Research On Building Energy Consumption Prediction Method Based On Generative Adversarial Network

With the acceleration of social development,the trend of increasing energy consumption in large public buildings has become increasingly prominent.The analysis of the change and development characteristics of energy consumption in large public buildings can provide help for the task of building energy conservation in cities.In the current field of building energy conservation,energy consumption prediction plays a very important role.Accurate energy consumption prediction is not only conducive to the analysis of energy consumption in the future so as to provide a certain basis for building energy conservation,but also can help judge building fault information according to the difference between the predicted value and the real value.Therefore,it is very necessary to predict the energy consumption of large public buildings.In this paper,the energy consumption prediction of large public buildings is studied.Firstly,the generative adversarial networks is used to expand the energy consumption data to solve the problem of lack of energy consumption data.Secondly,the energy consumption prediction model was constructed by combining the generative adversarial networks with reinforcement learning to improve the prediction accuracy of building energy consumption by using the advantages of reinforcement learning.Finally,the clustering algorithm is added into the energy consumption prediction model to improve the prediction accuracy of energy consumption between multi-building buildings.The specific research methods are described in the following three points:(1)To solve the problem of insufficient real energy consumption data of large public buildings,this paper proposes a reinforcement learning energy consumption prediction algorithm based on generative adversarial networks.This algorithm combines the advantages of generative adversarial networks and uses sample evaluation unit to evaluate the quality of generated samples when the sample size of energy consumption data is increased.Under the condition that the sample quality reaches a certain level,the original energy consumption data is mixed with the generated samples,which are used as the input data of the energy consumption prediction model of reinforcement learning.Experimental results show that the proposed prediction model can obtain smaller prediction errors compared with other models.(2)To solve the problem of low accuracy in predicting energy consumption of large public buildings,this paper proposes an adaptive sequential generation Re-GAN building energy consumption prediction algorithm.This algorithm combined reinforcement learning with generative adversarial networks,and the generator and discriminant in GAN were respectively constructed as Agent and reward functions in reinforcement learning.The experimental results show that the proposed algorithm has higher prediction accuracy compared with other comparison algorithms.In addition,the experimental part further verifies the influence of different lengths of the real sequence on the prediction accuracy when the current real energy consumption sequence is taken as the input state.(3)In view of the problem that the prediction accuracy of energy consumption among multi-buildings is generally low,this paper proposes a multi-building energy consumption prediction algorithm based on generative adversarial networks.The algorithm uses the clustering algorithm after the data preprocessing stage,that is,before the prediction process of Re-GAN prediction algorithm,the known energy consumption data is clustered first.Then,the data of each cluster are used to train the energy consumption prediction model of Re-GAN separately.Clustering analysis of energy consumption data can help select the appropriate energy consumption prediction model and improve the accuracy of energy consumption prediction.The experimental results show that the proposed algorithm can improve the accuracy of energy consumption prediction in the face of multi-building energy consumption prediction.