人工神经网络(ANN)具有的自适应、自学习、非线性重构等特点,使之成为解决电力系统负荷建模的有效途径。该文利用模糊理论对典型BP神经网络(TBP)的学习速度因子和权值惯性因子进行修正,采用自构形学习算法对网络拓扑2个方面进行改进,提出自适应神经网络(ABP)。结合现场试验和仿真数据,对TBP和ABP在负荷建模的速度和精度2方面进行了比较。同时,就负荷建模问题对自适应神经网络模型阶次和隐层神经元个数等因素进行了探讨。
Artificial neural network (ANN) is an effective way in electric power system load modeling. The paper presents an adaptive back-propagation network (ABP), which modifies learning rate and momentum by using fuzzy logical theory. The structure of network is optimized by self-structure learning algorithm. And the model is applied to field tests data and the identification effect is also verified. Furthermore, some factors, such as network orders and hidden neuron numbers, are discussed in load modeling.