Bi-GRUI and Transformer Variant Based Missing Value Imputation Method for Electricity Consumption Data

The accuracy of electricity consumption data collection has a direct impact on the billing of users. In view of the inevitable existence of some missing data problems in the collection of electricity consumption data of ten million users, especially with the development of time-sharing tariff business, the inconsistency of tariff periods of different user groups and the frequent changes in the collection time period have further reduced the success rate and the completeness rate of electricity consumption data collection. For this reason, this paper proposes a Bi-GRUI and Transformer variant of missing value imputation method for electricity consumption data. The model as a whole utilizes a two-stage imputation method. In the first stage, an AutoEncoder (AE) structure based on Bi-Gated Recurrent Unit Imputation Unit (Bi-GRUI) is used to perform missing value completion of the data after masking operation. Considering the influence of time-sharing tariff influencing factor as well as the time decay factor, which is introduced into the global attention in the decoder, in order to fully excavate the time-sharing variability and the temporal correlation of the consumer electricity consumption data. The second stage further extracts relevant information from different dimensions, such as time and features, by putting the complementary values into the Transformer variant to obtain the final missing value complementary results. The experimental results show that the model proposed in this paper can effectively fit the missing values of user load, which can be used to support the construction of a new generation of collection system.