Regional carbon emission prediction method, equipment and medium based on Beidou positioning and remote sensing image detection

Abstract

The present invention discloses a regional carbon emission prediction method, device and medium based on Beidou positioning and remote sensing image detection, including: obtaining satellite remote sensing images of a target area, identifying the types of buildings and the number of each type of buildings therein; using the carbon emission prediction model of each building to predict the current carbon emission of the building; collecting the position information of each motor vehicle in the target area based on Beidou positioning to obtain its speed, estimating the carbon emission of each motor vehicle in the target area according to the speed of the motor vehicle and the energy emission factor, the net calorific value of energy, and the characteristic factor; using the hierarchical analysis method to determine the weight of all buildings and motor vehicles in the solution layer to the regional carbon emission; calculating the carbon emission in the target area according to the carbon emission and weight of each building and motor vehicle in the target area. The present invention can effectively solve the problem of obtaining carbon emission data in a large area, and is flexible in application and relatively low in cost.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image detection, and in particular to a method, device and medium for predicting regional carbon emissions based on Beidou positioning and remote sensing image detection. Background Art

[0002] In view of the current international situation and hot trends, carbon dioxide emissions are one of the research hotspots, and it is extremely urgent to build a method for regional carbon emissions prediction. The existing methods for obtaining regional carbon emissions data are mainly equipment monitoring, which obtains accurate data. However, the cost of equipment monitoring is high. For example, the carbon emission trading quota of a company may not exceed tens of thousands of yuan, but other transaction costs such as measuring and verifying its emission data may be as high as hundreds of thousands or even millions of yuan. Moreover, equipment monitoring also has the disadvantages of low flexibility and small monitoring area. Therefore, a large-scale, flexible and relatively low-cost carbon emission prediction method is needed to deal with the problem of monitoring and regulating carbon emissions in any region. Summary of the invention

[0003] The present invention provides a regional carbon emission prediction method, device and medium based on Beidou positioning and remote sensing image detection, which can effectively solve the problem of regional carbon emission data acquisition, provide strong guarantee for the supervision and regulation of regional carbon emissions, and provide new ideas for the subsequent control of the carbon market and the development of clean intelligent navigation.

[0004] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0005] A regional carbon emission prediction method based on Beidou positioning and remote sensing image detection, comprising:

[0006] S1, obtain satellite remote sensing images of the target area and identify the types of buildings and the number of each type of buildings;

[0007] S2, for each building identified by S1, uses the carbon emission prediction model corresponding to its type to predict the current carbon emission of the building based on the historical carbon emission data of buildings of this type;

[0008] S3, based on Beidou positioning, collects the location information of each motor vehicle in the target area, and then obtains the real-time speed information of each motor vehicle, and estimates the carbon emissions of each motor vehicle in the target area according to the speed of the motor vehicle, energy emission factor, energy net calorific value, and characteristic factor;

[0009] S4, taking the weights of all buildings and motor vehicles for regional carbon emissions as the target layer, environmental impact factors as the criterion layer, and all buildings and motor vehicles in the target area as the solution layer, the analytic hierarchy process is used to determine the weights of all buildings and motor vehicles for regional carbon emissions in the solution layer;

[0010] S5, calculating the carbon emissions in the target area according to the carbon emissions of the buildings and motor vehicles in the target area and the weights obtained in S4.

[0011] Furthermore, for buildings that are public buildings, a simple regression model is used to predict the current carbon emissions of the buildings; for buildings that are daily buildings, a moving average model is used to predict the current carbon emissions of the buildings; for buildings that are industrial buildings, a differential autoregressive moving average model is used to predict the current carbon emissions of the buildings, and the dependent variable is set in months as the time unit.

[0012] Furthermore, the types of buildings that are public buildings include but are not limited to stadiums and airports, the types of buildings that are everyday buildings include but are not limited to residential buildings and farms, and the types of buildings that are industrial buildings include but are not limited to factories and chimneys.

[0013] Furthermore, a rotating object detector based on the R3det deep neural network is used to identify building types from satellite remote sensing images.

[0014] Furthermore, the method for estimating the carbon emissions of each motor vehicle in the target area in step S3 is:

[0015]

[0016] Where v represents the speed of the motor vehicle, ρ represents the fuel density of the motor vehicle, and N i represents the emission factor corresponding to the exhaust gas of the i-th fuel of the motor vehicle, F represents the net calorific value of the fuel of the motor vehicle, I i It represents the characteristic factor corresponding to the i-th fuel exhaust gas of a motor vehicle; n represents the number of types of fuel exhaust gas of a motor vehicle.

[0017] Furthermore, environmental factors include temperature, humidity and light intensity.

[0018] Furthermore, step S4 is specifically as follows:

[0019] For the relationship between the target layer and all environmental influencing factors of the criterion layer, a corresponding judgment matrix is ​​constructed, which is recorded as the first type of judgment matrix; and for the relationship between each environmental factor of the criterion layer and the solution layer, a corresponding judgment matrix is ​​constructed, which is recorded as the second type of judgment matrix;

[0020] Calculate the eigenvector and maximum eigenroot of each judgment matrix, perform consistency check on the constructed judgment matrix according to the maximum eigenroot, and determine whether to readjust the judgment matrix to meet the eigenvector corresponding to the judgment matrix of the consistency test;

[0021] A first weight matrix is ​​obtained by performing a cross multiplication calculation on the eigenvectors corresponding to all first-type judgment matrices and satisfying the consistency test; a second weight matrix is ​​obtained by performing a cross multiplication calculation on the eigenvectors corresponding to all second-type judgment matrices and satisfying the consistency test;

[0022] Multiplying the first weight matrix by the second weight matrix yields the weight vector of regional carbon emissions from all buildings and motor vehicles at the scenario level.

[0023] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor implements the method described in any one of the above technical solutions.

[0024] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method described in any of the above technical solutions.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention can effectively solve the problem of obtaining carbon emission data in a large area, and is flexible in application and relatively low in cost. In addition, on the basis of using deep learning neural network models and big data statistical models to calculate carbon emissions, it also fully considers regional environmental factors and provides a hierarchical analysis model that assigns different weights according to the different emphases of the user's consideration of influencing factors, so that the data is more in line with the user's expectations. In addition, the invention provides a strong guarantee for the supervision and regulation of regional carbon emissions, provides new ideas for the subsequent prediction of carbon prices in the carbon market and the development of clean intelligent navigation, and provides strong data support for individuals and enterprises to better participate in and integrate into the carbon market. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the method described in the embodiment of the present application;

[0028] Figure 2 It is a hierarchical structure diagram established when the hierarchical analysis method obtains weights in the embodiment of the present application. DETAILED DESCRIPTION

[0029] The following is a detailed description of an embodiment of the present invention. This embodiment is based on the technical solution of the present invention, and provides a detailed implementation method and a specific operation process to further explain the technical solution of the present invention.

[0030] This embodiment provides a method for predicting regional carbon emissions based on Beidou positioning and remote sensing image detection. Figure 1 As shown, including:

[0031] S1, obtain satellite remote sensing images of the target area and identify the building types and the number of each type of buildings.

[0032] Specifically, a rotating object detector based on the R3det deep neural network is used to identify building types from satellite remote sensing images.

[0033] In this embodiment, building types are divided into three categories: public buildings, daily buildings and industrial buildings. The types of buildings belonging to public buildings include but are not limited to stadiums and airports, the types of buildings belonging to daily buildings include but are not limited to residential buildings and farms, and the types of buildings belonging to industrial buildings include but are not limited to factories and chimneys.

[0034] Before using the rotating object detector to identify building types, you need to build the R3det deep neural network first, and then use satellite remote sensing image samples with labeled building types for learning and training.

[0035] Among them, the training samples of the rotating target detector and the satellite optical remote sensing images of the target area input to the rotating target detector during actual prediction have a spatial resolution of 30 meters, a wavelength range of 0.58 microns, and a radiation resolution of 12 bits. In this embodiment, partitioning is first performed to divide the area into 600*600 remote sensing images, and then preprocessing is performed using the software ENVI5.2Classic, including geometric precision correction and image registration--digital image mosaicking and cropping--atmospheric correction, etc.

[0036] The R3det neural network is an improved single-stage detector based on the RetinaNet neural network, and its AP reaches 40.8. R3det and RetinaNet have the same network structure, including the backbone network and the classification regression subnet.

[0037] The backbone network FPN is mainly divided into three parts when extracting image features: bottom-up lines, top-down lines, and lateral connections.

[0038] The bottom-up circuit is the feedforward calculation of the main convolution layer (RESNET) with a scale step of 2. In the forward process, image features are extracted through convolution operations, and downsampling layers are constructed through pooling operations to reduce the size of features and prevent overfitting. In this way, a series of feature maps are formed. The size of the featuremap will change after passing through some layers, but will not change when passing through other layers. The layers that do not change the size of the featuremap are grouped into one stage, so each extracted feature is the output of the last layer of each stage, thus forming a feature pyramid. Therefore, this part can be used as a feature extraction network for the classification and regression network.

[0039] The top-down process is performed by upsampling, which generally uses an interpolation method, that is, based on the original feature image pixels, a suitable interpolation algorithm is used to insert new elements between pixels, so as to expand the high-level feature map to the same size as the next layer of feature maps, and then merge downward. Here, bidirectional convolution is used to add feature maps to obtain new features. In the refinement stage, only the detection box with the highest score for each feature point is retained to increase the speed and ensure that one feature point corresponds to only one refined bounding box. For each feature point in the feature map, the corresponding five feature vectors are obtained on the feature map according to the five coordinates of the refined box. A more accurate feature vector is obtained through bilinear interpolation, and then these five feature vectors are added and replaced with the current feature vector. After traversing the feature points, the entire feature map is reconstructed. Finally, the reconstructed feature map is added to the original feature map to complete the entire process.

[0040] The horizontal connection is to perform one-dimensional convolution on the up-sampled result and the feature map of the same size generated from the bottom up, and then fuse them. Add a 1×1 convolution layer on C5 to generate the coarsest resolution map, and upsample it by 2 times. Since there are no other layers above P5, P5 is directly mapped from C5; P4 is first upsampled by P5, and then fused with the features mapped from C5; each layer is like this, and finally, a 3*3 convolution kernel and a step size of 1 are used to convolve each fusion result for all the fused layers to eliminate the aliasing effect of upsampling. So corresponding to C2, C3, C4, C5 in the figure below, we get P2, P3, P4, P5 with the same spatial size. The structure of the main network part of RetinaNet is not exactly the same as the structure mentioned in FPN. RetinaNet uses feature pyramid layers P3, P4, P5, P6, P7. Among them, P3, P4, and P5 are generated in the same way as in FPN, and are generated from C3, C4, and C5 through upsampling and lateral connections. P6 is obtained on the basis of C5 through a 3x3 convolution kernel and a step size of 2 convolution. P7 is obtained on the basis of P6 by adding a RELU and then through a 3x3 convolution kernel and a step size of 2 convolution. In addition, for each feature layer, the scale and ratio are changed from 1 and 3 to 3 and 3. In other words, the number of generated anchors increases from 3 to 9.

[0041] In the network, set [32 2 --512 2】Anchors of different sizes are set to {1:2, 1:1, 2:1}. To achieve dense coverage, three sizes are set for each aspect ratio, so that each layer has a total of 9 anchors, and the size range covered by different layers is 32-813. For each anchor, there is a K-dimensional one-hot vector (K is the number of categories) and a 4-dimensional position regression vector.

[0042] Each layer of FPN is associated with a classification regression subnetwork. The predictor used in each feature layer from P3 to P7 in FPN is the same, that is, their weights are shared. In the predictor, the parameters of the classification subnet and the regression subnet are separated, but the structure is similar. They all use a small FCN network with the feature layer as input, and then connect four 3x3 convolutional layers. After each convolutional layer, there is a ReLU activation function. Finally, the classification subnet connects the 3x3 volume layer with the channel (KA) without the activation function. K is the number of detection target types, and the K value selected during training in this embodiment is 28. A is the number of positioning of each feature layer, where A is equal to 9. In order to reduce training parameters, the channel selected for the regression subnet is 4A.

[0043] Based on the above network, in order to adapt to the recognition of rotated images in remote sensing images, the anchor generation method is improved. In order to make it suitable for detecting rotated targets, it is necessary to predict the additional angle offset in the sub-network. Combined with the anchor, the following two sets of data will be obtained.

[0044] t x =(xx a ) / w a , t y =(yy a ) / h a , t w =log(ww a )

[0045] t h =log(hh a ), t θ =(θ-θ a )

[0046] t x ′=(x′-x a ) / w a , t y ′=(y′-y a ) / h a , t w ′=log(w′-w a )

[0047] t h′=log(h′-h a ), t θ ′=(θ′-θ a )

[0048] x, y, w, h, θ——the manually marked anchor center coordinates, width, height and angle;

[0049] x a ,y a , w a ,h a ,θ a ——Generated anchor center coordinates, width, height and angle

[0050] x′, y′, w′, h′, θ′ — predicted anchor center of gravity coordinates, width, height and angle

[0051] t x , t y , t w , t h , t θ Composition nj Target vector, t x ′, t y ′, t w ′, t h ′, t θ ′ constitutes v′ nj The predicted offset vector is used to construct a new multi-class loss function as follows:

[0052]

[0053] t n ' takes the value of 0 or 1 (foreground is 1, background is 0, background has no regression)

[0054] L cls ——focal loss

[0055] L reg ——smooth L1 loss

[0056] p n ——Probability distribution of each category calculated by sigmoid;

[0057] t n - target category;

[0058] λ 1 ,λ 2 ——Hyperparameter, controlling the tradeoff, set to 1 by default

[0059] The anchors are modified using the loss function value L and refined using different IoU thresholds; the thresholds for the foreground and background in the first stage are 0.5 and 0.4 respectively, and the first refinement stage uses 0.5 and 0.4 respectively. If the refinement stage is repeated multiple times, the remaining values ​​are 0.7 and 0.6 respectively. The overall loss of refinement is:

[0060]

[0061] L i ——loss of the i-th refining stage;

[0062] ∝ i ——Weighting coefficient (default is 1);

[0063] S2, for each building identified by S1, uses the carbon emission prediction model corresponding to its type to predict the current carbon emission of the building based on the historical carbon emission data of buildings of this type.

[0064] (1) For buildings that are public buildings, their carbon emission data changes little and the trend of change is close to being flat, so a simple regression model can be used to predict the current carbon emissions of the building;

[0065] The simple regression model uses the least squares method to optimize the parameters in the formula y=ax+b for the loss, and obtains the best linear curve model. The parameters are solved as follows:

[0066]

[0067]

[0068] (2) For buildings of the everyday type, their carbon emissions have certain fluctuations within the range, but are generally mild, so the moving average model can be used to predict the current carbon emissions of the building;

[0069] The moving average model, or MA model, is a linear combination of past residual terms, that is, white noise. Its starting point is to observe the vibration of the residual by combining the residual terms. The model is defined as follows:

[0070]

[0071] (3) For buildings of industrial type, their carbon emissions mostly show seasonal cyclical changes with large fluctuations. Therefore, the differential autoregressive moving average model is used to predict the current carbon emissions of the buildings, and the month is set as the time unit for the dependent variable.

[0072] The ARIMA time series model requires the data to be stationary, so the input data needs to be tested for stationarity first: calculate the mean and variance of the historical carbon emissions time series to see if there is any significant change. If not, the data is stationary. If there is a significant change, perform difference processing on the data, and then calculate the mean and variance of the sample time series again until it is stationary.

[0073] The ARIMA time series model is defined in this embodiment as:

[0074]

[0075] In the formula, p is the number of autoregressive terms, q is the number of moving average terms, and ε t is the error term, μ is the constant term, γ i is the autocorrelation coefficient, θ i is the error term coefficient, y t-i and t are the carbon emissions at time points ti and t respectively.

[0076] The above model is trained using historical data in MATLAB, and ACF (autocorrelation function) and PACF (partial autocorrelation function) are plotted to determine the goodness of fit of the model. If the deviation is too large, the difference order is readjusted and training is performed again.

[0077] The above models trained for different types of buildings can be used to predict the emissions at the current moment using the historical emission data of the buildings, and obtain the current carbon emissions of each building identified from the satellite remote sensing image.

[0078] S3, based on Beidou positioning, collects the location information of each motor vehicle in the target area, and then obtains the real-time speed information of each motor vehicle. According to the speed of the motor vehicle, energy emission factor, energy net calorific value, and characteristic factor, the carbon emissions of each motor vehicle in the target area are estimated.

[0079] In this step S3, firstly, the boundary coordinates of the target area are used as constraints, and the Beidou positioning system is used to obtain the position information of motor vehicles in the area. The speed of the motor vehicle is updated according to the vehicle model and position information to predict the carbon emission data at the current moment.

[0080] Specifically, based on the transformation of the vehicle position information obtained by the Beidou positioning system, the vehicle's moving distance is calculated using the Euclidean distance method in combination with the scale (since the time interval used here is short, about 1s, and the vehicle movement is approximately a linear motion, the Euclidean distance method is used to measure the distance), and the vehicle's moving speed is calculated in combination with the time interval for information collection.

[0081] The method for estimating the carbon emissions of each motor vehicle in the target area in step S3 is:

[0082]

[0083] Where v represents the speed of the motor vehicle, ρ represents the fuel density of the motor vehicle, and N i represents the emission factor corresponding to the exhaust gas of the i-th fuel of the motor vehicle, F represents the net calorific value of the fuel of the motor vehicle, I i represents the characteristic factor corresponding to the i-th fuel exhaust gas of a motor vehicle; n represents the number of types of fuel exhaust gas of a motor vehicle. In this embodiment, it is assumed that the exhaust gas generated by the fuel is mainly carbon dioxide, methane and nitrous oxide, and other gases are relatively negligible.

[0084] The following is an explanation of the above carbon emissions calculation formula:

[0085] (1) Relationship between fuel consumption and speed

[0086] The experiment was conducted on a straight lane with good weather and light traffic. Curve fitting was performed based on the experimental data, and the relationship between vehicle speed (km / h) and fuel consumption per 100 kilometers (L) was obtained as follows:

[0087] y=0.0027v 2 -0.255v+31.8

[0088] Calculate the goodness of fit R 2 =0.93>0.8, meeting the curve fitting requirements;

[0089] (2) Relationship between energy and carbon emissions

[0090] After the above experiments, we collected relevant emission parameters and, based on previous knowledge, preliminarily screened and obtained the following five parameters: energy emission factor, energy net calorific value, characteristic factor, carbon oxidation rate, and carbon dioxide emission coefficient.

[0091] Table 1 Vehicle operation energy emission factors mg / MJ

[0092]

[0093] Table 2 Energy net calorific value and 95% confidence interval limits

[0094]

[0095]

[0096] Table 3 Characteristic factors

[0097]

[0098] Table 4 Carbon oxidation rate

[0099]

[0100] Table 5 Carbon dioxide emission coefficient

[0101]

[0102] Stata software was used to compare and analyze the correlation between the above coefficients and carbon emissions. First, we proposed the formula of carbon emissions as multiplication of each parameter, and then used carbon emissions as the dependent variable and each parameter as the independent variable for correlation analysis.

[0103] Finally, the correlation coefficients were 0.88, 0.83, 0.78, 0.56 and 0.62, respectively, among which the energy emission factor and the net calorific value of energy showed significance. Therefore, in this embodiment, these two parameters were selected to participate in the setting of the carbon emission formula. Since the characteristic factor parameter also showed a high correlation, and in the correlation analysis, it was highly correlated with the energy emission factor, that is, there was a large degree of overlap between the two, so it was also taken into account.

[0104] The above values ​​are used to determine the carbon emissions during fuel combustion, and the formula is as follows:

[0105]

[0106] Combined with the relationship between fuel consumption and speed, we can get the relationship between carbon emissions of motor vehicles and vehicle parameters:

[0107]

[0108] S4, taking the weights of all buildings and motor vehicles for regional carbon emissions as the target layer, all environmental influencing factors including temperature, humidity and light intensity as the criterion layer, and all buildings and motor vehicles in the target area as the scenario layer, the analytic hierarchy process is used to determine the weights of all buildings and motor vehicles for regional carbon emissions in the scenario layer.

[0109] The analytic hierarchy process (AHP) refers to treating a complex multi-objective decision-making problem as a system, decomposing the goal into multiple goals or criteria, and then decomposing it into several levels of multiple indicators (or criteria, constraints). The single ranking (weight) and total ranking of the levels are calculated through the fuzzy quantification method of qualitative indicators as a systematic method for optimizing decision-making with goals (multiple indicators) and multiple plans.

[0110] For the same goal - carbon emissions, because the objects they are produced by are different and are affected differently by the same environmental factors, there will be differences in emissions. This embodiment decomposes environmental factors into different components, and clusters and combines factors at different levels according to the interrelated influences and affiliations between the factors to form a multi-level analysis structure model, so that the problem is ultimately reduced to the determination of the relative importance of the lowest level (schemes and measures for decision-making, etc.) relative to the highest level (overall goal).

[0111] Step S4 specifically includes:

[0112] (1) Establish a hierarchical model: take the weights of all buildings and motor vehicles on regional carbon emissions as the target layer, take all environmental factors including temperature, humidity and light intensity as the criterion layer, and take all buildings and motor vehicles in the target area as the solution layer. Figure 2 The hierarchy diagram shown.

[0113] (2) According to the relationship between each environmental factor in the target layer and the criterion layer, that is, the degree to which the weight of carbon emissions is affected by each environmental factor, a corresponding judgment matrix is ​​constructed. For example, the selected environmental factors are temperature C1 and humidity C2, and the temperature influence degree of the currently selected area is three times that of humidity. Then the constructed judgment matrix is:

[0114]

[0115] This judgment matrix is ​​denoted as ZC, which is the first type of judgment matrix. Similarly, according to the relationship between the influence of each environmental factor on the scheme layer and the criterion layer, a judgment matrix corresponding to each environmental factor is constructed, C1-P, C2-P, which is denoted as the second type of judgment matrix.

[0116] (3) Calculate the eigenvector and maximum eigenroot of each judgment matrix, perform consistency check on the constructed judgment matrix according to the maximum eigenroot, and determine whether to readjust the judgment matrix to obtain the eigenvector corresponding to the judgment matrix that satisfies the consistency check;

[0117] Normalize each column of each judgment matrix, sum the columns of the matrix normalized by column, and then normalize it again to get the eigenvector of the judgment matrix. max Using the formula calculate.

[0118] This embodiment uses the consistency index CI and the consistency ratio CR to perform consistency check: the maximum characteristic root λ of each judgment matrix is max Substitute into the formula The random consistency index RI can be obtained by looking up the table.

[0119] If CI is close to 0, the consistency is better; if CR < 0.1, the matrix inconsistency is within the allowable range. Otherwise, the judgment matrix should be readjusted and the consistency test should be performed again.

[0120] (4) All the eigenvectors corresponding to the first type of judgment matrix and satisfying the consistency test are cross-multiplied to obtain the first weight matrix W1; all the eigenvectors corresponding to the second type of judgment matrix and satisfying the consistency test are cross-multiplied to obtain the second weight matrix W2;

[0121] The first weight matrix W1 is multiplied by the second weight matrix W2, and the result is the weight vector W=W1×W2 of the regional carbon emissions of all buildings and motor vehicles at the scenario level.

[0122] S5, calculating the carbon emissions in the target area according to the carbon emissions of the buildings and motor vehicles in the target area and the weights obtained in S4.

[0123] The above embodiments are preferred embodiments of the present application. Ordinary technicians in this field can also make various changes or improvements on this basis. Without departing from the overall concept of the present application, these changes or improvements should fall within the scope of protection required by the present application.

Claims

1. A regional carbon emission prediction method based on Beidou positioning and remote sensing image detection, characterized in that: include: S1, obtain satellite remote sensing images of the target area and identify the types of buildings and the number of each type of buildings; S2, for each building identified by S1, uses the carbon emission prediction model corresponding to its type to predict the current carbon emission of the building based on the historical carbon emission data of buildings of this type; S3, based on Beidou positioning, collects the location information of each motor vehicle in the target area, and then obtains the real-time speed information of each motor vehicle, and estimates the carbon emissions of each motor vehicle in the target area according to the speed of the motor vehicle, energy emission factor, energy net calorific value, and characteristic factor; S4, taking the weights of all buildings and motor vehicles for regional carbon emissions as the target layer, environmental impact factors as the criterion layer, and all buildings and motor vehicles in the target area as the solution layer, the analytic hierarchy process is used to determine the weights of all buildings and motor vehicles for regional carbon emissions in the solution layer; Step S4 is specifically as follows: For the relationship between the target layer and all environmental influencing factors of the criterion layer, a corresponding matrix is ​​constructed, which is recorded as the first type of judgment matrix; and for the relationship between each environmental factor of the criterion layer and the solution layer, a corresponding judgment matrix is ​​constructed, which is recorded as the second type of judgment matrix; Calculate the eigenvector and maximum eigenroot of each judgment matrix, perform consistency check on the constructed judgment matrix according to the maximum eigenroot, and determine whether to readjust the judgment matrix to meet the eigenvector corresponding to the judgment matrix of the consistency test; A first weight matrix is ​​obtained by performing a cross multiplication calculation on the eigenvectors corresponding to all first-type judgment matrices and satisfying the consistency test; a second weight matrix is ​​obtained by performing a cross multiplication calculation on the eigenvectors corresponding to all second-type judgment matrices and satisfying the consistency test; Multiply the first weight matrix by the second weight matrix to obtain the weight vector of regional carbon emissions from all buildings and motor vehicles at the scenario level; S5, calculating the carbon emissions in the target area according to the carbon emissions of the buildings and motor vehicles in the target area and the weights obtained in S4.

2. The regional carbon emission prediction method according to claim 1, characterized in that: For buildings of the public building type, a simple regression model is used to predict the current carbon emissions of the building; for buildings of the daily building type, a moving average model is used to predict the current carbon emissions of the building; For buildings of industrial type, the differential autoregressive moving average model is used to predict the current carbon emissions of the buildings, and the dependent variable is set in months.

3. The regional carbon emission prediction method according to claim 1, characterized in that: The types of buildings that are public buildings include but are not limited to stadiums and airports, the types of buildings that are everyday buildings include but are not limited to residential buildings and farms, and the types of buildings that are industrial buildings include but are not limited to factories and chimneys.

4. The regional carbon emission prediction method according to claim 1, characterized in that: A rotating object detector based on the R3det deep neural network is used to identify building types from satellite remote sensing images.

5. The regional carbon emission prediction method according to claim 1, characterized in that: The method for estimating the carbon emissions of each motor vehicle in the target area in step S3 is: ； In the formula, represents the speed of the motor vehicle, represents the fuel density of motor vehicles, Indicates the motor vehicle The emission factors corresponding to the emission gases of the fuel are: Indicates the net calorific value of fuel for motor vehicles, Indicates the motor vehicle Characteristic factors corresponding to the exhaust gas of the fuel; Indicates the number of types of fuel exhaust gas from motor vehicles.

6. The regional carbon emission prediction method according to claim 1, characterized in that: Environmental factors include temperature, humidity and light intensity.

7. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the computer program is executed by the processor, the processor is caused to implement the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images