Kernel Evolving Participatory Fuzzy Modeling for Time Series Forecasting
2018 IEEE INTERNATIONAL CONFERENCE ON FUZZY SYSTEMS (FUZZ-IEEE)(2018)
摘要
Evolving fuzzy models constitute a new paradigm for time series forecasting. Such models operate in dynamic non-stationary environments with a high degree of autonomy, and automatically adjust their structures and parameters as data are input. This paper suggests ePL-KRLS, an evolving fuzzy modeling approach that combines the participatory learning (PL) clustering algorithm and a kernel recursive least squares method (KRLS) for time series forecasting. While the PL clustering algorithm gives a fast and computationally efficient mechanism to update the model structure over time, the KRLS acts as an adaptation mechanism to maintain and store past knowledge in a robust and efficient manner. The effectiveness of the evolving kernel modeling algorithm is evaluated using the Mackey-Glass time series benchmark, and actual data of wind speed from three wind farms. Computational results show that the ePL-KRLS performs more consistently and accurately when compared against classic forecasting methods, and state of the art evolving models.
更多查看译文
关键词
evolving fuzzy models, time series forecasting, adaptive modeling
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络