基于变参数粒子群的模糊时间序列模型的研究

发布时间：2019-01-20 09:40

【摘要】：对于具有不确定性或模糊性的时间序列,传统的预测方法的适应性较差,很多学者将研究重心转移到模糊时间序列预测问题上来。目前,模糊时间序列已被应用于游客数量预测、温度预测、股市预测、网络用户数量预测等诸多领域。为了提高模型的预测精度,针对1993年Song和Chrisom[1-3]提出的四步预测框架中的论域划分和模糊规则两个步骤进行优化。区间划分是模糊时间序列模型建立的基石,早期的研究中采用的大都是等间隔论域划分,但等间隔平分论域的方法存在预测精度低,具有较强刻意性等缺陷,所以后来许多学者提出非等间隔划分论域并取得了较高的预测精度。随着许多先进的智能优化算法(遗传算法、蚁群算法、粒子群优化算法)被应用到预测模型中,有效地将优化算法融入预测框架中对预测模型进行改进也成为研究热点。本文将采用变参数的粒子群算法对论域合理划分。运用粒子群优化算法得到分割点作为论域划分的端点,在此基础上定义模糊集,模糊化时间序列并建立模糊规则。接着,运用特征展开法算得预测小前提关于推理规则的特征系数,对模糊规则进行约简,优化了模糊规则；最后,通过去模糊化得到预测结果。通过参数调整,本文将带有收缩因子的粒子群优化算法与特征展开法相结合运用于Alabama大学1971年到1992年间入学注册人数和亚马逊2013年上半年的股票收盘价的预测研究,因而,验证了本文所提算法的有效性。
[Abstract]:For uncertain or fuzzy time series, the traditional prediction methods have poor adaptability, many scholars will focus on the fuzzy time series prediction problem. At present, fuzzy time series have been applied in many fields, such as tourist number prediction, temperature prediction, stock market prediction, network user number prediction and so on. In order to improve the prediction accuracy of the model, the domain partition and fuzzy rules in the four-step prediction framework proposed by Song and Chrisom [1-3] in 1993 are optimized. Interval partitioning is the cornerstone of the establishment of fuzzy time series model. In the early research, the equal-interval domain partition is mostly used, but the method of equal-interval bisection has some defects, such as low prediction precision and strong intention. Therefore, many scholars later put forward the nonequidistant partition domain and obtained high prediction accuracy. With the application of many advanced intelligent optimization algorithms (genetic algorithm, ant colony algorithm, particle swarm optimization algorithm) to the prediction model, it has become a research focus to effectively integrate the optimization algorithm into the prediction framework to improve the prediction model. In this paper, the variable parameter particle swarm optimization (PSO) algorithm is used to divide the domain reasonably. Particle swarm optimization (PSO) algorithm is used to obtain segmentation points as the endpoints of domain division. On this basis, fuzzy sets are defined, time series are fuzzied and fuzzy rules are established. Then, the feature coefficients of the inference rules are calculated by using the feature expansion method, the fuzzy rules are reduced and the fuzzy rules are optimized. Finally, the prediction results are obtained by defuzzification. By adjusting the parameters, the particle swarm optimization algorithm with shrinkage factor and the feature expansion method are combined in this paper to predict the enrollment from 1971 to 1992 of Alabama University and the closing price of Amazon stock in the first half of 2013. The validity of the proposed algorithm is verified.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：O159;O211.61

【相似文献】