Research On Prediction Method Of Building Energy Consumption Based On Machine Learning

Construction industry is the backbone of China's national economy which promoted the development of society as an energy-intensive industry.Energy conservation and emission reduction in the construction industry contributes to achieving China's environmental goals of peaking carbon dioxide emissions around 2030.As refined management is one of the methods of building energy saving and emission reduction,so it is necessary to accurate the prediction of building energy consumption to optimizing building operation management and saving energy and reducing emission.In order to complete the building energy consumption prediction work efficiently,this paper applies the machine learning algorithm to the field of building energy consumption prediction.Based on the public building energy consumption data,through the building energy consumption data preprocessing,the building energy consumption feature analysis,the machine learning algorithm energy consumption prediction model,the analysis and evaluation of model prediction results,etc.,established the adaptation relationship among different types of buildings and machine learning algorithm models,and finally concluded a standardized machine learning-based building energy prediction method.The main research contents and achievements of this paper are as follows:First,this paper proposes a method for identifying and repairing abnormal building energy consumption data based on the KNN algorithm and K-means algorithm,and using this method to complete the preprocessing of historical energy consumption data for case buildings.The results of building energy data preprocessing show that this method can repair the abnormal data of building energy consumption reasonably and accurately.Subsequent research is based on this step.Secondly,this paper analyzes the energy consumption characteristics of four public buildings with different uses,and obtains the similarity and difference of energy consumption characteristics among office buildings,commercial buildings,hotel buildings,and medical buildings.This provides the basis for the matching between types of public buildings and different algorithm models in subsequent research.Thirdly,this paper establishes three typical machine-learning-based building energy consumption prediction models based on SVR algorithm,LSTM algorithm and XGBOOST algorithm,and uses the prediction model to carry out energy consumption prediction experiments on four public buildings and conducts experiments evaluation of accuracy and time cost.The results show that different types of buildings and different algorithm models have different adaptability characteristics.When choosing a suitable matching relationship,the building energy consumption prediction model based on machine learning can improve prediction accuracy.Finally,this paper summarizes the matching relationship between different types of buildings and different algorithm models,and finally proposes a complete machine learningbased building energy consumption prediction method,which provides a reference for other building energy consumption prediction research.