Combined Prediction Of Building Energy Consumption Based On Max-relevance And Min-redundancy

With the growth of global population and economy,the ratio of energy consumption of new energy buildings account for the global energy consumption has gradually increased.Effective prediction and analyzation of building energy data can promote the planning and control of the energy consumption ratio of the building energy system.This paper takes the energy consumption data of a building in one year provided by the Kaggle as an example.Firstly,the energy consumption data are analyzed by quarterly and monthly time series,so as to dig out the features that are helpful for prediction modeling.Then,according to the nonlinear characteristics of building energy consumption data,multiple models as ARIMA,machine learning model,neural network are used to fit and predict the energy consumption value per hour in the next week,and the prediction performance of each single model is compared according to the prediction error.Finally,An alternative model set is established based on the above single model prediction results,and the MRMR algorithm based on neighborhood mutual information is used to select the sub-model for a combined prediction model.By comparing and analyzing the prediction accuracy,the experiment proves that the combined model established by Max-Relevance and Min-Redundancy algorithm can improve the accuracy of building energy consumption prediction to a certain extent.