A neural network approach for creating a NTC thermistor model library for PSPICE

most sensors can not be modeled easily, which leads to the problem that a circuit with sensors can not be simulated in PSPICE. A method based on the neural network for modeling NTC thermistors and creating a NTC thermistor model library for PSPICE is presented to solve the problem. Firstly, a multi-layer feedforward neural network is used to approximate the characteristics of a NTC thermistor. Secondly, the achieved structure of the neural network is described in the PSPICE language to form a subcircuit. Thirdly, the structure is used to model the same series of NTC thermistors by changing weights and biases of the neural network. Finally, the subcircuits for the series of NTC thermistors can be packed into a file to create a model library. During PSPICE simulation, the variations of a non-electric quantity imposed on a sensor are replaced with those of an electric quantity. The availability of this method is verified in circuit simulation. This method can be extended to model other sensors.