Research On Energy Consumption Prediction Of Campus Buildings And Optimization Of Personnel Spatial And Temporal Distribution Based On Machine Learning

China announced that it would strive to reach a peak in carbon dioxide emissions by 2030,and strive for carbon neutrality by 2060.Since our building energy consumption accounts for about 35% of total energy consumption,carbon neutralization and carbon peak in the building industry play an important role in realizing our goals.Low carbon operation of the campus building has a strong demonstration role in realizing carbon neutralization in the building fields.Generally,the low-carbon operation of buildings needs to optimize the building envelope,equipment system,operation mode,number of users,indoor setting parameters,etc.,and then evaluate the building energy consumption prediction model.In this process,the accurate establishment of energy consumption prediction model and the corresponding optimization method are very important.Therefore,this paper mainly studies the prediction models of campus building energy consumption and establishes the prediction models of campus building energy consumption with sufficient information and lacking information respectively.Based on the energy consumption model and particle swarm optimization algorithm,taking the number of students in class as an example,the time and space distribution of staff in the teaching building is optimized to achieve the purpose of reducing the power consumption of the teaching building.The main conclusions of this paper are as follows:(1)A power consumption prediction model based on BP neural network was established for campus buildings with sufficient information,and the influence of different input variables and the number of hidden layer nodes on the model prediction accuracy was compared.When the input variables increased,MSE changed from0.0044 to 0.0037,and the performance evaluation index decreased,indicating that the prediction accuracy was improved.The structure of neural network has no obvious effect on the performance evaluation index,but increasing the number of hidden layer nodes will lead to longer training time.The application of recursive combination prediction method in the power consumption prediction model will reduce the MSE changed from 0.0032 to 0.0037,indicating that the cumulative error in the multi-step combination prediction will lead to the increase of performance evaluation index.(2)For campus buildings lacking in information,this paper proposes a power consumption prediction method based on wavelet decomposition and transfer learning,and uses other building historical data with sufficient information to improve the accuracy of power consumption prediction for buildings lacking in information through transfer learning.The power consumption prediction accuracy of teaching buildings and dormitory buildings was compared,and the influence of source domain,trend term,input variable and wavelet function on the model prediction accuracy was analyzed.The prediction accuracy of structural transfer model and Hephaestus model is compared and verified,and the effectiveness of wavelet decomposition and transfer learning is proved.The recursive combinational prediction method is used to verify the effectiveness of the wavelet decomposition and transfer learning method for cross-time range prediction.By comparing the accuracy evaluation indexes,it is found that the prediction accuracy of buildings lacking information can be improved by 4.03% at the lowest and 98.76% at the highest.(3)Put forward the optimization method for the spatial and temporal distribution of personnel in campus buildings.In order to reduce the power consumption of campus buildings,this paper combines the power consumption prediction model with particle swarm optimization algorithm to optimize the number of users and the opening time of multiple teaching buildings during class.The best number allocation for certified Building A and Building B from Monday to Friday is: In spring semester,the number of students in Building A is [5600,1432,1930,3020,1400],and the number of students in teaching Building B is [3400,7568,7070,5980,7600].At that time,the minimum power consumption of teaching building is 63,538.4 k W·h.In autumn semester,the minimum power consumption of teaching building is 47117.26 k W·h.The opening time is optimized,by calculation,the school spring semester and autumn semester class arrangement,can be appropriate in the spring semester and autumn semester in postpone of 1 weeks,can reach the minimum power consumption of the teaching building,spring semester can reduce 23.24%,autumn semester can reduce 29.17%.