Short-Term Power Load Forecasting with Deep Belief Network and Copula Models

The complexity and uncertainty in scheduling and operation of the power system are prominently increasing with the penetration of smart grid. An essential task for the effective operation of power systems is the power load forecasting. In this paper, a tandem data-driven method is studied in this research based on deep learning. A deep belief network (DBN) embedded with parametric Copula models is proposed to forecast the hourly load of a power grid. Data collected over a whole year from an urbanized area in Texas, United States is utilized. Forecasting hourly power load in four different seasons in a selected year is examined. Two forecasting scenarios, day-ahead and week-ahead forecasting are conducted using the proposed methods and compared with classical neural networks (NN), support vector regression machine (SVR), extreme learning machine (ELM), and classical deep belief networks (DBN). The accuracy of the forecasted power load is assessed by mean absolute percentage error (MAPE) and root mean square error (RMSE). Computational results confirm the effectiveness of the proposed semi-parametric data-driven method.