Building Load Forecasting Based on Sensor Network and Gaussian Process Regression

The accurate prediction of building power consumption not only plays an important role in the economy and security of distribution network operation, but also possesses important reference significance for the formulation of building energy-saving programs. Because the building load is affected by many factors, it is hard to greatly improve the accuracy of building load forecasting. To improve the forecasting accuracy of building load, based on sensor network and Gaussian process regression a building load forecasting method was proposed. Firstly, the indoor building occupancy was detected by the personnel presence detection sensor based on ultra-wideband (UWB) radar network, and the building occupancy was taken as one of the characteristics of the load prediction model. Secondly, a Gaussian process regression model was constructed, and this model was utilized to fit the nonlinear function of the load and related influencing factors, and based on the sampling approximate inference algorithm the maximum posteriori estimation of the model's hyperparameters was deduced to improve the accuracy of short-term load forecasting. Finally, by means of comparing prediction effects of Gaussian process regression model with different covariance functions, the optimal covariance function was identified to further improve prediction accuracy. The result of analyzing the computing example show that the average absolute percentage error obtained by the proposed method is reduced by 9.68% than that obtained by traditional Gaussian process regression, in which the building occupancy is not taken into account, thus both efficiency and accuracy of the proposed method are verified.