摘要
提出了一种基于广义回归神经网络采用实测数据的谐波源建模方法。引入电压运行度和功率负荷度概念,通过广义回归神经网络将它们与各次谐波电流幅值之间的非线性映射关系建立谐波源模型。在该模型中,对网络平滑系数进行了优化设计,将最小检测误差对应的平滑系数用于网络训练;对谐波源在不同运行条件下的负荷度-电流特性进行了研究,根据电压运行度和功率负荷度估计各次谐波电流幅值。以某中频炉实测数据为例,结果表明该模型计算值与实测值的误差很小,具有人为确定参数少、训练时间短、精度高等优点,是一种有效的谐波源建模方法。
A generalized regression neural network （GRNN） is proposed for harmonic source modeling based on measurement. The concept of voltage operating-degree and power load-degree is introduced; and the nonlinear mapping between them and each order of harmonic current amplitude is established with GRNN. In the model, the network smooth coefficient is optimized, and the coefficient that is related to the minimum detection error is selected for network training. The load-degree-current characteristics in different operating conditions are studied, and the harmonic amplitude is estimated by the two degrees. This model can be used for harmonic emission assessment and condition monitoring of harmonic load, and the harmonic current can be calculated according to the relationship between load-degree and harmonic current. Take a middle-frequency furnace based on measurement as an example, the results show that error of the model is very small. The proposed method has the advantages of few artificial parameters, short training time and high precision, and is an effective technique for building up harmonic source model.
出处
《电工电能新技术》
CSCD
北大核心
2012年第3期64-67,72,共5页
Advanced Technology of Electrical Engineering and Energy
