The Investigation Of Optimization Methods In Modeling Nonlinear Electron Devices Based On Neural Network

Aiming at the problem that it is hard to model nonlinear electronic device with complex characteristic, a new method based on neural network and the idea of"model in subsections, integration of the whole"is proposed. The method of structural optimization for the BP neural network is also discussed.Firstly, the modeling methods of nonlinear electron devices at present are introduced in this paper, as well as the discussion about the advantages and disadvantages of these methods. Secondly, the principle of neural network, especially the BP network is presented. The problems existing in the structural optimization of BP network is discussed. Thirdly, the idea of"model in subsections, integration of the whole"is tried to model a Zener Diode, which has complex characteristic and is hard to be modeled with single BP network. The number of neurons in the hidden layer nerve is optimized by the structural optimization method mentioned in the second part. Finally, the experimental results are presented to show the availability of these methods.There are mainly two aspects to investigate in this paper.On the one hand, hidden node learning algorithm is introduced, and then verified by experiments. Using this method, it has been successful to select the number of optimal hidden nodes. This provides a new method for the structure optimization of neural network.On the other hand,"model in subsections, integration of the whole"modeling method based on neural network is promoted. To achieve the method, the modeled device's input and output data are acquired through experiments at first, the dividing points are selected likewise. Then the neural networks are trained in the MATLAB network using the data for each section, respectively. When the error goal is achieved, the neural network is completed, which can approach to the characteristics of different sections. The weights and biases are recorded after the training, and then the obtained network structure is described using the description language of PSpice to get the device model. Finally, the two models are integrated into one model to achieve the complete device model. In this way, model for the nonlinear electronic device, which has the complex characteristic curve, is achieved. The method is significant to modeling device and circuitry simulation system.