Hourly Energy Consumption Forecasting for Office Buildings Based on Support Vector Machine

Abstract

Aimed at the nonlinearity and uncertainty of building energy consumption, a forecasting approach based on the support vector machine is proposed in this paper for the prediction of hourly energy consumption of an office building. The univariate model test is used to determine the input parameters. Superior model hyper-parameters are found by grid search optimization. The confidence interval of the model fitting error is applied to describe the uncertainty of building energy consumption. A case study is conducted using the data collected from an actual office building to verify the proposed approach. The results show that the overall mean absolute percentage error (MAPE) of the model after grid search optimization is reduced by 31.3%, and a higher model precision is achieved. After combining the prediction with the confidence interval, MAPE is found to be lower than 1.5% in different seasons and the building operation fluctuations are embodied. This approach can be used in the diagnosis and optimization of building operation.

Keywords

• energy consumption
• forecasting approach
• support vector machine
• grid search
• confidence interval