摘要
针对智能电表故障具有的突发性、复杂性以及多面性等特点,提出一种基于时空卷积神经网络(ST-CNN)的故障预测方法。该方法首先采用滑动窗口将时间信息融入特征变量中,构建具有时空特性的输入矩阵,然后与CNN相结合,建立智能电表故障预测模型,并采用Adam算法对模型参数进行优化。最后应用现场的实际数据对基于ST-CNN的智能电表故障预测模型进行仿真,结果表明该方法预测精度高,泛化能力强。
The faults of smart meters are sudden,complex and multifaceted.A fault prediction method based on spatio-temporal convolutional neural network(ST-CNN)is proposed.Firstly,the sliding window is used to integrate the time information into the characteristic variables,and the input matrix with space-time characteristics is constructed.Then,combined with CNN,the fault prediction model of smart meter is established,and the model parameters are optimized by adaptive momentum estimation(Adam)algorithm.Finally,the actual field data are used to simulate the fault prediction model of smart meter based on ST-CNN.The results show that this method has high prediction accuracy and strong generalization ability.
作者
高文俊
薛斌斌
庞振江
Gao Wenjun;Xue Binbin;Pang Zhenjiang(Beijing Zhixin Microelectronics Technology Co.,Ltd.,Beijing 102299,China)
出处
《电子技术应用》
2022年第3期59-63,共5页
Application of Electronic Technique
作者简介
高文俊(1981-),男,硕士,高级工程师,主要研究方向:用电技术和电力自动化;薛斌斌(1987-),男,硕士,工程师,主要研究方向:电力采集;庞振江(1978-),男,硕士,高级工程师,主要研究方向:电力自动化。
