Prediction of Government-Owned Building Energy Consumption Based on an Rrelieff and Support Vector Machine Model

Abstract

Accurate prediction of the energy consumption of government-owned buildings in the design phase is vital for government agencies, as it enables formulation of the early phases of development of such buildings with a view to reducing their environmental impact. The aim of this study was to identify the variables that are associated with energy consumption in government-owned buildings and to propose a predictive model based on those variables. The proposed approach selects relevant variables using the RReliefF variable selection algorithm. The support vector machine (SVM) method is used to develop a model of energy consumption based on the identified variables. The proposed approach was analyzed and validated on data for 175 government-owned buildings derived from the 2003 Commercial Building Energy Consumption Survey (CBECS) database. The experimental results revealed that the proposed model is able to predict the energy consumption of government-owned buildings in the design phase with a reasonable level of accuracy. The proposed model could be beneficial in guiding government agencies in developing early strategies and proactively reducing the environmental impact of a building, thereby achieving a high degree of sustainability of buildings constructed for government agencies.

Towards developing a systematic knowledge trend for building energy consumption prediction

“…Capitalises on the merits of individual methods Gradient boosting machines, Random forest [45], [75], [85] [86], [116], [187], [195] Clustering algorithms: k-Means [41], [50], [69], [71], [207], k-Medians, Expectation maximization [50], [73], Hierarchical clustering [41], [50], [207] Useful for making sense of data  Results are sometimes difficult to interpret  Very limited when dealing with unfamiliar datasets Dimensionality reduction algorithms: Principal component analysis [41], [64], [73], [79], [83], [131], [135], [165], Principal component regression [41], Partial least squares regression [21] Good for handling large datasets without necessarily making assumptions on data  Not effective when dealing with non-linear data  It is sometimes difficult to understand the meaning of the results…”

“… Superior in solving nonlinear problems with high-dimensional datasets  Can handle large and incomplete datasets  Self-adapting, selforganizing and real-time learning network  Easy to construct the network models  Requires a large amount of data  Extremely computationally expensive to train  The internal working is unknown  The meta parameter and network topology selection is hard Table 3. Summary of articles that explored hybrid methods, their associated sub-methods and functions Reference Algorithm Function [67] Physical Simulates the building energy performance Genetic Algorithm Identifies the buildings' internal mass model parameters [68] GM (1,1) Predicts the buildings' energy consumption Radial basis neural network Revises the residual errors of the grey model [69] RReliefF Accounts for interdependencies between variables so as to select the optimal variable subset SVM Predicts the buildings' energy consumption [70] Improved real coded genetic algorithm…”

Transformer-Based Model for Electrical Load Forecasting

“…In order to assess the significance of the improvements the proposed transformer model achieves over the state-of-the-art S2S [36], the two-sided Wilcoxon signed-rank test [49] was performed with respect to MAPE and RMSE. This method has been used in the context of energy predictions and electrical load forecasting [50][51][52] and allows us to establish whether the reported model errors are significantly different. For each of the experiments, the errors obtained by the two models, S2S and our transformer model, were assessed for each of the 20 streams.…”

Energy planning and forecasting approaches for supporting physical improvement strategies in the building sector: A review

“…However, recent work done by Son et al (2015) used SVM regression models to predict the electricity consumption of government owened buildings in the early design stages. This was achieved by first retrieving the relevant parameters through applying a variable selection algorithm called RreliefF.…”