Research On University Building Energy Consumption Prediction Based On Deep Learning

In recent years,with the rapid development of the large-scale public building industry,a growing trend appears in the public building energy consumption.Because of their scientific research and teaching tasks,universities have a large number of public buildings so that more energy demand need to be satisfied.In this case,as a place where energy consumption is relatively concentrated,the accurate prediction of energy consumption has great practical significance to the energy conservation planning and energy consumption strategies of university building.However,there are still some problems in the current energy consumption prediction methods of university building in terms of the feature extraction of university building and the correlation between features.Therefore,in order to solve the above problems,this paper will forecast the energy consumption of university building based on the deep learning method,and the main research contents are as follows::(1)Aiming at the characteristics of large amount and multidimensional of building energy consumption data,this paper proposes a pre-processing method that meets these characters.First,outlier recognition is performed by K-Means algorithm.Then the historical data are classified on the basis of the main influencing factors of building energy consumption.In addition,the mean and variance of energy consumption values are extracted as the characteristic values under different energy consumption modes.Finally,as the traditional missing value filling method ignores the problem that the mutation value contains valid information,this article proposed a new missing value filling method based on energy consumption modes classification.Moreover,its effectiveness is also verified by the comparative experiment.(2)Aiming at the problem that the time series prediction model Geo MAN(Multi-level Attention Networks for Geo-sensory Time Series Prediction)cannot model the correlation between features and the traditional attention mechanism does not perform well in long-term real-time series,this paper proposes a Building MAN(Multi-level Attention Networks for Building Energy Consumption Prediction)model.On the one hand,the Building MAN model introduces a full-dimensional convolution mechanism to learn the correlation between input features;on the other hand,it introduces a multimodal fusion attention mechanism to capture the latent information in long-term sequences in a better way.Based on the real building energy consumption data set of a university and two public data sets,this paper designs and conducts several sets of comparative experiments.The experimental results show that compared with the current mainstream models in the industry,the model proposed in this paper performs better on the RMSE and MAE indicators,verifying the effectiveness of the model.