在本論文中，針對多目標數據預測提出了一種方法，其中以複數型模糊類神經系統（Complex neuro-fuzzy system，CNFS）為計算模型，並以混合式機器學習算法—多群基因演算法搭配遞迴式最小平方估計法（記為MGGA-RLSE）用於複數型模糊類神經系統的參數學習。在數據選擇和模型構建中，以特徵選取出對目標有貢獻的特徵組成的子集。複數型模糊類神經系統使用複數模糊集（Complex fuzzy sets，CFSs）和Takagi-Sugeno（T-S）模糊If-Then規則。複數模糊集是以複數平面中單位圓盤內定義出的複數型歸屬程度來描述集合與其元素之間關係的模糊集合。由於複數型模糊類神經系統使用的複數模糊集，該模型可以同時處理最多六個實數型的目標。MGGA-RLSE結合了多群基因演算法（Multi-group genetic algorithm，MGGA）和眾所周知的遞迴式最小平方估計法（Recursive least squares estimator，RLSE），其中前者用於演化T-S中If-parts的參數，後者用於更新T-S中Then-parts的參數。在多群基因演算法中包含數個族群，每一個族群中都有很多個體。每一個個體被認為是模型中If-part規則參數的潛在解決方案。由於有多個族群，故期許多群基因演算法能找到效能好的最佳解決方案。最後，以四個實驗檢視本論文所提出的方法之優勢。;In this study, an approach has been presented for multi-target data prediction, where a complex neuro-fuzzy system (CNFS) is used as a computing model and a hybrid ma-chine learning algorithm, denoted as MGGA-RLSE, is used for the parameter learning of the CNFS. For data selection and model construction, feature selection has been used to select a compact subset of features contributive to targets. The CNFS uses complex fuzzy sets (CFSs) and Takagi-Sugeno (T-S) fuzzy If-Then rules. CFS is a fuzzy set using com-plex-valued membership degrees defined within the unit disc of the complex plane to de-scribe the relationship between the set and its elements. Due to CFSs used in the CNFS, the model can simultaneously apply on data application with up to six real-valued targets. The MGGA-RLSE integrates the multi-group genetic algorithm (MGGA) and the well-known Recursive least squares estimator (RLSE), where the former is used to evolve the parameters of If-parts and the latter is used to update the parameters of T-S Then-parts. For the multi-group genetic algorithm, several populations are involved, each population with many individuals. Each individual is regarded as a potential solution to If-part pa-rameters in the model. Due to multiple populations, the MGGA is expected to search for the optimal solution with good efficacy. Four experiments for multi-target data prediction have been used to test the proposed approach.
