作　　者： ; ; ; ;

机构地区： 北京航空航天大学

出　　处： 《电光与控制》 2013年第4期72-76,共5页

摘　　要： 为了解决现有电子式电能表故障检测方法精度偏低、训练速度慢的问题,提出一种BP-AdaBoost复合神经网络故障预测方法。首先,在单一BP神经网络的基础上,利用组合分类器算法AdaBoost对其进行改进,将多个单一BP神经网络作为弱分类器,多次迭代训练得到强分类器;随后,将该故障预测方法应用于电子式电能表的典型故障——整流桥故障的分类判别中;利用Simulink搭建电子式电能表仿真模型,选取故障注入点与观测点,仿真运行生成的故障数据作为BP-AdaBoost算法的处理对象。仿真结果表明,BP-AdaBoost故障预测方法与单一BP神经网络故障预测方法相比,能提高预测精度,显著减小误差,在实际应用中具有一定可行性。 Aiming to solve the shortcoming of the existing failure supervising method for electricity meter, such as low forecasting accuracy, the neural network prediction algorithm, named BP-AdaBoost, was proposed. Firstly, the combination classification algorithm called AdaBoost was used to improve the single BP neural network. Working as the weak classifiers, several BP neural networks constituted a strong classifier after training repeatedly. Then this prediction method was applied to the classification and discrimination of the typical failure of the electricity meter that is the rectifier bridge failure. The simulation model of the electricity meter was established by using Simulink. The fault injection points and observation points were chosen, and the failure data generated by the simulation was used as the processing object of the BP-AdaBoost algorithm. Finally, the computer simulations shows that compared with the single BP neural network fault prediction method, the BP-AdaBoost algorithm can improve prediction accuracy and significantly reduce the error, which is applicable to practice as well.

关 键 词： 电能表 故障检测方法 算法 神经网络

领　　域： [航空宇航科学与技术] [航空宇航科学技术]