Study On The Prediction Of Building Energy Consumption Based On Deep Learning

At present,the prediction of building energy consumption based on machine learning at home and abroad is basically limited to specific areas,and the energy consumption is predicted under the premise of specific building types.Most of the models use Linear Regression,Support Vector Machines,Gaussian Processes,Multiple Adaptive Regression Splines and Random Forests.The forecasting target is divided into qualitative classification and Energy Use Intensity(EUI)values.The data samples of the model have some limitations in terms of quantity and diversity,so the scope of the model is also limited to specific regions and specific building types.This thesis proposes new sources of building energy consumption data and building energy consumption feature extraction methods,using Autodesk’s Green Building Studio and related services as data sources for feature extraction,and then based on deep learning models for EUI predictions were studied.The following aspects have been included in the research:(1)Identification and definition of building energy consumption parameters,addressing geographical and building type constraints;(2)Research on the extraction method of building energy consumption parameters,research and implementation of normalization method;(3)The applicability of deep learning model to building energy consumption prediction is studied;(4)Program implementation of deep learning model and ideas and methods for adjusting and optimizing hyperparameters.(5)Evaluate the prediction results and compare them with the results of Multivariate Linear Regression and Support Vector Regression.Compared with Multivariate Linear Regression and Support Vector Regression,the deep learning model achieved better prediction results of building energy consumption: The Root Mean Square Error(RMSE)reached 0.0357002,the Mean Absolute Percentage Error(MAPE)reached 12.8661,and the Coefficient of Determination(R2)reached 0.93370587.Therefore,to some extent,it is an effective method to predict the building energy consumption by the deep learning model.