concentration and so on. The diagnostic results in medical data were standardized, and the items encoded by ICD-10 into I00-I99 were extracted for the emergency volume data of this article. A preliminary statistical analysis of the data shows that there is a complex linear correlation between meteorological and pollution factors, which is not suitable for modeling methods sensitive to the number of input variables. The emergent data have the effect of annual growth, and the year dummy variable is introduced as input variable to be controlled, and the linear correlation between input variable and output variable is not high, so all variables are smoothed for 7 days. Artificial intelligence has unique advantages for complex, nonlinear models. The 1455 data processed above are randomly divided into three groups: training set, test set and independent sample set according to the need of modeling process. BP neural network with driving factor and support vector machine regression model are used for fitting experiment. The characteristics of training data and the selection of model parameters affect the final modeling effect, so this paper compares the two modeling methods such as the number of hidden layer units. Therefore, the model parameters are optimized by the influence of the representative parameter changes on the average absolute error of the test set. Finally, we choose the most suitable model for the data of this paper under the two modeling methods. In the test of prediction effect on independent sample set, the prediction value series of Ann model and support vector machine regression model are highly linearly correlated with original value series, and the average absolute error of the results obtained from the latter is lower, and the average absolute error of the results is lower, and the average absolute error of the results is higher than that of the support vector machine regression model. The prediction efficiency of a few samples such as low value is better than that of the former. Finally, the support vector machine regression model with radial basis function as kernel function is chosen as the optimal model for the influence of pollution factors on emergency volume of cardiovascular and cerebrovascular diseases in this study.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP18;R122.2

【参考文献】