To build electromagnetic pulse response models for complex electronic circuits without knowing their internal structure and parameters

a dynamic model based on NARX neural network is established

and the sine wave sweep frequency signal and the circuit response are taken as training data. The theoretical basis and design steps of the modelling with NARX neural network are proposed

and it is proved that the response model of lumped parameter circuit can be replaced by the present dynamic model. The filter circuit and radio-frequency amplifier circuit are designed and simulated by ADS software. The satisfactory simulation results indicate that the proposed method can be applied to predicting the electromagnetic pulse response of lumped parameter circuit. In addition

the defects of NARX neural network are analyzed

and the corresponding improvement method is suggested by using genetic algorithm to optimize the network parameters and by using support vector machine（SVM）or extreme learning machine（ELM）to take the place of the feedforward neural network section in NARX neural network.