Constructing Time Series Forecasting System Type Methods Based on the Separability of Seasonal and Chaotic Phenomena

• Main Authors: Zeng Bo-Jian, 曾柏健
• Other Authors: Ping-Teng Chang
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/964d2v

碩士 === 東海大學 === 工業工程與經營資訊學系 === 106 === Chaos is a non-linear but implied rule system and it is worth exploring in the development of prediction models. In recent years, with the development of research related to predictive models, the related research is also continuously put forward, which Combining various theories (such as fuzzy theory, neural networks) with chaos phenomenon, and grasping the variability caused by chaos phenomenon to improve the prediction accuracy. In addition to incorporating the above discussion into the distortion of predictive results caused by chaotic phenomena, and then analyzing the variability and reducing the error related research, this study will also explore another variation factor that distort the prediction results: Seasonal variation, which is characterized by regularity and periodic changes in presentation. At present, the related research based on chaos theory combined with seasonal variation is still lacking, but the influence of seasonality on forecast results cannot be ignored. This study observes that the chaos phenomenon and the seasonal variation factor have the difference between separability and inseparability. However, the existing research method does not consider the separability of the seasonal variation, and only directly performs the chaos prediction and seasonal adjustments, it is more prone to situations where the prediction results are distorted. In the forecasting system proposed in this study, in the early stage of forecasting, the data processing methods will be divided into three major types by the separability and inseparability of the seasonal variations of data. Type-I: removal of post-seasonal data is still chaotic; Type-II: After the removal of seasonal variation, the data chaos disappears; Type-III: seasonal index chaos processing, these three types of data processing methods will be applied and studied multiple prediction methods. In the prediction method, this study not only considers fuzzy set concepts and prediction methods under fuzzy and uncertain environments, proposing non-fuzzy and fuzzy prediction systems, but also discusses the concept of intuitionistic fuzzy sets derived from fuzzy theory and makes the expression of data closer to practical problems. The long-term and short-term prediction results produced by the prediction method will be compared with other traditional methods such as regression analysis and t ANN-type-model-applied chaotic forecasting methods. At the same time, this study will also examine the data of different areas (such as traffic flow, raw material production, electric power load, birth number, air pollution index, etc.) as a verification model. The research results show that the prediction methods based on fuzzy and intuitionistic fuzzy theory combined with seasonal and chaotic processing can be closer to the actual prediction results than the traditional methods. In the long-term prediction results, the prediction results of fuzzy neural methods can grasp the trend in long-term data, which represents the method of this study, can overcome the problem that if the prediction period of the time series prediction method in past research is too long, it will lead to distortion of the prediction result.