Imputation of Electrical Load Profile Data as Derived from Smart Meters

With the advent of smart meter technologies utility providers have acquired a wealth of knowledge regarding their energy distribution network. This allows for the identification of stress points in the network, calculation of technical losses and exploration of methods for their reduction. Furthermore, techniques may be adopted to also estimate non-technical losses. However, it is inevitable that a certain amount of data fails to be collected by the smart meter network, which must be imputed before any further analytical activity may be performed. This work explores techniques for imputing such missing data in smart meter load profiles and implements a k-nearest neighbors' approach. In this method, the imputed part is determined by analyzing the past consumption of the consumer, seeking patterns that best resemble the portion around the missing data. The algorithm developed was tested on a sample of 335 consumer load profiles resulting in an average normalized RMSE of 7.47%. That is, for a consumer with an actual maximum consumption of 10kWh, the RMSE of the imputed value is on average 747Wh.