Building Energy Consumption Prediction Based On Generative Adversarial Network

In recent years,with the rapid development of society,China’s building energy consumption is also increasing.Excessive energy consumption will lead to environmental problems such as increased greenhouse gas emissions.Therefore,it is necessary to promote building energy conservation and emission reduction.The key to building energy conservation and emission reduction is to predict the future trend of building energy consumption.It is helpful for building energy conservation planning and energy consumption strategy that predict the accurate trend of building energy consumption.However,the existing building energy consumption prediction methods are difficult to model the building energy consumption data with high complexity and nonlinearity,and it is difficult to capture the time dependence of the data.For the non-stationary and abrupt multivariable building energy consumption series,how to consider the interaction between various variables and capture the dependence patterns in the data is still a major challenge in building energy consumption prediction.Therefore,this thesis will use generative adversarial network to study the above problems.The main research contents are as follows:（1）In view of the characteristics that building energy consumption is affected by many factors and building energy consumption data is highly complex,a building energy consumption prediction model based on generative adversarial network is proposed in this thesis.Using the confrontation mechanism of generating confrontation network,we can learn the hidden layer relationship between building energy consumption and its influencing factors.By designing an appropriate objective function and network structure for the generative adversarial network,the generation adversarial network is applied to the prediction of building energy consumption.The experimental results show that its prediction performance is good.（2）In view of the non-stationary and abrupt characteristics of building energy consumption series,this thesis integrates the retrospective prediction idea into the generative adversarial network,and proposes a building energy consumption prediction model based on the retrospective generative adversarial network.The idea of retrospective prediction is that if the prediction result is realistic,then the generator should give a realistic past result even the predicted future result is given as input.Based on this idea,we add a retrospective generator in GAN,change the input format of the discriminator,and restrict the trend consistency of future energy consumption value and past energy consumption value through the loss function.Finally,the prediction results of the model are compared with those of many common methods through experimental operation.The results show that our model can well analyze and process the non-stationary and abrupt time series of building energy consumption,capture the time dependence of data and improve the prediction accuracy.