神经网络负荷模型的综合能力研究

Neural Network Load Modeling Synthesis Ability Research

针对电力电子设备综合负荷模型难以用机理模型描述的现状,构造了动态综合负荷的模糊神经网络模型.该模型具有模糊推理和神经网络的优点,能很好地逼近动态负荷的模型输出.通过对已知实测建模数据的训练,分析了模糊神经网络负荷模型的前件参数、结论参数的辨识策略,阐述了模糊隶属度和模糊规则的形成过程.对负荷构成相异的4组实测变电站负荷数据,用其中1组建模数据进行训练,得出模糊模型结构和参数,用该模型去拟合其他3组数据,对模糊神经网络负荷模型的综合能力进行验证.实例表明,该模糊神经网络负荷模型不仅具有很强的自描述能力和收敛性,而且具有良好的综合描述能力.

In order to overcome the trouble that mechanism power load models are difficult to describe composite load characteristics of power electronics equipment, this paper put forward a kind of fuzzy neural power load model based on ANFIS （adaptive-network-based fuzzy inference system）. Integrating the advantages of fuzzy inference and neural network, the model can accurately represent the output behavior of dynamic power load. Through training and optimizing the neural network with the measured field data, the authors obtained the before-condition parameters and conclusion-parameters of the model. Combining the application instance, the authors elaborated the procedure forming fuzzy subordination parameter and fuzzy rule. To verify the synthesizing ability of the model, the authors applied 4 data samples at power substation field to model dynamic power load. One of them was used to identify the model, and the others were used to test the model. The results show that the fuzzy neural network model has not only excellent selfdescription ability but also strong synthesizing ability.