Abstract: The existing watt-hour meter fault prediction models in the State Grid information system are relatively simple and insufficient, and there is no specific model for predicting the monthly fault number of watt-hour meters. Based on time series, an integrated time series prediction model is established for an accurate prediction of the monthly fault number of batch watt-hour meters. Firstly, the moving average sequence is calculated for the monthly fault number of watt-hour meters to remove small fluctuations. And then, the ARIMA model or exponential smoothing model is selected to predict the moving average sequence according to the long-term trend of the sequence. Finally, the reverse moving average is used to realize the accurate short-term prediction of the monthly fault number of the whole batch of watt-hour meters. By comparison with the BP neural network model, the practicability and accuracy of the proposed time series model is verified. On this basis, a monthly fault prediction model is established. The measurement asset management departments can use the proposed method to predict the number of faulted watt-hour meters, and prepare the stock according to the prediction results, consequently improving the rationality of resource allocation and work efficiency.

Key words: watt-hour meter, monthly fault number, time series, BP neural network, reasonable distribution of watt-hour meters