Uncertainty prediction of energy consumption in buildings under stochastic shading adjustment

• Published in: Energy (Oxford) 2022-09, Vol.254, p.124145, Article 124145
• Main Authors: Feng, Yayuan, Yao, Jian, Li, Zhonghao, Zheng, Rongyue
• Format: Article
• Language: English
• Subjects: Hyperparameter optimization; Machine learning; Shapley value method; Stochastic shading
• ISSN: 0360-5442
• DOI: 10.1016/j.energy.2022.124145

The prediction of building energy consumption is indispensable to reduce energy consumption, improve energy efficiency and achieve carbon neutrality. The stochastic adjustment of shading has an important impact on energy consumption due to the uncertainty in the use of window shades in common office buildings. This study is based on a stochastic shading building model established in the previous study and uses time, temperature, solar radiation, and shading coefficient as input variables for predicting shading related energy uncertainty. Firstly machine learning algorithm is used for modeling, then Shapley Value Method is applied to refine the model variables, and finally, the model is optimized by hyperparameter optimization. The resulting model can perform uncertainty prediction of building energy consumption under stochastic shading adjustment. The results indicate that the Gaussian process regression is suitable for the prediction, and the final model prediction accuracies of R2 are all above 0.9, which can be used in practical applications. This study is the first to address the uncertainty prediction of building energy consumption under stochastic shading adjustment using machine learning methods without the use of energy simulation tools. •Energy uncertainty fluctuation under stochastic shading adjustment was predicted.•Machine learning method was used for prediction instead of physical energy models.•Shapley value and hyperparameter optimization are used to reduce prediction errors.