基于模糊规则预测模型的急性高血糖诊断

摘要

We build an algorithm based on fuzzy rules for creating prediction model.In order to generate prediction model rule base, the model uses Takagi-Sugeno (T-S) structure and IF-THEN expressions to describe rules.Meanwhile, the coefficients are determined by Gaussian membership functions.The experimental data are derived from electrophysiological information of cardiac conduction systems and are divided into training set and test set.The rule base is generated using the training set and the duration of acute hyperglycemia is predicted by the test set.Meanwhile, experimental measurement outputs and prediction model outputs of test sets are compared and prediction root mean square error (RMSEP) is used to evaluate the prediction accuracies.After that, the prediction results are compared with the ones of three traditional models which are partial least squares (PLS), least squares support vector machine (LSSVM) and back propagation neural network (BPNN).The result shows that the prediction model based on fuzzy rules has the highest prediction accuracy.The model is suitable for predicting the actuation duration of acute hyperglycemia, providing guidance and advice for basic medical research and clinical diagnosis.%构建了一种基于模糊规则的算法来建立预测模型.该模型采用Takagi-Sugeno (T-S) 结构, 通过IF-THEN的表述方式来描述规则, 并通过高斯隶属函数确定模型系数, 生成预测模型规则库.实验数据来源于心脏传导系统的电生理信息, 将实验数据分成训练集和验证集, 通过训练集生成规则库, 并使用验证集来预测急性高血糖的持续时间, 对比验证集的实验测量输出与预测模型输出, 使用预测均方根误差 (RMSEP) 评价预测精度.将预测结果与偏最小二乘法 (PLS) 、最小二乘支持向量机 (LSSVM) 和反向传播神经网络 (BPNN) 三种经典预测模型的结果进行比较, 实验结果表明, 基于模糊规则的预测模型预测精度最高, 适合用来预测急性高血糖的持续时间.该模型可以为医学基础研究和临床诊断提供指导与建议.