Research On Building Energy Consumption Prediction Method Based On Deep Learning And Transfer Learning

The rapid development of the construction industry is usually accompanied by serious building energy consumption problems.In this context,the realization of building energy efficiency is particularly important,and it has become an important strategy for the sustainable development of my country’s construction industry.For the building operation stage,relatively accurate and objective building energy consumption prediction results can provide a reference for energy consumption management and energy saving plans in the building operation stage.However,due to the uncertainty and complexity of the building energy consumption system,this puts forward higher requirements for the rationality and accuracy of the building energy consumption prediction method.This article uses this as a starting point,First,optimize the parameters of the deep neural network by improving the intelligent algorithm,use the powerful modeling capabilities of deep learning methods to improve the accuracy of building energy consumption predictions.On this basis,transfer learning is used to improve the prediction performance of deep learning methods in the absence of new buildings and energy consumption sample data.The main research contents are as follows:First,the historical data of building energy consumption was cleaned and normalized,and the relevant influencing factors of building energy consumption were analyzed.The environmental variables are extracted by the method of distance correlation coefficient to remove the redundant features.At the same time,particle swarm algorithm is introduced as the parameter optimization method of LSTM neural network.Aiming at the shortcomings of poor convergence and low accuracy of the standard particle swarm algorithm,a new improvement strategy is proposed from the two aspects of inertia weight and learning factor,and the effectiveness of the algorithm improvement strategy is verified by comparing and analyzing different test functions.Experimental results show that the proposed improvement strategy significantly improves the optimization performance of the standard particle swarm algorithm.Secondly,the super parameters of traditional LSTM neural network are selected according to human experience or grid search,which is subjective and takes a long time to search.Therefore,the IPSO-LSTM building energy consumption prediction model was constructed by optimizing the super parameters of LSTM neural network with improved particle swarm optimization algorithm.The simulation results on the example energy consumption data set show that the prediction effect of IPSO-LSTM building energy consumption prediction model is better than the basic LSTM model and the traditional machine learning prediction model.Finally,the problem of energy consumption prediction in the absence of building energy consumption data and new buildings is studied.On the basis of the constructed prediction model,the maximum mean difference method is used to measure the difference between the source domain data and the target domain data distribution,and the transfer learning method is used to improve the prediction effect of the LSTM neural network under small sample data.From the experimental results,it can be seen the prediction performance of the migrated model is significant compared with the unmigrated prediction model and other machine learning prediction models.improve.In summary,this article combines deep learning and migration learning to apply to building energy consumption prediction,which has certain novelty and reference value.