查詢結果分析
相關文獻
- A Kalman Filter Type of Neural Network for Dynamic Modeling and Data Rectification
- 類神經網路應用於適應性控制之研究與設計
- 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
- 限制性卡門濾波器應用於馬斯金更洪水演算
- 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
- 臺灣汽保費率之估計--對數線性費率模式與類神經網路之比較
頁籤選單縮合
題 名 | A Kalman Filter Type of Neural Network for Dynamic Modeling and Data Rectification=卡門濾波器形式的類神經網路於動態模式化及資料調合之應用 |
---|---|
作 者 | 董欣華; 欒大年; 張德賢; 余政靖; | 書刊名 | Journal of the Chinese Institute of Chemical Engineers |
卷 期 | 27:6 1996.11[民85.11] |
頁 次 | 頁493-509 |
分類號 | 460.02 |
關鍵詞 | 卡門濾波器; 類神經網路; 動態模式化; 資料調合; Data Rectification; Nonlinear dynamic modeling; Kalman filter; Neural network; Network architecture; |
語 文 | 英文(English) |
中文摘要 | 非線性動態模式在系統監測、診斷及控制上皆相當有價值,但是一個實際上有用的模式必需能夠處理量測雜訊的問題。本文提出一個可變結構的類神經網路,此架構介於前饋式與迴饋式網路中間且其迴饋程度取決於雜訊的大小。類似卡門濾波器,我們以一個濾波器增益來調整其迴饋程度,而且此增益是由預方差及雜訊方差的比值來調整。接下來我們提出一個適合此架構的網路訓練方式。我們以一個酸鹼中和的實驗系統以及一個模擬的放熱反應器來測試此網路在動態模式化及資料調合的能力。結果顯示此網路能自我調整其迴饋結構來正確的描述動態系統而且當雜訊方差異動時此網路顯示了韌性的特質。 |
英文摘要 | A nonlinear dynamic model is useful for the purposes of monitoring, diagnosis and control of chemical processes. Since most process measurements are corrupted with different degree of measurement noises, any realistic online application of nonlinear dynamic model should take the effect of noises into account. In this work, a variable-structure neural network architecture is explored. This network architecture is a hybrid between the feedforward network (FFN) and the recurrent network (RecN). The degree of recurrence in the hybrid network depends on the noise level of the process. That is: part of the measured outputs and part of the predicated outputs are fed back to the network for the future output prediction. A procedure is proposed to adjust the feedback gain. Similar to the adjustment of the Kalman filter gain, the filter gain is changed according to an approximated noise covariance to the error covariance ratio. Furthermore, a variable back-propagation through time (BPTT) algorithm is proposed for the training of this variable-structure network. A simulated nonlinear example and an experimental nonlinear pH neutralization system are used to illustrate the effectiveness of the proposed hybrid network. Simulation as well as experimental results show that the hybrid network adapts itself to operating conditions by adjusting its structure and, moreover, good description of dynamical behavior can be achieved over a wide range of noise levels. The hybrid network is then extended toa multiple-feedback-path system for dynamic data rectification. Three neural network architectures, the hybrid network, the Elman net and the FFN, are tested on a literature CSTR example. Simulation results show that the hybrid network gives comparable performance to the Elman net at the nominal condition. However, as the noise level change, the hybrid network shows the property of robustness. |
本系統中英文摘要資訊取自各篇刊載內容。