Failure Rate Prediction Method Of Smart Meter Based On Improved Stacked Denoising Autoencoder

As the measuring terminal of smart grid,smart meter is an indispensable bridge to ensure the common interests of enterprises and users.It plays an important role in the development of smart grid and future energy management.The reliability of smart meter directly affects the safe and stable operation of smart grid.Nowadays,people put forward higher requirements for the research on the failure rate prediction of smart meter due to the various kinds and large number of smart meters under complex working environment and long continuous operating time.To solve the above problems,this paper proposes a failure rate prediction method of smart meter based on improved stacked denoising autoencoder.The main contribution is shown as follows:Firstly,this paper carried out the failure rate test of key components of smart meter,including the design and implementation of test process and the performance test of samples.The basic failure rate of key components is calculated according to the test results and the basic failure rate model of components.Then,according to the reliability prediction handbook of electronic equipment,this paper provides the working failure rate model of components,establishes the module failure rate model of smart meter considering IEC62059-41 standard,and presents the whole meter failure rate model of smart meter based on Weibull distribution function.Secondly,a failure rate prediction method of smart meter based on improved stacked denoising autoencoder is proposed in this paper.In the unsupervised training stage of stacked denoising autoencoder,Adadelta algorithm is used to adaptively adjust the learning rate.Nadam algorithm is used in the supervised fine-tuning stage to add Nesterov momentum,and Dropout is used for regularization,which improves the failure rate prediction accuracy of smart meter.Finally,based on the data from smart meter manufacturers and measurement of integrated production dispatching system,the proposed method is verified by experiment.Through a practical example,the specific calculation process of module failure rate and whole meter failure rate data required for prediction is demonstrated,and the accuracy of failure rate calculation model is verified;The comparison between different algorithms on perspective of failure rate prediction results of smart meters from different manufacturers and models are carried out and verifies that the proposed method has lower prediction error and can realize the accurate prediction of failure rate of smart meter.