Study on WN applied in ferment process

Generally, ferment process has properties of strong nonlinearity, time-variation and correlation. In many cases, simple linearization in identifying the process would not lead to satisfactory results, and the problem of convergence and local minimum occurs if a normal neural network is used for identification. To solve the problem, this paper studies the modeling and predicting of cell concentration, an important parameter in the ferment process, by using wavelet neural networks with variable wavelet units. Based on analyzing the internal relations of input data, a simple initializing method is presented to solve the initializing problem of high dimensional neural networks. Simulation results show that the method is effective and could track the variation of cell concentration well.