本土性硫氧化硫化桿菌數學模型的敏感度分析與類神經網路最適化高密度培養之探討

Sensitivity Analysis of the Semiempirical Model of the Indigenous Acidithiobacillus thiooxidans and the Optimal Cell Cultivation using Artificial Neural Networks

楊復強(Fu-Chiang Yang)

指導教授 ： 鄭陽助

國立臺北科技大學/工程學院/生物科技研究所/碩士(2006年)

摘要

本土性之嗜酸性硫氧化硫化桿菌（Acidithiobacillus thiooxidans）高密度培養及批次反應器的瀝取作用，常使用數學模型加以描述。在我們先前的研究中，藉由輸送現象及驅動力原理的推導，提出一個半經驗的數學模型。為了解此模型中各參數對菌體生長的影響及重要性，本研究利用敏感度分析中的模擬及Morris兩種方法進行探討。模擬的結果顯示KL、BL、CL分別與最大菌體濃度、對數生長期的生長速率、以及停滯期的長短有密切的關係。而在Morris 的敏感度分析中，我們測試在不同的So、CaCl2、MnSO4、及pH值對於菌體生長的影響。實驗結果指出，大多數的菌體生長在第14天左右即到達最大值；換言之，這四個環境因子雖然對BL及CL具有些微的影響，但在第14天後這些影響便可忽略，我們於是轉而研究這四個環境因子對於KL的影響。分析結果指出，KL對於So較為敏感；此外CaCl2對KL的影響也不容忽視，但CaCl2 與其它因子具交互作用影響且不穩定。相較之下，MnSO4的影響幾乎可忽視。另外，pH值對於KL雖具有影響，但在接下來理想培養基的實驗中，改變起始pH值對於KL幾乎沒有影響。基於以上分析，我們最後只迴歸培養基中So對於KL、BL、CL的影響。本研究的結果也證實，在本土嗜酸性硫氧化硫化桿菌最佳化培養的模擬中，敏感度分析確實可以有效率地探討環境因子如何影響模擬參數。
此外，我們也利用類神經網路模擬本土性硫氧化硫化桿菌的生長狀況及最佳生長條件。在本研究中，模擬使用的類神經網路是廣泛應用的多層前饋類神經網路，其中包含三個部份，分別是輸入層、隱藏層及輸出層，調整三者之間的運算關係與轉換函數即能模擬出相近結果。在輸入層部分，我們以KH2PO4、(NH4)2SO4、MgSO4、及So 這些培養基成分作為類神經網路的輸入值，隱層與輸出層的轉換函數為 Gaussian 和 Sigmoid 轉換函數而輸出值則是OD(菌體濃度)。實驗的結果，其R2 = 0.991 與平均相對偏差(RD) = 0.026，顯示使用類神經網路預測，其結果相當的準確。此外所搜尋到的 A.thiooxidans 的最佳化培養基為KH2PO4 = 1.0, (NH4)2SO4 = 3.5, MgSO4 = 0.65, and So = 23 (g/l) 與產生0.722 g/l 的細胞乾重。雖然進行探討的實驗數據有限，但是圖表顯示類神經網路在菌體生長的模擬以及搜尋培養基最適化上，具有良好的效果。

關鍵字

敏感度分析 ； Acidthiobacillus thiooxidans ； Morris 法 ； 類神經網路

並列摘要

Mathematical models to describe the high cell density cultivation and bioleaching of the indigenous Acidithiobacillus thiooxidans in a batch reactor are highly desired. Based on the concept of transport phenomena and driving force, we had previously developed a semiempirical model. In order to determine which parameters of this model play an important role in bacterial growth, simulation and Morris method for sensitivity analysis were further applied in this study. Simulation results show that KL, BL, and CL are closely related to the maximal cell concentration, growth rate in the exponential phase, and the residual time in the lag phase, respectively. Four factors including the initial concentrations of sulfur (So), CaCl2, MnSO4, and pH, were further analyzed using Morris methods. Most of the experimental results show that the optimal cell concentration can be obtained at the 14th day of cultivation. The effects of these four factors on BL and CL can be neglected after the 14th day. Thus, we only focused on the effects of these four factors on KL. The results indicate that So plays an important role in obtaining the maximal cell concentration. Although the effect of CaCl2 is significant, it interacts with the other factors and is unstable. In contrast, the effects of MnSO4 and initial pH can be neglected. Based on the above sensitivity analyses, we conclude that among the four factors, only So exhibits stable significant effects on KL, BL, and CL. This study also demonstrates that sensitivity analysis is an effective method to investigate how the cultivation factors influence the parameters in the semiempirical model.
In addition, artificial neural network (ANN) was used to predict the bacterial growth and the optimal growth condition of the indigenous Acidithiobacillus thiooxidans. The ANN adopted in this study is the most widely used multi-layer Feed-forward Neural Network. Our ANN consists of three layers, a three layer feed forward neural network model consisting of including an input layer, a hidden layer, and an output layers. Four parameters, the concentrations of KH2PO4, (NH4)2SO4, MgSO4, and So, were fed as input to the network, while cell concentration (OD) is the output. The Gaussian and Sigmoid transfer functions were selected for the hidden and the output layers. The resulting ANN shows satisfactory prediction of the DCW with R2 = 0.991 and mean relative deviation (RD) = 0.026. The optimal medium composition of the indigenous A. thiooxidans was further predicted to be KH2PO4 = 1.0, (NH4)2SO4 = 3.5, MgSO4 = 0.65, and So = 23 (g/l) with the optimal DCW being 0.722 g/l. ANN is an effective method for predicting the optimal medium concentrations and obtaining the maximal cell concentration .

並列關鍵字

sensitivity analysis ； Acidithiobacillus thiooxidans ； Morris method ； artificial neural network (ANN)

參考文獻

Lan, Y.-W.; Liu, H.-L.; Cheng, Y.-C. Optimized physical parameters for high cell density cultivation of the ind;igenous Thiobacillus thiooxidans by response surface methodology. J. Chin. Inst. Chem. Engrs., 2005, 36, 243-251.

Acharya, C.; Mohanty, S.; Sukla, L.B.; Misra, V.N. Prediction of sulphur removal with Acidithiobacillus sp. using artificial neural networks. Ecol. Modell., 2006, 190, 223-230.

Abasaeed, A.E. Sensitivity analysis on a sequencing batch reactor model I. Effect of kinetic parameters. J. Chem. Tech. Biotechnol., 1997, 70, 379-383.

Abasaeed, A.E. Sensitivity analysis on a sequencing batch reactor model II: Effect of stoichiometric and operating parameters. J. Chem. Tech. Biotechnol., 1999, 74, 451-455.

Asai, S.; Konishi, Y.; Yoshida, K. Kinetic model for batch bacterial dissolution of pyrite particles by Thiobacillus ferrooxidans. Chem. Eng. Sci., 1992, 47, 133-139.