| 摘要: |
| 量化和预测用地的碳排放是实现规
划控碳的前提和基础。基于城镇建设用地分
类体系，从城市用地建筑能源消费的碳排放
核算视角，提出以“用地”作为碳排放的核算
终端和核算单元，基于PCA-BP神经网络建
立规划用地碳排放预测模型来预测用地碳
排放。将调研获得的样本地块的碳排放数据
作为因变量，以其用地特征指标（包括：容积
率、建筑单体数量、用地面积、建筑密度、建
筑高度、用地类型、用地兼容性、人口密度）
作为自变量，建立用地碳排放预测模型。以长
兴县老城区为实例，应用该模型预测用地碳
排放，从模型预测结果来看，该方法能较准
确地预测用地的碳排放，为城市的低碳规划
和碳排放管控提供了量化依据。 |
| 关键词: 用地碳排放 碳排放预测 用地指
标 BP神经网络 主成分分析（PCA） |
| DOI：10.13791/j.cnki.hsfwest.20210601 |
| 分类号: |
| 基金项目:国家重点研发计划资助项目（2018YFC0704700）；
国家自然科学基金项目（51878441） |
|
| Prediction of Carbon Emission for Land Use Based on PCA-BP Neural Network |
|
YAN Fengying,LIU Sixian,ZHANG Xiaoping
|
| Abstract: |
| Because the carbon emission intensity of urban construction land is much higher than
that of other land use types, and the carbon emissions produced by different land use methods
are quite different, it is of great significance to make rational use of spatial planning control
means and adjust land use at the spatial planning level with “land use” as the breakthrough
point, so as to achieve carbon control and low-carbon development. Due to the constraints
of weak foundation, few references and difficult measurement, the domestic research on the
quantification of carbon emissions from planned land is generally lagging behind, and it is
impossible to reveal the relationship between carbon emissions from land use and the scale,
type and spatial characteristics of specific land plots from the perspective of land use units.
Quantifying and forecasting the carbon emissions of land use is the premise and foundation
for realizing planning and controlling carbon. However, at present, the barriers of quantitative
analysis of structural emission reduction data in spatial planning are difficult to break, and there
is still a lack of relatively objective and effective tools to measure the carbon emissions of urban
land. In view of this deficiency, from the perspective of carbon emission accounting for building
energy consumption of urban land, a land-use carbon emission prediction method is proposed,
which takes “land use” as the carbon emission accounting terminal and accounting unit, and
then uses PCA-BP neural network to establish a planning land-use carbon emission prediction
model to predict land-use carbon emissions. The core of the land-use carbon emission prediction
model based on PCA-BP neural network is to analyze the known sample data by statistical
analysis method, and establish a mapping relationship, so as to analyze the unknown data, and
replace the artificial consideration of various non-linear relationships which cannot be estimated
by data with machine learning internal intelligent logic. The carbon emission data of sample
land obtained by investigation is taken as dependent variable. The independent variables mainly
consider selecting typical indicators representing land use characteristics, such as floor area ratio,
number of building units, land area, building density, building height, land use type, land use
compatibility, population density and so on. The prediction model of land-use carbon emissions
based on PCA-BP neural network includes eight steps: 1) Standardize data; 2) Dimension
reduction of principal components; 3) Input neural network for training; 4) Initialize weights
and thresholds; 5) Set initial values of variables; 6) Calculate the weight and threshold of reverse
modification; 7) Complete all sample learning and reach the expected error value; 8) Save and
end. Taking the old city of Changxing County as an example, the prediction model of land-use carbon emissions based on PCA-BP neural network was simulated to verify the feasibility and operability of the model. Then it inputs the collected
sample data into the model for calculation. After much training, the average error rate between the simulated value and the actual value of the model
is about 10%, and the error is within the expected range, thus representing the basic success of the model establishment. Some plots in the central
city of Changxing County were selected as sample plots to verify the feasibility and applicability of the model. The predicted values calculated by
the input model were compared with the real values, and the error rate of the predicted results was still below 10%. The prediction results show that
the model has certain feasibility and applicability. In this paper, the carbon emission application model of planning land in the old city of Changxing
County is predicted. According to the classification of planned land use, the carbon intensity values of all kinds of land use were calculated. The
adjusted plots use the model to simulate the intensity value, while the unadjusted plots use the current carbon intensity value for calculation. This
method has a good accuracy in measuring carbon emissions of predicted land. The predicted results can reflect the carbon emission level of various
types of land under the current actual economic development level in a specific region, and can provide reliable judgment and quantitative basis
for carbon emission control of planned land. Guide the adjustment of land structure and indicators in the planning scheme is to achieve the goal of
structural carbon reduction. The operation is relatively simple and practical. The measurement can alleviate the current difficult situation of complex
operation and lack of technical means in carbon emission measurement at the spatial planning level. After the model training is completed, when
it is applied to the same area again, there is no need to repeat the model training. When applied to different regions, it is necessary to re-enter data
and train models to learn the linear and nonlinear relationship between land use characteristic indicators and carbon emissions in this region. After
successful training, it can be used for carbon emission prediction and simulation of planned land in this region. |
| Key words: Carbon Emissions for Land Use Forecasting Carbon Emission Land Index BP Neural Network Principal Component Analysis |
