Abstract: With the increase of the complexity of microwave device structure and the requirement of product performance,microwave device modeling should not only be able to describe the ideal electromagnetic characteristics, but also be able to quickly and accurately reflect the influence of multiple physical parameters on device performance. Although neural networks have been introduced into the field of microwave devices, there are few studies on their application in the multiphysical modeling of devices. In this paper, a multi-physical parameter modeling method based on artificial neural network is proposed to represent the nonlinear relationship between input and output variables. An efficient neural network multi-physical parameter model is proposed, and a new training algorithm is introduced for the model. The proposed model can quickly and accurately predict the multi - physical response of microwave devices , such as the S - parameter characteristic curve of filters and the output characteristic curve of ion-sensitive field-effect transistor. Compared with the finite element method, this method can save about 98% of the calculation cost and 99% of the calculation time, which provides a feasible method for fast and efficient behavior-level modeling of microwave devices.