應用演化式計算為基之幅狀基底函數神經網路於工業電腦銷售量預測之研究

Application of Evolutionary Computation-Based Radial Basis Function Neural Network to IPC Sales Forecasting

陳振耀(Zhen-Yao Chen)

指導教授 ： 郭人介 教授 ； 胡同來 教授

國立臺北科技大學/管理學院/工商管理研究所/博士(2010年)

摘要

由於預測在實務應用上可確保產能與適恰存貨量的有效配置，其的確是個至關重要的因素。而在採取線性預測方式的自廻歸整合移動平均式模型中，會受到實體世界於預測複雜資料問題上的限制，致使有其他的一些方法被發展出來並克服了在非線性預測上的挑戰。因此，本研究即為了預測非線性資料而欲發展出三個以整合性演化式計算為基的演算法，藉此來訓練幅狀基底函數神經網路。這些以演化式計算為基的演算法包括：基因演算法、粒子群最佳化和類免疫系統。

有三個標竿的連續性測試函數被用來驗證所發展出的這三個以整合性演化式計算為基的演算法，其實驗上的結果著實亦大有可為。此外，藉由臺灣的一家國際知名工業電腦製造商所提供的工業電腦銷售資料，也被進一步的用以評估這些所發展出的演算法。而模型評估結果指出，其所發展的演算法確實能預測得更為準確。更有甚者，若再考慮國際匯兌要素，便可改善其預測結果。

關鍵字

幅狀基底函數神經網路 ； 基因演算法 ； 粒子群最佳化 ； 類免疫系統 ； 演化式計算為基之演算法 ； 工業電腦 ； 銷售預測

並列摘要

Forecasting is one of the crucial factors in practical application since it ensures the effective allocation of capacity and proper amount of inventory. Since auto-regressive integrated moving average (ARIMA) models which are more suitable for linear data have their constraints in predicting complex data for the real-world problems, some approaches have been developed to conquer the challenge of nonlinear forecasting. Therefore, for the purpose of forecasting nonlinear data, this study intends to develop three integrated evolutionary computation (EC)-based algorithms for training radial basis function neural network (RBFnn). The EC-based algorithms include genetic algorithm (GA), particle swarm optimization (PSO), and artificial immune system (AIS).

In order to verify these three developed integrated EC-based algorithms, three benchmark continuous test functions were employed. The experimental results of three integrated EC-based algorithms are really very promising. In addition, industrial personal computer (IPC) sales data provided by an international well-known IPC manufacturer in Taiwan is also applied to further assess these developed algorithms. The model evaluation results indicated that the developed algorithms really can forecast more accurately. Furthermore, if foreign exchange (FX) factor is considered, the forecasting results can be improved.

並列關鍵字

Radial basis function neural network ； Genetic algorithm ； Particle swarm optimization ； Artificial immune system ； Evolutionary computation-based algorithm ； Industrial personal computer ； Sales forecasting

參考文獻

[1] Aburto, L., Weber, R., 2007. Improved supply chain management based on hybrid demand forecasts. Applied Soft Computing 7, 136-144.

[2] Agrawal, D., Schorling, C., 1996. Market share forecasting: An empirical comparison of artificial neural networks and multinomial logic model. Journal of Retailing 72 (4), 383-407.

[3] Aguirre, A., Aguirre, L. 2000. Time series analysis of monthly beef cattle prices with nonlinear autoregressive models. Applied Economics 32, 265-277.

[4] Akaike, H., 1974. A new look at the statistical model identification. IEEE Trans. on Automat. Control AC. 19, 716-723.

[5] Ahn, B.S., Cho, S.S., Kim, C.Y., 2000. The integrated methodology of rough set theory and artificial neural network for business failure prediction. Expert Systems with Applications 18, 65-74.