基于改进广义回归神经网络的雷达故障预测

Fault Prediction of Radar Based on Modified General Regression Neural Network

针对目前雷达故障预测存在的问题,提出了一种基于遗传算法(GA)的广义回归神经网络(GRNN)模型。该模型以前10个时刻的雷达状态为输入,以下一时刻状态及其变化速度为输出;利用遗传算法对网络平滑因子以及网络结构进行优化,以均方差(MSE)最小构造适应度函数。仿真结果表明,所提出的GRNN模型预测值与计算值的偏差系数2.62%,期望偏差率2.07%。

A new general regression neural network （GRNN） veloped to fault prediction of radar. The previous states of radar changing rate of state as output. The fitness function aiming at the model based on genetic algorithm （GA） was dewere used as network input, the next state and mean square error （MSE） was constructed when optimizing the smooth factor of GRNN. Simulation shows the coefficient of variation and expected error percentage between the predictive values and calculative values are 2.62% and 2.07%, respectively.