基于BP神经网络的交流接触器剩余电寿命预测

交流接触器广泛应用于各种用电控制系统,对其剩余电寿命进行准确预测可以提高用电控制系统运行的可靠性。建立了交流接触器剩余电寿命预测的神经网络模型,并提出网络模型结构参数的确定方法。采用平均影响值(MIV)筛选方法对预测模型输入参量进行筛选,确定了累积燃弧能量和吸合时间为预测模型的主要输入参量,能够反映影响交流接触器电寿命的关键因素。分析了不同神经网络模型下交流接触器电寿命的预测误差,其中自适应遗传算法优化BP神经网络(AGA-BP)模型的预测精度最高。分析了输入参量对神经网络预测结果的影响,对比了输入参量无筛选、因子分析法、MIV筛选下预测的误差,结果表明采用MIV方法筛选出累积燃弧能量和吸合时间进行交流接触器电寿命预测的效果最好。将不同试品的试验数据分别作为训练样本和验证样本进行预测,其最大预测误差在11%以下,因此预测模型满足工程需要。

Residual Electrical Life Prediction for AC Contactor Based on BP Neural Network

AC contactor is widely used in electrical systems,and the prediction of its residual electrical life is essential for improving the reliability of these systems.A neural network model is established to predict the residual life of AC contactor,and a method is proposed especially to determine the structure parameters of the model.The accumulation of arc energy and the making time are selected as inputs of the model by the method of Mean Impact Value (MIV),which are two main factors affecting the electrical life of the contactor.Among the neural network models analyzed in this paper,BP neural network model optimized by adaptive genetic algorithm (AGA-BP) has the highest prediction accuracy.The prediction errors of the models with unprocessed inputs,inputs processed by factor analysis and inputs processed by MIV are analyzed.The results show that MIV is suitable for predicting the electrical life of AC contactor.The test data of each sample is used as training sample and testing sample,and the maximum prediction error is less than 11％,so the model is acceptable in actual use.