Smart Building Behavior Modeling And Prediction Method Based On Machine Learning

With the rapid development of science and technology,the application of smart building in people's lives gradually penetrate,machine learning also plays a more important role in daily life.In this paper,we use EnergyPlus software to build complex simulation models for smart buildings,extract historical data from models as eigenvectors for machine learning,and propose smart building behavior modeling and forecasting methods based on machine learning.In this paper,two different machine learning algorithms,support vector regression(SVR)and recurrent neural network(RNN)are used to model the smart building and put the two algorithms into the same data coordinate system and eigenvector.Based on the accurate prediction of this model,smart buildings can be indoor lighting systems,air conditioning systems,ventilation systems for more energy efficient,more environmentally friendly control,thereby reducing unnecessary energy consumption.In addition,the method can also be applied to real-time monitoring or security monitoring and other fields,so as to better protect public property and personnel safety.The smart building model is generated by the software EnergyPlus programming to get the time series data of the simulated smart building,such as ambient temperature,outdoor temperature,heating heat,ventilation heat consumption,air conditioning system heat,electronic equipment energy consumption and lighting consumption..These data are extracted into the eigenvectors of machine learning,and we use the machine learning algorithm to successfully complete the prediction model.By observing the difference between the accuracy of the training data and the test data,the size of the eigenvectors and the parameters of the model can be optimized to make the prediction more accurate.The results show that the smart building model based on EnergyPlus is a more accurate simulation scheme,and the prediction accuracy of machine learning is high,and different machine learning algorithms have different characteristics in the prediction results.