Research On Building Energy Consumption Prediction And Energy Consumption Evaluation Based On Data-Driven

With the rapid development of social economy and the yearly growth of population,the demand and consumption of energy in the world are increasing dramatically,and it is imperative to save energy.According to statistics,building energy consumption accounts for the highest percentage of total energy consumption in China.How to reduce building energy consumption has become a key issue in energy conservation.In view of this,this paper investigates an integrated model for building energy consumption prediction based on data-driven methods such as clustering,decision trees and proposes a benchmark assessment framework for building energy consumption accordingly.Aiming at the difficulty that the features of building energy consumption data sets are nonlinearly related and there are outliers,an integrated prediction model for building energy consumption based on feature extraction,clustering,and Light Gradient Boosting Machine is proposed.Firstly,the mutual information method and recursive feature elimination are used to make secondary feature selections on the original building dataset to remove redundant features.Then the data with similar features are reasonably clustered based on Gaussian Mixture Model algorithm.The integrated model consisting of a lightweight gradient boosting decision tree fused with random forest is used to train and predict the energy consumption data for each cluster separately.The experimental results show that the prediction results of the proposed model are optimal for each cluster.Based on the above,an energy consumption benchmark evaluation framework based on data-driven is further proposed.Firstly,the integrated energy consumption prediction model is used to predict and calculate the energy efficiency ratio for a subset of buildings after clustering.Secondly,considering the advantages and disadvantages of conventional Energy Star and clustering benchmark evaluation methods,a custom benchmark evaluation method based on cluster-centered energy consumption and predicted energy consumption is proposed.Then the above benchmark evaluation indexes are used to count the cumulative distribution pattern of building energy efficiency ratios,and the Sigmoid fitting algorithm is used to fit them with statistical data to determine the energy consumption evaluation benchmark values and finally get the energy efficiency evaluation grade.The comparative experimental analysis of the energy consumption data sets of commercial and residential buildings shows that the proposed building energy consumption prediction model and benchmark evaluation framework can adapt to different types of building data,and have better prediction accuracy and convergence speed,and the consistency and robustness of the proposed energy consumption benchmark evaluation method are optimal.