Modeling of switched reluctance machine with few samples based on chaotic fuzzy neural network

A method to obtain the simulation model of the switched reluctance machine (SRM) is presented. First, the flux-linkage characteristics of a 1kW 3-phase 12/8-pole SRM are quickly measured without any rotor clamping device and position sensor, and few samples can be obtained. Then, to build the accurate simulation model of the machine with these few samples, chaos theory is applied to the training of fuzzy neural network with gradient descent method, and local optimum is effectively avoided by the chaotic characteristics of the weights and the parameters of the membership functions. Finally, the accuracy of the model is verified by the comparisons between the phase currents from simulations and experiments under different control methods and operating conditions.