Forecasting model of building energy consumption based on parallel Kriging sampling algorithm

Abstract

Parallel Kriging sampling approach (PEIGF) with minimum computing resourcesis proposed to promote the speed ofbuilding energy consumption. In each iteration of PEIGF, the initial sample point is obtained by maximizing the standard expected improvement global fit (EIGF) criterion andthe parallel EIGF function is constructed to obtain multiple new sample points by maximizing the PEIGF criterion. Performance of PEIGF is evaluated by comparing the obtained results with four other existing sequence approaches and four artificial intelligence (AI) based models. Theresults reveal that PEIGF stands out for most of the test cases with goodpredictive accuracy and computing efficiency. PEIGF runs 4.7–11.6 times, 5.1–12.4 times, 5.0–14.8 times, and 19.7–158.6 times faster than thoseof combined expectation (CE), EIGF, expected improvement for global fit based on gradient (EIGFG) and maximum mean square error (MMSE) sampling.Itobtains the least root mean squared errors for energy consumption prediction, and reduces the total number of samples by 14.9% (heating load) and 14.1% (cooling load), as well as 16.7%compared with other four AI-based models.It isprovento bea promising technique toplan low-cost energy managementandfacilitate early designs or renovation of energy conserving buildings.