Based On Neural Network Model Of Switched Reluctance Motor Analysis And Research,

Switched reluctance motor (SRM) system is a new kind of variable-speed drives,which has been developed in 1980s. The system is composed of double-salientreluctance motor, power converter, rotor position sensor and controller. The motor andconverter have more simple structure, higher reliability, and more flexiblecontrollability of the system and high efficiency. It could be looked as an importantrespect in the research field of motor and variable-speed drives. But, the magneticcharacteristic of SRM is nonlinear, which makes it very difficult to derive an analyticmathematical model of SRM.The paper generalizes several current frequently-used SRM models, which includeideal linear model, half-linear model and nonlinear model, based on basic principle andmathematical model of Switched Reluctance Motor, and analyzes these models'application area.Magnetic properties model of SRM is a nonlinear map problem, while artificalneural networks has a very strong nonlinear mapping ability. So it is significative toexplore a SRM modeling method based on neural networks. Nowadays correlativeresearches have been carried on and people have achieved good experiment effect.These researches are primarily based on mature neural network models, such as BPneural network, RBF neural network and so on. But the networks themselves havesome shortcoming, so it is not enough to set up models. People begin to use modelingmethod of hybrid neural networks to offset the disadvantage of traditional neuralnetworks aimed at this situation, such as fuzzy-neural networks. This kind of modelingmethods has achieved very good effect in practical application.The paper explores a modeling method for SRM based on hybrid neural network.Because traditional BP neural network has slower convergence rate and poorer globaloptimization ability, while Genetic Algorithms have strong global searching abilitywhich can offset the disadvantage of BP neural network, the two methods can becombined together to optimize the structure and parameter of BP neural network byGenetic Algorithms. The method has wide application and mature theory.Optimization method of Genetic Algorithms has been improved aiming atpractical SRM modeling problems, which mostly adjust code rule and genetic operation.After simulation experiment, we find this improved algorithm shows good robustnessand fitting while modeling. Efficiency of algorithm has improved compared withformer methods. At the same times it is found that there are some disadvantages inalgorithm in the course of emulation. Because the training of neural network needplenty of samples to get stable network, we need to provide lots of representativemeasured experimental data, which is the disadvantage of neural networks modelingcompared with other nonlinear modeling methods.