Towards missing electric power data imputation for energy management systems

摘要: Abstract Demand for electricity is gradually increasing in many countries. Efforts related studies have been made the application of data mining techniques over electric power development more effective energy management systems. However, one major challenge how to compensate parts collected dataset, such as consumption, voltage, or current that may be missing a specific period time. In literature, several methods employed imputation data, especially single feature value imputation. performance different types methods, i.e. statistical and machine learning multiple features has not fully explored. Moreover, variations their during summer/non-summer seasons peak/off-peak/semi-peak times investigated. this paper, five well-known processing two autoregressive integrated moving average (ARIMA) linear interpolation (LI) models, three k-nearest neighbor (K-NN), multilayer perceptron (MLP), support vector regression (SVR) compared. The experimental results, based on two-year Taiwan, show generally perform better than ones, with K-NN SVR performing best. particular, all produced higher error rates summer season non-summer seasons. (especially K-NN) are choices peak times, whereas LI) off-peak semi-peak times.