基于信息粒化和误差补偿的模糊系统建模及其应用

In this project, fuzzy modeling method for different types of data including numeric data, interval-valued data and linguistic data, prediction technique, model assessment and optimization are researched. The studies would help in risk assessment and therapeutic intervention for detrimental complication, such as cardiovascular diseases in maintenance hemodialysis. The main contents are as following. (1) The generation method of information granular, which can integrate various types of data together, is proposed. Based on fuzzy rules of information granular, a kind of fuzzy system is designed. Meanwhile, the approximation capabilities of fuzzy system to different types of data are investigated. And some criterions of measurements for corresponding models are also considered. (2) According to error compensation strategy, combined with contraction-expansion factor a type of fuzzy system is raised. By proposed method prediction for heterogeneous time series data are further researched. (3) In order to evaluate the risk of cardiovascular complications in hemodialysis, using collected clinical data and medical experts’ knowledge, a fuzzy system modeling method which can express the relation between the concentration of dialysate and the fluctuation range of cardiovascular indexes is built, and corresponding adjustment decisions are provided for patients with variant risk levels. This research can enrich method of hybrid data modeling and enhance efficiency of complex information utilization. Relevant results could provide references to doctors and nurses, which could help them to develop advanced timely therapy for patients to prevent cardiovascular complications in hemodialysis.

本项目探讨含有数值型、区间值型和语言值型等异构数据的模糊系统建模、预测以及模型评价与优化等问题，并将相关结果运用于血液透析治疗过程中心血管疾病的风险评估和干预调控，主要研究内容包括：(1)设计能够对不同类型数据进行融合处理的信息粒生成方法；提炼信息粒的模糊规则建立模糊系统模型；考察模型对异构数据的逼近能力，建立模型的评价准则；(2)运用误差补偿策略构建带有变论域伸缩因子的模糊系统模型，对时序异构数据进行预测和模型优化研究；(3)搜集血液透析患者临床数据，运用模糊系统建模方法确定反映透析液浓度与患者心血管指标波动范围的数学模型，结合专家知识与误差补偿模型的预测结果，开展透析中心血管疾病发作的风险评级，辅助医护人员制定对应的治疗方案。项目的研究将丰富异构数据的建模方法，提高多类型信息的利用效力，为血液透析治疗中医护人员对患者采取前置及时的心血管疾病防控措施提供参考依据。

在工业生产、经济管理以及医疗诊断等众多领域中，普遍存在着同时含有数值型变量、区间值型变量以及语言值型变量的复杂系统。本项目对多类型数据的系统建模与时序数据的预测开展了探讨，涵盖的研究内容和取得的主要结果包括：(1) 开展了基于信息粒化的模糊系统构建和性能分析，运用信息粒化技术建立隶属函数和模糊规则，实现了数据向数学模型的转化，从理论上证明了所建的模糊模型对非线性模型的逼近能力；(2) 通过覆盖度、特征度和信息粒度准则等指标，提出并设计了评估建模结果的依据，建立了适用于区间值型以及语言值型数据的建模方法，推导了相应模糊模型的概率表示，并刻画了建模数据的概率分布信息；(3) 开展了不同类型时序数据的预测方法研究，综合了回归模型和模糊系统的特征，设计了多阶段补偿策略，实现了数值型和区间值型时序数据的预测，进而利用信息粒度准则建立了对预测结果的度量方法；(4) 基于建模残差和信息粒化开展了模型结构与参数的优化，利用深度学习方法和回归模型，建立了适用于区间值数据的模型参数寻优方法；(5) 应用所建模型开展了透析患者生理指标数据的回归和分类，实现了对患者的分级和心血管指标波动范围的有效度量。本项目的研究成果拓宽了数据驱动的数学模型构建途径，丰富了对建模结果的评价依据和信息的度量方式，在工业数据建模、金融数据趋势预测以及医疗临床信息的智能决策等领域具有一定的应用前景。