一种新型的模糊神经网络电阻检测器

A New Type of Fuzzy Neural Network Resistance Observer

基于混合型Ｐｉ—Ｓｉｇｍ ａ网络，设计了一种新型模糊神经网络电阻检测器，用于检测感应电动机运行时的定子电阻。给出了模糊规则前件语言变量隶属函数和后件结果函数有关参数的学习方法。选择电机绕组端部温度及其时间变化率作为检测器输入，通过工业实验测取了２０００ 组数据分别用于网络的学习和推理结果的检验。检测器用于直接转矩控制系统时，依据检测出的电机定子电阻瞬时值，对定子电压进行补偿，进而改善电机低速运行时的速度稳定性。对基于不同补偿方法测得的转速误差率分析表明，本文介绍的补偿方法效果优于其它方法，这证明模糊神经网络电阻检测器具有更高的检测精度。

A new type of fuzzy neural network resistance observer is developed, on basis of the hybrid Pi—Sigma networks, to measure the real time stator resistance of an induction motor. A self learning algorithm is proposed for optimizing the parameters of membership functions belonging to the language variables of the rule preconditions as well as the parameters of the output functions. The two variables, T and ΔT, measured at terminal of winding were selected as the inputs of the fuzzy observer. 2000 groups of the data obtained through the industial experiments were used for the learning of networks as well as the comparison with the results observed respectively. Applied in direct torque control system, the observer could measure the instantanous stator resistances precisely. The stator voltage is compensated by the observed resistances in order to improve the low speed stability of the induction motor. With the analysis of the speed error accompanied with the differient compensating ways, it has been proved that the fuzzy neural network resistance observer enjoys a high accuracy.