類神經網路訓練程式之改善與熱力學模式之前饋式類神經網路自動訓練程式設計

Improving Neural Network Training Program and Programming of Auto-Training Feedforward Neural Network of Thermodynamic Model

張展綺(Chan-Chi Chang)

指導教授 ： 林顯聖

國立臺北科技大學/工程學院/化學工程研究所/碩士(2008年)

摘要

在本研究中解決了由本實驗室所開發之準牛頓訓練法前饋式類神經網路程式的當機問題。且類神經網路之網路輸入及輸出值不必再作正規化處理。
　　許多化工程序的模擬中，常利用到熱力學模式來得到沸點以及氣相平衡組成，但以熱力學模式計算時，需要使用相當花費時間的迭代法。故本研究希望開發一能自動訓練擬合熱力學模式數據的類神經網路副程式，然後將得到的類神經網路模式取代模擬程式中的熱力學模式，藉而減少模擬時的計算時間。
本程式使用Visual Basic 2005 開發，程式中的熱力學模式是使用NRTL ( Non Random Two Liquid)的雙成份模式。程式依參數計算相平衡數據，再利用類神經網路訓練擬合相平衡數據，訓練過程中程式自動判斷隱藏層最佳神經元個數。
本程式對18種共沸混合物及5種非共沸混合物範例作汽液平衡數據擬合的驗證，結果顯示本程式具有不錯的實用性。

關鍵字

前饋式類神經網路 ； 數據擬合

並列摘要

At this study, the crash problem of our laboratory developed program of Quasi-Newton training feedforward neural network was solved. And the normalization of input and output data of neural network model no longer required.
In many simulations of chemical engineering process, it is common to calculate the boiling point and the equilibrium composition of gas phase by thermodynamic model.But the calculation direct from thermodynamic model spends much time in iteration.At this work developing a auto-training neural network sub program, which can fit VLE data calculating from thermodynamic model to a neural network model.Then thermodynamic model is replaced by neural network model to reduce computation time in simulation.
This program is developed by Visual Basic 2005. Thermodynamic model is NRTL (Non Random Two Liquid) binary model in this program. Program according the parameters to get VLE data, and neural network training fit the data to get neural network model, program determine the numbers of neurons at hidden layer in training procedure automatically.
18 azeotrope and 5 non-azeotrope examples were used to test the applicability of the program. The result shows the program is practicable.

並列關鍵字

Feedforward Neural Network ； Data-fitting

參考文獻

[15] 林金貝，使用準牛頓訓練法之前饋式類神經網路視窗程式設計，碩士論文，國立臺北科技大學化學工程研究所，臺北 (2007)。

[1] W. S. McCulloch and W. Pitts, "A logical calculus of the ideas immanent in nervous activity," Bull. Math Biophy, Vol. 5, pp. 115-133 (1943).

[2] D. O. Hebb, The Organization of Behavior, Wiley (1949).

[3] F. Rosenblatt, "The perceptron: a probabilistic model for information storage and organization in the brain," Psychol. rex, Vol. 65, pp. 386-408 (1958).

[5] M. M. Minsky and S. A. Papert, Perceptrons: An introduction to Computational Geometry, MIT Press (1969).