本文提出廣義型類神經維此模式架構，並探討此模式的學習策略。在網路的架構設計上，廣義型類神經綱路模式將數學式與類神經綱路直接連結而成，而其中的數學式，可視爲綱路中的虛擬理單元，其所結合的鍵值亦可視爲虛擬鍵值，在學習的過程中不予調整。在學習策略上，因其整體架構爲－網路形態，所以可直接套用已發展完備的學習法則。其中包括了最受歡迎的向後傳遞法則。知識爲基礎的類神經綱路其模式的動態或是有物理意義的參數初值同事先給定、將可大量減少訓練類神經網路所需的資料、數據或時間，提昇了訓練效率與準確性甚至對訓練範圍以外的動臢瀆測，仍有一定的品質，而不致如傳統的類神經綱路一般地束手無策。最後，吾人成功地應用廣義型類示經綱路模式來建立饋料批次生物反應器之動態特性，經由電腦模擬驗證廣義型類神經綱路模式之準確性。
A generalized neural network modeling scheme is developed and used to model a fedbatch bioreactor. The generalized neural network scheme combines mathematic equations with standard neural networks. The mathematic equations in this scheme is a virtual processing node, and the associated connective weights corresponding to the node is also defined as virtual weights, which do not be updated through the training procedures. The well-known Generalized Delta Rule can be directly applied in training the connective weights of this network, since the structure of the generalized neural network is similar to the standard one. The results show that generalized neural network enhances the generalization capabilities of a standard neural network model. The approach is more efficient in training procedures, and produces more accurate and consistent predictions.
