A novel imputation approach for power load time series data based on tsDatawig.

Accurate power load forecasting possesses important decision-making value for optimizing unit scheduling and grid operation, in which real-time load monitoring data collected by sensor networks is the core foundation for constructing forecasting models. However, due to physical factors such as sensor node failure, communication interruption, signal interference, etc., the power load data may be missing. Therefore, in this paper, we analyze historical power load data using the data imputation method and propose a time-coding method based on tsDataWig to solve the problem. The tabular data are preprocessed and encoded using a continuous time strategy. Then, the dataset is masked using three different data missing mechanisms. A power load data imputation framework is constructed by filling in the missing data through the tsDataWig method. The experimental results show that the data imputation method proposed in this paper has significant advantages and its prediction errors are all lower than those of other methods, which confirms the effectiveness of the method in predicting the missing values of the power load data. The problem of data missing loophole caused by incomplete sensor network data can be effectively solved.