Multi-category and Multi-standard Project Selection under Constrainedly Periodical Budget and Ambiguous Value-based Time Limit

具預算限制及模糊價值期限下之多類別與多等級之專案選擇

本研究提出一個複數模糊類神經系統（CNFS），其結合複數模糊集合、類神經模糊系統以及差分自回歸移動平均模型（ARIMA），形成 CNFS-ARIMA 模型並應用於時間序列預測之研究。為了更新CNFS-ARIMA 模型之參數，本研究提出一複合式進化式學習演算法，其結合粒子群最佳化演算法與遞迴最小平方估計之學習方法；其中，粒子群最佳化演算法用來調整系統的前鑑部參數，而遞迴最小平方估計用以更新系統之後鑑部參數。為了測試本研究所提出之方法的效能，使用兩個標竿時間序列資料集作為實驗範例。在實驗中，將比較並分析本研究所提出之CNFS（複數型）以及其傳統的NFS（實數型）之效能差異，並與文獻所提出之方法進行比較。由實驗結果可證實本研究所提出之系統方法可以獲得良好的效能。

A complex neuro-fuzzy system, using complex fuzzy sets (CFSs), neuro-fuzzy theory, and autoregressive integrated moving average (ARIMA) model, is proposed to the problem of time series forecasting. The proposed computing system is denoted as CNFS-ARIMA. To update the free parameters of the proposed CNFS-ARIMA, a novel hybrid learning method is devised, combining both the particle swarm optimization (PSO) algorithm and the recursive least squares estimator (RLSE) algorithm. The PSO is used to adjust the premise parameters of the proposed predictor, and the RLSE is used to update the consequent parameters. To test the proposed approach, two benchmark time series datasets are used. The experimental results by the proposed approach are compared with those by its neuro-fuzzy counterpart and by other approaches in literatures. The experimental results have illustrated the merits of CFSs in the proposed approach with excellent performance for the two examples of time series forecasting. Through performance comparison, the experimental results indicate that the proposed approach outperforms the compared approaches.