基于改进GRNN的电网雷击跳闸预测

Predicting Lightning Outages of Transmission Lines Using Modified Generalized Regression Neural Network

摘要: 电力系统的安全可靠稳定有着重大意义，而雷击输电线路导致跳闸是一种常见的危害形式，因此有效预防预测雷击跳闸事故有着重要的意义。本文针对广义回归神经网络的超参数优化选择问题，提出基于和声搜索算法的优化超参数方法，将改进后广义回归神经网络用于建立雷击跳闸预测模型，并采用故障检测率、误报率、总预测精度以及平均绝对误差等性能指标评价该预测模型的预测性能。实验结果表明本文建立的模型能够准确预测雷击跳闸，预测模型性能优异。

Abstract: The security, reliability and stability of power system are of great significance. Tripping caused by lightning strikes on transmission lines is a common form of hazard. Therefore, it is of great signifi-cance to effectively prevent and predict lightning strikes. Aiming at the problem of optimal selec-tion of hyperparameters in generalized regression neural networks (GRNN), this paper proposes an optimal hyperparametric method based on harmony search (HS) algorithm. Then, the improved method is applied to establish the prediction model of lightning strikes. Fault detection rate (FDR), false alarm rate (FAR), total prediction accuracy (PA) and mean absolute error (MAE) are adopted to evaluate the prediction performance. The test results indicate that the prediction model of light-ning strikes based on improved GRNN can accurately predict lightning outages with an outstanding performance.