Prediction of Direct Carbon Emissions of Chinese Provincial Residents under Artificial Neural Networks in Deep Learning Environment

It is aimed to deepen the understanding of the consumption carbon emissions of Chinese provinces, establish an accurate and feasible carbon emission prediction model, develop an urban low-carbon economy, and ensure the sustainable development of Chinese cities. Through the national statistical data information, based on the artificial neural network model, mathematical statistics and deep learning methods are used to learn and analyze the carbon emission data of various provinces in China from 1999 to 2019. The neural network toolbox in Matlab is used to program separately to realize the prediction of carbon emissions by different neural network models. After comparing and analyzing the accuracy and prediction performance, the optimal model for the prediction effect is selected. Finally, based on ArcGIS Engine (Arc Geographic Information Science Engine) and C#.NET platform, the call to Matlab neural network toolbox is realized. The selected model is embedded in the prediction system to complete the development of the entire system. The results show that the carbon emissions of residents in the north are distinctly higher than those in the south. Also, with the passage of time, the rate of carbon emissions continues to accelerate. Compared with other models, Elman neural network has higher accuracy and smaller error in carbon emission prediction. Compared to BP (Back Propagation) neural network, the accuracy is improved by 55.93%, and the prediction performance is improved by 19.48%. The prediction results show that China is expected to reach the peak of carbon emissions from 2027 to 2032. This investigation will provide a theoretical basis to control and plan carbon emissions from Chinese urban residents.


25
Recently, with the continuous expansion of the field of human activities, the ecology of nature is facing 26 many problems. A series of environmental problems such as global warming, frequent extreme weather, and 27 melting glaciers threaten human survival and health [1]. Among them, global warming has become an important 28 issue of common concern in today's society and related investigations have also emerged endlessly [2]. It is 29 found that the main factors affecting climate warming come from the continuous emission of CO 2 [3]. To reduce 30 carbon emissions, countries have implemented the most stringent measures [4]. China is the largest developing 31 country in the world today and one of the major energy-consuming countries [5]. As a big country that takes 32 charge of the human living environment, China has the responsibility to actively undertake emission reduction 33 tasks according to its ability while doing well in national economic development [6]. The carbon emissions 34 brought about by rapid urbanization and industrialization as well as the direct carbon consumption of residential 35 energy have become the main part of China's greenhouse gas emissions [7]. Relevant investigations show that in 36 developed countries, as the industrial level continues to decrease, more residents in cities continue to increase 37 their carbon emissions. Some regions have exceeded industrial carbon emissions [8]. Therefore, according to the 38 current energy consumption and carbon consumption levels of residents in various provinces of China, it is 39 greatly significant to predict the direct carbon emissions of Chinese residents.

40
The neural network is a newly developed computer technology in recent years. It relies on its unique 41 network structure characteristics and data processing methods to achieve fruitful results in many fields [9]. It 42 includes engineering automation, image recognition, model prediction, and signal processing. Among them, the 43 use of neural networks to build prediction models is a relatively common method [10]. There   76 (1) 77 Where: i is the i-th region and j is the j-th fuel. When j is 1, 2, 3...19, it represents 19 kinds of energy 78 sources such as coal, oil, natural gas, heat, and electricity.
is the carbon emission of the j-th fuel in the i-th 79 region. is the final consumption of the j-th fuel in the i-th region. is the CO 2 emission coefficient of the 80 j-th fuel. This method needs to count plenty of residents' energy consumption data, which takes a long time.

155
[ Figure. 2 The schematic diagram of the RBF neural network structure] 156

157
Elman neural network is a dynamic feedback neural network, which is a neural network model with local 158 memory and feedback capabilities. For the model, the convergence speed is good and the prediction accuracy is 159 high. Therefore, it is adopted in many fields [22]. Compared with the above two network models, it has an 160 additional undertaking layer with memory and feedback functions. The specific structure is shown in Figure 3.  The above data are analyzed for errors and relative errors.

260
Affected by factors such as the availability of statistical data over the years and different statistical calibers, the 261 calculated direct carbon emission data of residents in various years and provinces will inevitably produce errors.

262
It leads to problems with the conclusions reached.
(2) At present, more empirical methods are used to obtain a 263 more ideal network model. The network testing process is relatively cumbersome and the workload is large. In 264 this process, subjective factors will have a certain impact on network performance. Next, in-depth investigations 265 will be conducted in these two aspects, quickly and efficiently predicting and analyzing the carbon emissions of 266 residents.