Artificial neural networks are information processing systems having achieved great success in many areas such as speech recognition, image processing and more. In this paper, we describe neural network approaches to learn the complex behavior of high-frequency electronic circuits through learning. The training data which embed the information of high-frequency electronic behavior and their relationships with structural parameters are obtained by electromagnetic simulation. We address the issue of data generation expenses for training neural networks by incorporating prior knowledge of electronic behavior in the form of semi-analytical equations and equivalent circuits. The knowledge based neural network model can be trained with less data while retaining neural network accuracy, and can exhibit good tendency of electronic behavior even used outside the training region.