Research On Carbon Emission Prediction Of Electric Power During The Operation Stage Of Teaching Buildings In A University Based On Recurrent Neural Network

In recent years,China has actively responded to the call of the United Nations Framework Convention on Climate Change on reducing greenhouse gas emissions,and proposed the double carbon goals of "carbon peak" in 2030 and "carbon neutrality" in2060,as well as corresponding policies and regulations.As a typical representative of large-scale public buildings,university buildings have significant energy consumption and carbon emissions,which cannot be ignored.They have become a key regulatory object for energy conservation and emission reduction in the 14 th Five Year Plan.To control the total carbon emissions of university buildings,the accounting and prediction of carbon emissions from university buildings is the foundation and prerequisite.Improving the accuracy of building carbon emission prediction models and accurately predicting future building carbon emission trends can provide methodological support and basis for formulating policies and strategies related to energy conservation,emission reduction,and environmental pollution control.Therefore,it is particularly urgent to explore methods to improve the accuracy of carbon emission prediction for university buildings in China.The traditional method of estimating and predicting building carbon emissions using multiple regression equations is linear regression,which cannot infinitely approximate the true value,and the accuracy is often not high enough,with certain limitations in application.To address this issue,this study proposes a new method for calculating building carbon emissions-cyclic neural network prediction of university building carbon emissions-based on 99 months of electricity carbon emissions time series data from four university teaching buildings from January 2015 to March 2023,with the goal of minimizing prediction errors.Firstly,by conducting a detailed investigation and in-depth analysis of the current status of carbon emissions from electrical equipment during the operation phase of a certain university building and the key factors affecting carbon emissions,a carbon emission prediction model is established using a recurrent neural network(RNN)algorithm with excellent nonlinear mapping ability.Its characteristic is that potential patterns can be discovered through carbon emission time series data,and only a small number of explicit influencing factors need to be input as independent variables into the model,There is no need to laboriously search for many undiscovered potential influencing factors,as well as factors that cannot be quantified and data is difficult to obtain.In this paper,Matlab numerical analysis software is used to build a model,and the idea of transfer learning is used to design a cyclic neural network time series model.The model training and testing of the time series data of power carbon emissions in the operation phase of university teaching buildings are carried out,and the dilemma of small sample data is solved.This method has been applied to the prediction of electric power carbon emissions during the operation phase of four teaching buildings in a practical university.The results show that this method can effectively extract temporal features and potential laws,and approximate the true values.Compared with traditional carbon emission prediction,it has higher prediction accuracy.The conclusions drawn from this study are as follows:(1)The number of computers,lighting equipment,and air conditioning usage are significantly related to building carbon emissions.Only by vigorously introducing energy-saving technologies for electrical equipment such as computers,air conditioning systems,and lighting systems,and carrying out energy-saving renovations,can building energy conservation be achieved and the construction of green universities be promoted as scheduled;In addition,the teaching area,experimental area,and administrative area of university buildings have a certain impact on building carbon emissions.During the construction process of university buildings,teaching tasks and functional layout should be reasonably arranged;(2)The influencing factors of carbon emissions from teaching buildings in universities are complex,with diverse hidden variables and difficult to discover patterns.It is necessary to use deep learning methods such as recurrent neural networks to analyze the potential patterns of carbon emission time series data and accurately predict them;(3)To solve the prediction problem in small sample scenarios with a few data,the construction method of transfer learning can be used.This method uses the universality of the feature extraction layer on different datasets,uses stable network models trained on large-scale datasets,freezes the neuron weight information of the feature extraction layer,and then retrains its fitting output layer on small sample datasets,Reestablish the mapping relationship between predicted values and underlying features,so that buildings with short construction years or difficult data acquisition can use a large amount of data from similar large public buildings for model original training.This method is more flexible,making it possible to predict in small sample data scenarios and greatly improving accuracy;(4)This study uses a recurrent neural network(RNN)to construct a prediction model for carbon emissions in university buildings.Through training and testing of the time series data of electricity carbon emissions from four teaching buildings in a certain university,the experimental results validate the prediction error within the range of4.6%,indicating that the hidden laws and fluctuation trends of electricity carbon emissions during the operation phase of these four teaching buildings have been successfully captured by the prediction model,And achieved accurate prediction of building carbon emissions within the error range.The establishment of a carbon emission prediction model for the electricity of teaching buildings in universities has a certain positive contribution and practical value for low-carbon and energy-saving transformation and reasonable arrangement of teaching tasks.It can predict and adjust school energy management measures,thereby reducing the cost of energy use in university buildings and reducing greenhouse gas emissions such as building CO2.The research results of this study not only enrich the literature research on building carbon emission prediction in theory,but also provide important reference basis for universities to develop energy management and carbon emission warning mechanisms,and then lay the groundwork for universities’ "carbon quotas" and "carbon trading",helping countries achieve the "dual carbon goals" as scheduled.