开关磁阻电机的Pi-sigma模糊神经网络建模

摘要

The high saturation of magnetic circuit and the doubly salient structure of switched reluctance motor (SRM) make the flux linkage a nonlinear function of rotor position and phase current. In this paper, the pi-sigma fuzzy neural network, which has the merit of T-S type fuzzy logic and neural network, is adopted to develop the nonlinear model of SRM and an adaptive learning rate training algorithm with momentum is applied. Relatively high precision model of SRM is implemented. It has simple structure, less training epoch and fast online calculation. The sampled phase current and rotor position are non-equally spaced. Thus a reasonable distribution of measured data is reached. The precision and generalization ability of the model are improved. Meanwhile the number of measured data is reduced. Compared with the measured data and generalization validate data, the output data of the model are in good agreement with those data. This proves that the precision of the model developed in this paper is relatively high. The model has the merits of relatively strong generalization ability, simple structure and fast calculation.%开关磁阻电机的磁路高度饱和及双凸极结构导致了相绕组的磁链是转子位置和相电流的非线性函数.本文采用兼具Takagi-Sugeno(T-S)模糊逻辑和神经网络优点的Pi-sigma模糊神经网络来建立开关磁阻电机的非线性模型并采用了附加动量项的自适应学习速率训练算法.实现了开关磁阻电机的较高精度建模,减少了学习训练次数,简化了结构,使其可在线快速运算.本文通过对相电流与转子位置角的非均匀间隔采样和对论域的全面覆盖,来达到测量数据的合理分布,以提高建模精度和泛化能力并减少测试数据量.通过对模型输出数据与实测数据进行比较及对泛化样本数据的校验表明,本文所建立的模型具有精度较高、泛化能力较好、结构较简洁、运算速度较快等特点.