Sensitivity Analysis Of Building Energy Consumption Based On Sequential Sampling

In response to global climate change,many countries have proposed carbon peaking and carbon neutrality goals.The Chinese government has proposed peaking carbon dioxide emissions by 2030 and striving to achieve carbon neutrality by 2060.In 2019,my country’s construction sector carbon emissions accounted for about 38% of the total carbon emissions,and the control of carbon emissions in the construction sector is the key to emission reduction.It is of great significance to reduce the energy consumption of buildings by making a reasonable energy-saving renovation plan by analyzing the energy consumption sensitivity of new buildings and existing buildings.Sensitivity analysis is to study the influence of input parameter uncertainty on output energy consumption.Through sensitivity analysis,a reliable solution can be provided for energy-saving renovation of new buildings and existing buildings.However,the current sensitivity analysis in the field of building energy consumption usually adopts the method of fixed sampling number,and most of the studies do not make convergence judgments on the sensitivity results,so the reliability and stability of the sensitivity results cannot be guaranteed.Therefore,this paper proposes a research on building energy consumption sensitivity based on sequential sampling.The basic principle of sequential sampling method is to dynamically adjust the sampling times of building energy consumption simulation,and to evaluate the convergence of sensitivity analysis results in time after each sampling.Compared with the sensitivity analysis method of fixed sampling,it significantly reduces the running times of the building energy consumption model,saves the calculation time of dynamic simulation of building energy consumption,and provides accurate and stable sensitivity results.This paper adopts three global sensitivity analysis methods: Morris screening method,standard regression coefficient method and Sobol sensitivity method.Based on sequential sampling,a sensitivity analysis of building energy consumption in an office building in Tianjin is carried out.The sensitivity results include variable importance ranking and sensitivity indicator.The research results show that the sensitivity results tend to be stable with the increase of sampling times,but the sampling times required by different sensitivity methods are significantly different,and it is found that the sampling times required for ranking convergence is significantly less than that for index convergence.At the same time,it is found that there are certain differences in the sensitivity results obtained by the three sensitivity analysis methods.It is recommended to use more than two sensitivity analysis methods to test the robustness of the results.If there is non-monotonic or non-linear relationship between input parameters and building energy consumption,variance-based sensitivity analysis method(such as Sobol sensitivity method)is recommended.When using the Morris screening method,only a few to dozens of trajectories are needed to satisfy the ranking convergence,while the index convergence requires dozens or even hundreds of building energy consumption model calculations.When applying the standard regression coefficient method,four different sampling methods of random,uniform design,optimized latin hypercube and Sobol sequence are used,only dozens of samplings are required to satisfy the ranking convergence,and the index convergence requires about 1200-1400 sampling times.Moreover,uniform design sampling and Sobol sequence sampling require fewer sampling times to satisfy the index convergence.When the Sobol sensitivity method is applied,the sensitivity analysis sampling times need to be several thousand or tens of thousands to satisfy the ranking convergence,while the index convergence requires hundreds of thousands or even millions of times,and the Sobol sequence sampling method can meet the convergence requirements faster.The sensitivity analysis method of building energy consumption based on sequential sampling is presented in this paper,which can reduce the computational cost significantly while ensuring the convergence of sensitivity analysis results,not only can it be used in building energy consumption machine learning modeling to reduce unnecessary variables and improve the stability of building energy consumption model,but also can be used to quickly develop reliable and efficient building energy saving scheme.