基于正交小波神经网络的广域紧急直流功率支援在线预测

摘要

As the BP neural network is easy to fall into local minimum point and has slow convergence speed, and the RBF neural network has no sole approximation function,the OWNN(Orthogonal Wavelet Neural Network),which takes the Harr orthogonal scaling function as its activation function,is proposed and its weight training algorithm is presented. The OWNN is applied to EDCPS(Emergency DC Power Support) online prediction and control of Yun-Guang UHVDC and Gui-Guang H DC transmission. Simulation results show that,it ensures the sole and fast approximation with simple training algorithm and outputs the EDCPS control reference with higher reliability and precision by mapping the aggregated characteristic input data.%鉴于BP神经网络较易陷入局部极小点且收敛速度慢、RBF神经网络因其激励函数是冗余的非正交基故其逼近函数的表达式并不唯一等缺点,构造以Harr正交小波尺度函数为激励函数的神经网络并提出其相应的权值训练新方法,将该正交小波神经网络应用于实现对云广特高压直流和贵广Ⅱ直流的在线紧急直流功率支援在线协调预测控制.仿真结果表明:正交小波神经网络采用正交尺度函数作为激励函数,能保证网络逼近的唯一性,且训练算法简单、收敛迅速;正交小波神经网络能映射聚合成的特征输入数据,准确给出紧急直流功率支援控制量,具有较高的可靠性和准确性.