CAJ | 학술논문

针对传统BP神经网络用于谐波检测时存在收敛速度慢、易陷入局部最小值的缺点，提出用混合蛙跳算法代替BP神经网络中梯度搜索算法的混合蛙跳算法神经网络，并将其用于电力系统谐波幅值与相位测量。根据电力系统所含谐波特点，构建谐波检测的神经网络模型，阐述混合蛙跳算法神经网络的基本原理。以三次谐波为例，给出神经网络训练方法以及训练样本如何构建。仿真结果验证所提方法的可行性，其收敛速度、检测精度均优于BP神经网络。最后用训练好的神经网络检测未训练的样本，实验结果验证该网络具有良好的泛化能力。
침대전통BP신경망락용우해파검측시존재수렴속도만、역함입국부최소치적결점，제출용혼합와도산법대체BP신경망락중제도수색산법적혼합와도산법신경망락，병장기용우전력계통해파폭치여상위측량。근거전력계통소함해파특점，구건해파검측적신경망락모형，천술혼합와도산법신경망락적기본원리。이삼차해파위례，급출신경망락훈련방법이급훈련양본여하구건。방진결과험증소제방법적가행성，기수렴속도、검측정도균우우BP신경망락。최후용훈련호적신경망락검측미훈련적양본，실험결과험증해망락구유량호적범화능력。
According to the harmonic measuring for traditional BP neural network, compares the problems of slow convergence speed, easily falling into local minimum value. Proposes a Shuffled Frog-leaping algorithm neural network using Shuffled Frog-leaping Algorithm, instead of a Gradient Search Algorithm in BP neural network method for Harmonic amplitude and phase measurements of power of system. The neural network model is developed according to the requirements of measuring harmonic. Expounds the basic principle of Shuffled Frog-leaping Algorithm neural network. Gives the training method of SFLA neural network and how to construct the training sample in the three har-monic as an example. The simulation results verify the feasibility of the proposed method. SFLA neural network convergence speed and detection accuracy is better than the BP neural network. Uses the neural network detection trained without training samples, the result proves that the neural network has good generalization ability.