Office Building Energy Efficiency Benchmarking Method Based On Quantile Regression

Building energy consumption is one of the main components of the global energy consumption structure.Improving building energy efficiency has engineering guiding significance for reducing overall building energy consumption and maintenance costs,as well as strengthening building energy consumption management.Due to the lack of a quantitative assessment standard for building energy efficiency benchmarking,it is difficult to estimate the investment risk and return period for building energy efficiency improvement projects,and it’s also a challenge for governments to draft relevant building energy consumption policies.In this regard,this thesis proposes a building energy-saving potential benchmarking method based on probabilistic forecasting modeling strategies.The specific research work is as follows:The "twice-boxplot analysis" was proposed to separate and eliminate the abnormal energy consumption extreme values in the data set.Based on the pre-processed data,combined with the high importance of extreme energy using situations in the process of building energy consumption level evaluation,quantile regression is used as the underlying algorithm to nest the time series prediction algorithm modeling strategy.Instead of the traditional point prediction model,three probabilistic prediction models were built to output the prediction confidence interval for improving the reliability of the forecast results.And three different indicators were used to evaluate the performance of the models on HVAC energy consumption,lighting,and electrical energy consumption,and total building energy consumption forecasting respectively.The results show that among the three probabilistic prediction models,the long-short-term memory quantile regression model(LSTM-QR)performs well in all three prediction scenarios and has the best comprehensive performance.Based on the prediction results of the forecast model,the cumulative probability distribution curve of the total energy consumption for the whole year is obtained.After linearizing the cumulative probability distribution curve,a quantitative evaluation method for building energy-saving potential is proposed and tested on real office buildings.Then,the composite effect from the energy efficiency change of the subsystems was deeply explored.The probability of triggering additional benefits on the building from the composite effect was calculated.Based on the evaluation rules,the energy-saving potential of 8 sample buildings was compared and analyzed.The results show that: Building 1 gets the highest score while Building 11 gets the lowest one.And the No.1,No.7,and No.18 buildings may obtain additional benefits from the compound effect caused by the energy efficiency changes of the subsystems,and the probabilities are 53.74%,54.55%,and 42.35% respectively.In this thesis,by introducing probabilistic forecasting modeling strategies,combining quantile regression with commonly used time series forecasting algorithms,and replacing point forecasting models with interval forecasting models,the accuracy and credibility of energy consumption forecast results were improved.On this basis,based on the cumulative probability distribution,a method for benchmarking the office building energy efficiency potential is proposed,which provides an effective reference and guarantee for the risk assessment on building energy efficiency improvement plans making.