Research On Application Of Deep Belief Networks In Public Building Energy Consumption Prediction

Precise and effective prediction of building energy consumption has guiding significance for building energy management and smart building development.This thesis carries out the research of energy consumption prediction for a public office building in Henan Province during the summer cooling season.A public building energy consumption prediction model based on deep belief network is established by using unit operation data,local meteorological data and building energy consumption data.And through outlier detection and model parameter optimization,the model is optimized to improve model prediction accuracy.The unit operation data in the original data set is measured by the water source heat pump system experiment in the public building,and the meteorological data is collected by the local weather station.For the unreasonable data in the original data set,Seasonal-Trend decomposition procedure based on Loess(STL)method and the boxplot method are separately used to detect the outliers.Data that are detected as outliers by both methods are considered outliers and removed.We divide the outlier-processed data into a training set and a test set according to a ratio of 3:1.We use a deep belief network(DBN)algorithm to establish a public building energy consumption prediction model,and establish two prediction models as a comparison model.The results show that compared with the other two models,the deep belief network model has an average absolute error and a mean square error that are both less than the two comparison models,and the decision coefficients are greater than the two comparison models.This shows that the deep belief network model that can reconstruct the feature variables has a good prediction effect.Aiming at the influence of network parameters on the performance of prediction models in deep belief networks,a selection strategy for optimizing the parameters of deep belief networks is proposed.A combination of experiments and experience is used to determine the optimal number of hidden layers and nodes in the prediction model.Then we use the cuckoo search(CS)algorithm to optimize the kernel parameter g and the penalty factor C in the prediction model to obtain the optimal parameters of the model.When the number of hidden layers is 2 and the number of hidden layer nodes is 9 and 15,respectively,the optimal consumption prediction model is obtained.Compared with the model before optimization,the mean absolute error of the model after optimization is reduced from 1.72 to 1.05,the mean square error decreased from 11.94 to 1.36,and the coefficient of determination changed from 0.88 to 0.98.The optimization effect is significant.This thesis establishes a public building energy consumption prediction model based on the deep belief network algorithm,and optimizes the model parameters through the cuckoo search algorithm to obtain a public building energy consumption prediction model with higher prediction accuracy.The model can better predict the change trend of public building energy consumption,and provide decision-making and theoretical support for building energy efficiency.