Research On Energy Consumption Prediction Of Commercial Buildings Based On Deep Learning

The rapid development of social economy has led to a rapid increase in building energy consumption in my country.Relevant statistical results show that commercial building energy consumption in some regions accounts for more than 30% of total building energy consumption,showing the characteristics of high proportion and low energy efficiency.Energy consumption prediction is an important task in building energy saving.Accurate energy consumption prediction has important reference significance for commercial building energy saving planning,abnormal equipment detection,and energy use strategy optimization.The reliability of data plays a particularly important role in the research of energy consumption prediction.Appropriate data preprocessing is helpful to improve the overall quality of data set.Therefore,aiming at the problem of outliers in energy consumption data set,a method of abnormal data identification and completion is proposed.Based on the operating characteristics of commercial building,at all time points,on the basis of energy consumption data grouping,using boxplot quickly visually identify the abnormal data,using a certain range at the same time of the point of energy consumption value to mean fill of abnormal data,maximum retention after completion data due time sequence features,it provides a good data basis for the subsequent analysis and research of commercial building energy consumption data.The traditional attention mechanism does not perform well in the Seq2 Seq commercial building energy consumption prediction model with multivariable input over a long period of time.Aiming at the above shortcomings,an energy consumption prediction model based on the TPA attention mechanism,namely,the TPA energy consumption prediction model,is proposed.In this model,the Temporal Pattern Attention(TPA)mechanism was introduced to explore the potential connection between input characteristic variables and energy consumption.At the same time,Bigru network model in deep learning was used as the prediction unit to obtain more internal information between the time steps before and after using its bidirectional network structure.The accuracy of Seq2 Seq prediction model is improved in the case of multivariable inputs over a long period of time.Based on the real energy consumption data of a large commercial building in Shanghai and two open data sets of foreign commercial buildings,the proposed TPA energy consumption prediction model is analyzed in detail by comparative experiments.By comparing with several typical prediction models,it is proved that under the condition of real energy consumption data of commercial buildings,the TPA energy consumption prediction model has the characteristics of high accuracy and good applicability.In order to further improve the accuracy and generalization ability of the TPA energy consumption prediction model,an energy consumption prediction model based on the selection of characteristic information of ET-TPA,namely,ET-TPA energy consumption prediction model,was proposed.The model using limit tree(Extra Trees,ET)embedded feature selection algorithm importance analysis of the characteristic of input data,excluding has nothing to do with the energy consumption of redundant features,select the important features of the optimal information collection,to achieve dimension reduction for the model input,information collection based on the optimal character building energy consumption prediction model.The experimental results show that the method of feature information screening can effectively improve the prediction accuracy compared with the method without feature information screening.At the same time,considering the application of realistic scenarios,the sensitivity analysis of the proposed ET-TPA energy consumption prediction model is carried out,and the prediction performance of the model under different circumstances is discussed in detail.The strong applicability and robustness of the ET-TPA energy consumption prediction model is verified,which lays a good foundation for the practical application of the prediction model.