Predicting the Energy Meter Demand by a Window-based Gradient Boosting Regression Tree Model

The accurate prediction of energy meter demand is crucial for the formulation of the energy meter purchasing management plan of the metering centers, leading to effective reduction in distribution and storage costs. In this study, we use a window-based Gradient Boosting Regression Tree(GBRT(W-b)) prediction model for the forecast of energy meter demand. The model is used to train and fit historical data about the numbers of monthly newly installed energy meters in Fujian Province, China. We compare the prediction results obtained using the GBRT(W-b) model with those from Autoregressive Integrated Moving Average(ARIMA). The experimental results demonstrate that GBRT(W-b) outperforms the ARIMA model in forecasting the total amount of newly installed energy meters for the next three months, next six months, and next nine months. This also indicates that the proposed prediction method based on GBRT(W-b) achieves high prediction accuracy and can assist power grid companies in improving their purchasing management of energy meters.