查詢結果分析
來源資料
相關文獻
- Hybrid Parameter Analysis using Neural Network and Genetic Algorithms Given the Finite Training Data Set
- 應用類神經網路推估區域地下水中砷污染之研究
- 類神經網路於區域地下水砷濃度的推估
- 軍事投資最適化比例之研究
- Neural Network Procedures for Taguchi's Dynamic Problems
- A Fast and Efficient Competitive Learning Design Algorithm Based on Weight Vector Training in Transform Domain
- 專家系統振動訊號圖型判別之研究
- 反傳遞模糊類神經網路於流量推估之應用
- 類神經網路(Neural Networks)的種類及其在影像處理上的應用
- C++Fuzzy類神經網路物件導向發展系統之建立
頁籤選單縮合
題 名 | Hybrid Parameter Analysis using Neural Network and Genetic Algorithms Given the Finite Training Data Set=在有限輸入訓練資料下使用類神經網路和基因遺傳演算法之混合式參數分析 |
---|---|
作 者 | 張保榮; | 書刊名 | 高苑學報 |
卷 期 | 6:2 1997.08[民86.08] |
頁 次 | 頁433-438 |
分類號 | 448.5 |
關鍵詞 | 類神經網路; 基因遺傳演算法; 倒向傳遞誤差; 迪而塔法則; 適應性評估; 目標函數; Artificial neural network; Genetic algorithms; Backpropagation error; Delta rule; Fitness evaulation; Objective function; |
語 文 | 英文(English) |
中文摘要 | 近年來,智慧型計算法被引入做為參數估計之工具,例如,類神經網路或基因遺 傳演算法。在此情況下,類神經網路之權重被當成參數看待或把參數編碼成二進位字串送入 基因遺傳演算法中。然而,在類神經網路中可能在訓練階段中遭遇停滯在局部最佳值的狀態 。對基因遺傳演算法而言,它可能碰上花了一段甚長時間來尋找最佳值的問題。在本文�堙A 我們提出一混合式系統,它能避免上述問題發生。這個混合式系統結合上述二種方法優點,即 類神經網路的到向傳遞誤差修正法和基因遺傳演算法之平行搜尋特色。更精確來說,倒向傳 遞誤差被用為目標函數來提供做適應性評估,而應用迪而塔法則去更新侯選解答。另外一方 面,演迭計算法 -- 複製、交配、突變被採用來精鍊經迪塔法則更新後之輸出資料。在有限 輸入訓練資料下(取樣的資料), 我們應用些混合式系統去測量 H 經絡的參數,並在電腦 模擬後,我們獲得不錯的結果。因此,我們得一結論是此混合式系統在有限輸入訓練資料下 (取樣的資料)可提供一個重要的貢獻在參數估計的精準度和快速收斂。 |
英文摘要 | In recent year, the methods of computational intelligence were introdu ced for the parameter estimation; for instance, the artificial neural network or the genetic algorithms. In those cases, the weights of the neural network were used as a kind of parameters; the system parameters were encoded into the binary string (chromosome) for the genetic algorithms. However, the neural network could have the trouble with sticking around the local optimum state during the training phase. And, the genetic algorithms have the problem about taking a long period of time of finally reach the near global optimum. In this paper, we presented the hybrid system to avoid the aboye mentioned problems. This hybrid system combined the advantages of both methods, that is, the parallel search on the genetic algorithms and the backpropagation error correction on the neural network. More specifically, the backpropagation error is used as the objective function for the fitness evaluation and Delta Rule is a pplied to update the candidate solutions. On the other hand, the evolution compu tation which contains three operations-reproduction, crossover, and mutation was adopted to refine the output data of Delta Rule updating process. Given the fin ite training data set, the hybrid system in this paper was applied to measure th e parameters of the H meridian, and the good results were obtained after this si mulations. Therefore, we concluded that this hybrid system have a significant contribution on the accuracy and the fast convergence of parameter estimation as provided only the finite training data set. |
本系統中英文摘要資訊取自各篇刊載內容。