Research On The Prediction Of Building Energy Consumption Based On Transfer Reinforcement Learning

With the continuous development of Chinese economy,solving energy consumption has become an inevitable major challenge in the process of social development.At present,Chinese total energy consumption is increasing,and the proportion of total building energy consumption has exceeded 30%in total energy consumption.At the same time,with the rapid development of urban construction,the number of buildings has increased rapidly,and people's material level has increased.The requirements for the comfort of the environment continue to increase,resulting in building energy consumption continues to grow^[1].In order to alleviate the current demand for building energy consumption,the issue of intelligent building energy efficiency has become an important research point in the construction field.The analysis and prediction of building energy consumption is an important prerequisite for carrying out work related to intelligent building energy efficiency.Therefore,accurate and efficient energy consumption prediction is one of the important research tasks of intelligent building energy efficiency.This paper will focus on the study of building energy consumption prediction based on the methods of transfer learning and reinforcement learning.The main researches are divided into 3 parts:?1?Firstly,the accuracy of building energy consumption prediction affects the related work of building energy conservation.This chapter combines the transfer learning to improve the Sarsa algorithm,improve the accuracy of energy consumption prediction,a building energy based on value function transfer is proposed.By introducing Bisimulation metric and uses it to measure the similarity between source tasks and target tasks in which those two tasks have the same state space and action space.In addition,the algorithm introduces Bayesian inference and uses variational inference to measure information gain,finally,using the obtained information gain to build intrinsic reward function model as exploring factors,to speed up the convergence of the algorithm and the energy consumption prediction performance.?2?Secondly,the features of building energy consumption are complex,which affects the building energy consumption prediction performance.To reduce the complexity,a building energy consumption prediction method based on feature transfer reinforcement learning is proposed.This method uses the Stacked Denoising Autoencoder to learn the deep features of building energy consumption and transfer useful information of the different building energy consumption.On the other hand,the trained model states output set is used as the input of the reinforcement learning Sarsa algorithm,combined MDP to build energy Consumption modeling and construct reward function.At the same time,the Sarsa algorithm is used to realize building energy consumption prediction and improve energy consumption prediction accuracy.?3?Finally,the building energy consumption has many influencing factors,and the energy consumption data samples are insufficient,which affects the building energy consumption prediction performance.Therefore,a building energy consumption prediction method with dimensional reduction and self-transfer reinforcement learning is proposed,which uses sparse coding to unify the dimensions of MDP in different buildings and extract the important dimensions that affect the energy consumption of buildings,reducing the dimensions,using European metrics to transfer the source building MDP states that meet the transfer conditions,solving the problem of insufficient MDP energy consumption samples in the target building,and combining the Sarsa algorithm to build the building energy consumption and reward function model.Which realize energy consumption prediction and improve energy prediction performance.The algorithms proposed in the first and two parts all do energy consumption prediction experiments.The experimental results show that the proposed algorithms are both higher than the traditional Sarsa algorithm.At the same time,under the multi-dimensional influencing factors,the energy prediction performance of the proposed method is further improved.