Marine industry carbon emission estimation method based on industry density and residual network

Abstract

The invention discloses an ocean industry carbon emission estimation method based on industry density and a residual network. Comprising the following steps: acquiring spatial geographical distribution data of the marine industry in a target area, and preprocessing the spatial geographical distribution data; constructing a carbon emission estimation model, dividing the preprocessed spatial geographic distribution data into a training set and a test set, performing carbon emission estimation training on the carbon emission estimation model by using the training set, testing the trained carbon emission estimation model by using the test set, and obtaining a trained carbon emission estimation model after the test is passed; and performing carbon emission estimation on the spatial geographic distribution data of the marine industry in the target area to be estimated by using the trained carbon emission estimation model, and outputting an estimation result. Through the method, correlation analysis and regression estimation can be effectively carried out on the spatial layout and the industrial density of the ocean enterprises in the coastal region on the carbon emission, so that the accuracy of estimating the carbon emission of the ocean industry is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission estimation, and in particular to a method for estimating carbon emissions from marine industries based on industry density and residual networks. Background Art

[0002] As an important economic pillar of coastal cities, the marine industry has an increasingly prominent carbon emission problem. The rapid development of the marine industry has not only promoted economic growth, but also brought significant greenhouse gas emissions, facing severe pressure to reduce carbon emissions. Especially in the context of global climate change, the marine industry urgently needs to seek effective carbon reduction strategies to mitigate its impact on the environment. In addition, marine ecosystems are regarded as "blue carbon sinks" and play an important role in carbon capture and storage; however, the relationship between carbon emissions from the marine industry and its industrial density has not been fully studied. This lack of research makes carbon emission management and policy making in coastal areas difficult.

[0003] Existing carbon emission estimation techniques are mostly based on learning and estimating historical carbon emission data and correlative analysis of carbon emissions and other economic and social factors. However, these methods usually ignore the impact of industrial spatial layout on carbon emissions, especially in coastal areas where the concentration of marine industries is high. The shortcomings of traditional methods are: Data limitations: Most existing methods rely on historical carbon emission data for estimation and fail to take into account the impact of spatial industrial density on carbon emissions; Lack of deep learning applications: Currently, there are relatively few studies using deep learning technology in marine industry carbon emission estimation, and its powerful feature extraction and pattern recognition capabilities have not been fully utilized; Insufficient research on the relationship between industrial density and carbon emissions: There is a relative lack of research on the spatial layout of marine enterprises in coastal areas and the correlation analysis and regression estimation of industrial density on carbon emissions, resulting in the failure to consider key factors in carbon emission assessment.

[0004] In summary, the existing carbon emission estimation methods rely on historical carbon emission data for estimation and fail to take into account the impact of spatial industrial density on carbon emissions. They are unable to effectively perform feature extraction and pattern recognition through deep learning technology, and lack effective research on the correlation analysis and regression estimation of the spatial layout of marine enterprises in coastal areas and the impact of industrial density on carbon emissions. As a result, key factors are not considered in carbon emission assessment, which in turn leads to the inability to effectively estimate marine industry carbon emissions or the low accuracy of estimation results. Summary of the invention

[0005] In order to overcome the problems existing in existing marine industry carbon emission technologies, such as lack of effective research on the correlation analysis and regression estimation of the spatial layout of marine enterprises in coastal areas and the industrial density on carbon emissions, which leads to the inability to effectively estimate the marine industry carbon emissions or the low accuracy of the estimation results, the present invention proposes a marine industry carbon emission estimation method based on industrial density and residual network, which can effectively perform correlation analysis and regression estimation on the spatial layout of marine enterprises in coastal areas and the industrial density on carbon emissions, thereby effectively improving the estimation of marine industry carbon emissions and improving the accuracy and robustness of the estimation.

[0006] To achieve the purpose of the present invention, the present invention adopts the following technical solutions:

[0007] A method for estimating carbon emissions from marine industries based on industry density and residual networks, the method comprising the following steps:

[0008] Acquire spatial geographic distribution data of marine industries in target areas, and pre-process the spatial geographic distribution data;

[0009] Constructing a carbon emission estimation model, and dividing the preprocessed spatial geographic distribution data into a training set and a test set, using the training set to perform carbon emission estimation training on the carbon emission estimation model, and using the test set to test the trained carbon emission estimation model, and obtaining a trained carbon emission estimation model after the test passes;

[0010] The trained carbon emission estimation model is used to estimate carbon emissions based on the spatial geographic distribution data of the marine industry in the target area to be estimated, and the estimation results are output.

[0011] In the above technical scheme, by preprocessing the acquired spatial geographic distribution data of the marine industry in the target area, the practicability and reliability of the spatial geographic distribution data can be effectively improved; the constructed carbon emission estimation model can be trained on carbon emission estimation in the training set and tested in the test set to obtain a carbon emission estimation model that effectively grasps the spatial geographic distribution data and carbon emission changes, so that the carbon emission estimation model can be used to estimate the spatial geographic distribution data of the marine industry in the target area to be estimated, and can effectively perform correlation analysis and regression estimation on the spatial layout and industrial density of marine enterprises in coastal areas and carbon emissions, thereby effectively improving the estimation of carbon emissions in the marine industry and improving the accuracy and robustness of the estimation.

[0012] Furthermore, the process of preprocessing the spatial geographic distribution data includes:

[0013] The spatial geographic distribution data of the marine industry in the target area is rasterized to obtain the industry distribution density raster dataset;

[0014] Obtain the carbon emission raster dataset of the same region, and extract the industrial distribution point density data and corresponding carbon emission data of all geographical units based on the industrial distribution density raster dataset and the carbon emission raster dataset;

[0015] The industrial distribution point density data and corresponding carbon emission data of all extracted geographical units are normalized and outlier-free, and the normalized and outlier-free industrial distribution point density data and corresponding carbon emission data are divided into training set and test set.

[0016] Furthermore, the expression for normalizing the industrial distribution point density data and corresponding carbon emission data of all extracted geographical units is:

[0017]

[0018] Among them, x min is the normalized data set, x is the data set, x min is the minimum value in the data set, x max is the maximum value in the data set.

[0019] In the above technical scheme, the spatial geographic distribution data of the marine industry in the target area is rasterized and the carbon emission raster data set of the same area is obtained to extract the industry distribution point density data and the corresponding carbon emission data to improve the efficiency of data processing. The industry distribution point density data and the corresponding carbon emission data of all extracted geographic units are normalized and outliers are removed, which can effectively improve the practicality and reliability of the data.

[0020] Furthermore, the process of using the training set to perform carbon emission estimation training on the carbon emission estimation model includes:

[0021] The constructed carbon emission estimation model includes an initial layer module, a final layer module and several residual block modules connected in sequence;

[0022] Set N rounds of training and input the training set into the carbon emission estimation model;

[0023] Based on the input training set, the initial layer module extracts preliminary correlation features from the industrial distribution point density data and the corresponding carbon emission data;

[0024] According to the extracted preliminary correlation features, several sequentially connected residual block modules use residual networks to extract the deep features of industrial distribution point density data and corresponding carbon emission data;

[0025] The final layer module adjusts the feature dimension of the extracted deep features, estimates the carbon emissions of the marine industry based on the adjusted deep features, and outputs the estimated results;

[0026] The loss function is set to iteratively optimize the carbon emission estimation model during the training process, and the performance of the super-resolution network model after each round of training is tested using the test set. When the number of iterations reaches the preset number or when the loss function converges, the trained carbon emission estimation model is obtained.

[0027] Among them, the carbon emission estimation model is a residual network model based on deep learning.

[0028] Furthermore, the initial layer module extracts preliminary correlation features from industrial distribution point density data and corresponding carbon emission data, including:

[0029] The initial layer module includes an input layer, a convolution layer, a normalization layer, a ReLU activation function layer and a maximum processing layer;

[0030] The input layer is used to convert the input industrial distribution point density data and corresponding carbon emission data into one-dimensional data and transmit them to the convolution layer;

[0031] The convolution layer uses the convolution kernel to perform convolution operation on the input one-dimensional data, and uses the filter to filter the data after the convolution operation to extract the preliminary features of the industrial distribution point density data and the corresponding carbon emission data;

[0032] The extracted preliminary features are normalized using a normalization layer, and nonlinear learning is performed based on the normalized preliminary features using a ReLU activation function layer to learn the correlation between the spatial layout of the industrial distribution point density data and the corresponding carbon emission data in the training set and carbon emissions, so as to extract high-dimensional nonlinear features of the industrial distribution point density data and the corresponding carbon emission data;

[0033] The maximum pooling layer uses pooling kernels to perform feature dimensionality reduction processing on the extracted high-dimensional nonlinear features to obtain preliminary correlation features in the industrial distribution point density data and the corresponding carbon emission data.

[0034] Furthermore, the process of extracting the deep features of the industrial distribution point density data and the corresponding carbon emission data by using the residual network by a plurality of sequentially connected residual block modules includes:

[0035] Each of the residual block modules includes a first convolutional layer, a first normalization layer, a first ReLU activation function layer, a second convolutional layer, a second normalization layer, a jump connection layer and a second ReLU activation function layer, and a plurality of sequentially connected residual block modules form a residual network;

[0036] Based on the residual network design of deep learning, the first convolution layer is used to perform the first step of convolution on the extracted preliminary correlation features to extract the first deep learning features from the industrial distribution point density data and the corresponding carbon emission data. The expression is:

[0037] Y1 = conv(X,W1,S);

[0038] The first normalization layer is used to perform batch normalization on the extracted first-level deep learning features. The expression is:

[0039]

[0040] The first ReLU activation function layer introduces nonlinearity based on the first deep learning feature after batch normalization, and learns the nonlinear characteristics between the spatial layout of the industrial distribution point density data and the corresponding carbon emission data in the training set and carbon emissions. The expression is:

[0041] Y 1relu =max(0,Y 1norm );

[0042] The second convolutional layer performs the second convolution on the extracted nonlinear features to extract the second deep learning features from the industrial distribution point density data and the corresponding carbon emission data. The expression is:

[0043] Y2=conv(X,W2,S);

[0044] The second normalization layer is used to perform batch normalization on the extracted second-level deep learning features. The expression is:

[0045]

[0046] The skip connection layer is used to perform residual connection between the preliminary correlation features of the input residual block module and the second deep learning features after batch normalization processing to obtain the deep features of the industrial distribution point density data and the corresponding carbon emission data. The expression is:

[0047] Yres=X+Y 2norm ;

[0048] The second ReLU activation function layer is used to perform nonlinear processing on the extracted deep features to obtain nonlinear and high-dimensional deep features;

[0049] Among them, X represents the input industry distribution point density data and the corresponding carbon emission data, W represents the convolution kernel, S represents the step size, μ represents the mean, and σ represents the standard deviation.

[0050] Furthermore, the process of adjusting the feature dimension of the extracted deep features by the final layer module includes:

[0051] The final layer module includes a global maximum pooling layer, a fully connected layer and a linear regression layer;

[0052] The global maximum pooling layer is used to reduce the dimension of nonlinear and high-dimensional deep features. The expression is:

[0053] Ygmp=max(Yres);

[0054] The fully connected layer is used to map the depth features after dimensionality reduction to a 1-dimensional space. The expression is:

[0055] Y fc =W fc ·Ygmp+b fc ;

[0056] The linear regression layer is used to perform linear regression processing on the depth features in the 1D space, and the carbon emissions of the marine industry are estimated based on the depth features after linear regression processing, and the estimated value is output. The expression is:

[0057] Yregression=f(Y fc );

[0058] Among them, W fc represents the weight, b fc represents the bias, and f represents the regression function.

[0059] Furthermore, the Adam optimizer was used to optimize the carbon emission estimation model during the training process, and the initial learning rate was set to 0.001.

[0060] In the above technical scheme, by considering the spatial layout of the marine industry and deeply analyzing its carbon emission relationship, by introducing spatial point density, the carbon emission characteristics of the marine industry concentrated areas can be identified, and then the intrinsic connection between industry density and carbon emissions can be revealed, and the carbon emissions of different regions can be estimated more accurately, providing a scientific basis for carbon emission management and policy making in coastal areas; the idea of ​​deep learning residual network is used to estimate carbon emissions, so that the network can learn deeper features, while avoiding the gradient vanishing problem common in traditional deep learning models, and can more effectively extract complex features in carbon emission data, including nonlinear and high-dimensional features, thereby improving the accuracy and robustness of carbon emission estimates.

[0061] A marine industry carbon emission estimation system based on industry density and residual network, the system comprising:

[0062] The acquisition module is used to obtain the spatial geographic distribution data of the marine industry in the target area;

[0063] A processing module, used for preprocessing the spatial geographic distribution data;

[0064] Construction module, used to build carbon emission estimation model;

[0065] The processing module is also used to divide the pre-processed spatial geographic distribution data into a training set and a test set, use the training set to perform carbon emission estimation training on the carbon emission estimation model, and use the test set to test the trained carbon emission estimation model, and obtain the trained carbon emission estimation model after the test passes;

[0066] The carbon emission estimation module is used to use the trained carbon emission estimation model to estimate the carbon emissions of the spatial geographic distribution data of the marine industry in the target area to be estimated, and output the estimation results.

[0067] An electronic device comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of a method for estimating carbon emissions of marine industries based on industry density and residual network are implemented.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] The present invention proposes a method for estimating carbon emissions from marine industries based on industry density and residual networks. By processing the acquired spatial geographic distribution data of marine industries in target areas, the practicability and reliability of spatial geographic distribution data can be effectively improved. The constructed carbon emission estimation model can obtain a carbon emission estimation model that effectively grasps the spatial geographic distribution data and carbon emission changes after being trained on carbon emission estimation in a training set and tested in a test set. Therefore, the carbon emission estimation model can be used to estimate the carbon emissions of the spatial geographic distribution data of marine industries in target areas to be estimated, and can effectively perform correlation analysis and regression estimation on the spatial layout of marine enterprises in coastal areas and the industrial density of carbon emissions, thereby effectively improving the estimation of carbon emissions from marine industries and improving the accuracy and robustness of the estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A flowchart of a method for estimating carbon emissions from marine industries based on industry density and residual networks provided in an embodiment of the present application;

[0071] Figure 2 A schematic diagram of the structure of the carbon emission estimation model provided in the embodiment of the present application;

[0072] Figure 3 A schematic diagram of a residual network provided in an embodiment of the present application;

[0073] Figure 4 A schematic diagram of the training effect of the carbon emission estimation model provided in an embodiment of the present application;

[0074] Figure 5 A graph showing the estimated results of carbon emissions from the marine industry provided in the embodiments of this application;

[0075] Figure 6 A structural schematic diagram of a marine industry carbon emission estimation system based on industry density and residual network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0076] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Preferred embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0077] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0078] Embodiment 1:

[0079] This embodiment provides a method for estimating carbon emissions from marine industries based on industry density and residual networks. Figure 1 , the method comprises the following steps:

[0080] Step S1: obtaining spatial geographic distribution data of marine industries in a target area, and preprocessing the spatial geographic distribution data;

[0081] Step S2: construct a carbon emission estimation model, and divide the preprocessed spatial geographic distribution data into a training set and a test set, use the training set to train the carbon emission estimation model, and use the test set to test the trained carbon emission estimation model, and obtain the trained carbon emission estimation model after the test passes;

[0082] Step S3: Use the trained carbon emission estimation model to estimate carbon emissions based on the spatial geographic distribution data of the marine industry in the target area to be estimated, and output the estimation results.

[0083] In step S1, the process of preprocessing the spatial geographic distribution data includes:

[0084] The spatial geographic distribution data of marine industries in the target area are rasterized in ArcGIS to obtain the industry distribution density raster dataset;

[0085] Obtain the carbon emission raster dataset of the same region, and extract the industrial distribution point density data and corresponding carbon emission data of all geographical units based on the industrial distribution density raster dataset and the carbon emission raster dataset;

[0086] The industrial distribution point density data and corresponding carbon emission data of all extracted geographical units are normalized and outlier-free, and the normalized and outlier-free industrial distribution point density data and corresponding carbon emission data are divided into training set and test set.

[0087] Specifically, the expression for normalizing the industrial distribution point density data and corresponding carbon emission data of all extracted geographical units is:

[0088]

[0089] Among them, x min is the normalized data set, x is the data set, x min is the minimum value in the data set, x max is the maximum value in the data set.

[0090] It can be understood that by rasterizing the spatial geographic distribution data of the marine industry in the target area and obtaining the carbon emission raster data set in the same area, the industry distribution point density data and the corresponding carbon emission data can be extracted to improve the efficiency of data processing. The industry distribution point density data and the corresponding carbon emission data of all extracted geographic units are normalized and outliers are removed, which can effectively improve the practicality and reliability of the data.

[0091] In step S2, the process of using the training set to train the carbon emission estimation model for carbon emission estimation includes:

[0092] See also Figure 2 ,The constructed carbon emission estimation model includes an initial layer module, a final layer module and several residual block modules connected in sequence;

[0093] S21: setting N rounds of training times and inputting the training set into the carbon emission estimation model;

[0094] S22: Based on the input training set, the initial layer module extracts preliminary correlation features from the industrial distribution point density data and the corresponding carbon emission data;

[0095] S23: Based on the extracted preliminary correlation features, a plurality of sequentially connected residual block modules use a residual network to extract the depth features of the industrial distribution point density data and the corresponding carbon emission data;

[0096] S24: The final layer module adjusts the feature dimension of the extracted depth features, estimates the carbon emissions of the marine industry based on the depth features after the dimension adjustment, and outputs the estimation results;

[0097] S25: Setting a loss function to perform iterative optimization training on the carbon emission estimation model during the training process, and using a test set to perform a performance test on the super-resolution network model after each round of training. When the number of iterations reaches a preset number of rounds or when the loss function converges, a trained carbon emission estimation model is obtained. The loss function adopts a mean square error (MSE) loss function to measure the difference between the model estimation result and the actual data.

[0098] Among them, the carbon emission estimation model is a residual network model based on deep learning.

[0099] In step S22, see Figure 2 and Figure 3 The initial layer module extracts the preliminary correlation features from the industrial distribution point density data and the corresponding carbon emission data, including:

[0100] The initial layer module includes an input layer, a convolution layer, a normalization layer, a ReLU activation function layer and a maximum processing layer;

[0101] The input layer is used to convert the input industrial distribution point density data and the corresponding carbon emission data into 1x1x1 one-dimensional data, that is, a scalar value, denoted as X input , and transmitted to the convolutional layer;

[0102] The convolution layer uses 16 1x3 convolution kernels to perform convolution operation on the input one-dimensional data, and uses 16 filters to filter the data after the convolution operation to extract the preliminary features of the industrial distribution point density data and the corresponding carbon emission data, and uses 'same' padding to keep the output size the same as the input, with a step size of 1x2, and the expression is:

[0103] Y = conv(X input ,W,S)Y=conv(X input ,W,S);

[0104] Among them, W is the convolution kernel and S is the step size;

[0105] The normalization layer is used to normalize the extracted preliminary features to normalize the output of the previous convolutional layer to improve the training speed and performance. The expression is:

[0106]

[0107] Among them, μ is the mean and σ is the standard deviation;

[0108] The ReLU activation function layer is used to introduce nonlinearity, and nonlinear learning is performed based on the initial features after normalization. The correlation between the spatial layout of the industrial distribution point density data and the corresponding carbon emission data in the training set is learned to extract the high-dimensional nonlinear features of the industrial distribution point density data and the corresponding carbon emission data. The expression is:

[0109] Y relu =max(0,Y norm );

[0110] The maximum pooling layer uses the pooling kernel to perform feature dimensionality reduction processing on the extracted high-dimensional nonlinear features to obtain preliminary correlation features in the industrial distribution point density data and the corresponding carbon emission data; among them, the maximum pooling layer uses a 1x3 pooling kernel, 'same' padding and a step size of 1x2 to reduce the feature dimension and extract important features.

[0111] In step S23, see Figure 2 and Figure 3 The process of seven sequentially connected residual block modules using residual networks to extract the deep features of industrial distribution point density data and corresponding carbon emission data includes:

[0112] Each of the residual block modules includes a first convolutional layer, a first normalization layer, a first ReLU activation function layer, a second convolutional layer, a second normalization layer, a jump connection layer and a second ReLU activation function layer, and a plurality of sequentially connected residual block modules form a residual network;

[0113] Based on the residual network design of deep learning, the first convolution layer is used to perform the first step of convolution on the extracted preliminary correlation features to extract the first deep learning features from the industrial distribution point density data and the corresponding carbon emission data. The expression is:

[0114] Y1 = conv(X,W1,S);

[0115] The first normalization layer is used to perform batch normalization on the extracted first-level deep learning features. The expression is:

[0116]

[0117] The first ReLU activation function layer introduces nonlinearity based on the first deep learning feature after batch normalization, and learns the nonlinear characteristics between the spatial layout of the industrial distribution point density data and the corresponding carbon emission data in the training set and carbon emissions. The expression is:

[0118] Y 1relu =max(0,Y 1norm );

[0119] The second convolutional layer performs the second convolution on the extracted nonlinear features to extract the second deep learning features from the industrial distribution point density data and the corresponding carbon emission data. The expression is:

[0120] Y2=conv(X,W2,S);

[0121] The second normalization layer is used to perform batch normalization on the extracted second-level deep learning features. The expression is:

[0122]

[0123] The skip connection layer is used to perform residual connection between the preliminary correlation features of the input residual block module and the second deep learning features after batch normalization processing to obtain the deep features of the industrial distribution point density data and the corresponding carbon emission data. The expression is:

[0124] Yres=X+Y 2norm ;

[0125] The second ReLU activation function layer is used to perform nonlinear processing on the extracted deep features to obtain nonlinear and high-dimensional deep features;

[0126] Among them, X represents the input industry distribution point density data and the corresponding carbon emission data, W represents the convolution kernel, S represents the step size, μ represents the mean, and σ represents the standard deviation.

[0127] In step S24, see Figure 2 and Figure 3 ,The process of feature dimension adjustment of the extracted deep features by the final layer module includes:

[0128] The final layer module includes a global maximum pooling layer, a fully connected layer and a linear regression layer;

[0129] The global maximum pooling layer is used to reduce the dimension of nonlinear and high-dimensional deep features. The expression is:

[0130] Ygmp=max(Yres);

[0131] The fully connected layer is used to map the depth features after dimensionality reduction to a 1-dimensional space. The expression is:

[0132] Y fc =W fc ·Ygmp+b fc ;

[0133] The linear regression layer is used to perform linear regression processing on the depth features in the 1D space, and the carbon emissions of the marine industry are estimated based on the depth features after linear regression processing, and the estimated value is output. The expression is:

[0134] Yregression=f(Y fc );

[0135] Among them, W fc represents the weight, b fc represents the bias, and f represents the regression function.

[0136] In some preferred embodiments, the Adam optimizer is used to optimize the carbon emission estimation model during the training process, and the initial learning rate is set to 0.001, the maximum number of training rounds is 10, and the batch size is 32.

[0137] It is understandable that by considering the spatial layout of the marine industry and deeply analyzing its relationship with carbon emissions, and by introducing spatial point density, the carbon emission characteristics of marine industry concentrated areas can be identified, and then the intrinsic connection between industry density and carbon emissions can be revealed. It can more accurately estimate the carbon emissions in different regions and provide a scientific basis for carbon emission management and policy making in coastal areas; the idea of ​​deep learning residual network is used to estimate carbon emissions, so that the network can learn deeper features, while avoiding the gradient vanishing problem common in traditional deep learning models, and can more effectively extract complex features in carbon emission data, including nonlinear and high-dimensional features, thereby improving the accuracy and robustness of carbon emission estimates.

[0138] In this embodiment, see Figure 4 and Figure 5 , a carbon emission estimation result graph is given. By evaluating the quality and accuracy of the carbon emission data downscaling result graph, the following indicators are obtained: the root mean square error (RMSE) is about 0.17, which is basically consistent with the trajectory of the true value curve, indicating that the carbon emission estimation model proposed in the present invention has high performance.

[0139] In this embodiment, by processing the acquired spatial geographic distribution data of the marine industry in the target area, the practicality and reliability of the spatial geographic distribution data can be effectively improved; the constructed carbon emission estimation model can be trained on carbon emission estimation in the training set and tested in the test set to obtain a carbon emission estimation model that effectively grasps the spatial geographic distribution data and carbon emission changes, so that the carbon emission estimation model can be used to estimate the spatial geographic distribution data of the marine industry in the target area to be estimated, and can effectively perform correlation analysis and regression estimation on the spatial layout and industrial density of marine enterprises in coastal areas and carbon emissions, thereby effectively improving the estimation of carbon emissions in the marine industry and improving the accuracy and robustness of the estimation.

[0140] Embodiment 2:

[0141] This embodiment provides a marine industry carbon emission estimation system based on industry density and residual network, see Figure 6 , the system comprising:

[0142] The acquisition module is used to obtain the spatial geographic distribution data of the marine industry in the target area;

[0143] A processing module, used for preprocessing the spatial geographic distribution data;

[0144] Construction module, used to build carbon emission estimation model;

[0145] The processing module is also used to divide the pre-processed spatial geographic distribution data into a training set and a test set, use the training set to perform carbon emission estimation training on the carbon emission estimation model, and use the test set to test the trained carbon emission estimation model, and obtain the trained carbon emission estimation model after the test passes;

[0146] The carbon emission estimation module is used to use the trained carbon emission estimation model to estimate the carbon emissions of the spatial geographic distribution data of the marine industry in the target area to be estimated, and output the estimation results.

[0147] Embodiment three:

[0148] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of a method for estimating carbon emissions of marine industries based on industry density and residual networks are implemented.

[0149] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for estimating carbon emissions from marine industries based on industry density and residual networks, characterized in that: The method comprises the following steps: Acquire spatial geographic distribution data of marine industries in target areas, and pre-process the spatial geographic distribution data; Constructing a carbon emission estimation model, and dividing the preprocessed spatial geographic distribution data into a training set and a test set, using the training set to perform carbon emission estimation training on the carbon emission estimation model, and using the test set to test the trained carbon emission estimation model, and obtaining a trained carbon emission estimation model after the test passes; The trained carbon emission estimation model is used to estimate carbon emissions based on the spatial geographic distribution data of the marine industry in the target area to be estimated, and the estimation results are output.

2. The method for estimating marine industry carbon emissions based on industry density and residual network according to claim 1 is characterized in that: The process of preprocessing the spatial geographic distribution data includes: The spatial geographic distribution data of the marine industry in the target area is rasterized to obtain the industry distribution density raster dataset; Obtain the carbon emission raster dataset of the same region, and extract the industrial distribution point density data and corresponding carbon emission data of all geographical units based on the industrial distribution density raster dataset and the carbon emission raster dataset; The industrial distribution point density data and corresponding carbon emission data of all extracted geographical units are normalized and outlier-free, and the normalized and outlier-free industrial distribution point density data and corresponding carbon emission data are divided into training set and test set.

3. The method for estimating marine industry carbon emissions based on industry density and residual network according to claim 2 is characterized in that: The expression for normalizing the industrial distribution point density data and corresponding carbon emission data of all extracted geographical units is: Among them, x min is the normalized data set, x is the data set, x min is the minimum value in the data set, x max is the maximum value in the data set.

4. The method for estimating marine industry carbon emissions based on industry density and residual network according to claim 2 is characterized in that: The process of using the training set to train the carbon emission estimation model for carbon emission estimation includes: The constructed carbon emission estimation model includes an initial layer module, a final layer module and several residual block modules connected in sequence; Set N rounds of training and input the training set into the carbon emission estimation model; Based on the input training set, the initial layer module extracts preliminary correlation features from the industrial distribution point density data and the corresponding carbon emission data; According to the extracted preliminary correlation features, several sequentially connected residual block modules use residual networks to extract the deep features of industrial distribution point density data and corresponding carbon emission data; The final layer module adjusts the feature dimension of the extracted deep features, estimates the carbon emissions of the marine industry based on the adjusted deep features, and outputs the estimated results; The loss function is set to iteratively optimize the carbon emission estimation model during the training process, and the performance of the super-resolution network model after each round of training is tested using the test set. When the number of iterations reaches the preset number or when the loss function converges, the trained carbon emission estimation model is obtained. Among them, the carbon emission estimation model is a residual network model based on deep learning.

5. The method for estimating marine industry carbon emissions based on industry density and residual network according to claim 4 is characterized in that: The initial layer module extracts the preliminary correlation features from the industrial distribution point density data and the corresponding carbon emission data, including: The initial layer module includes an input layer, a convolution layer, a normalization layer, a ReLU activation function layer and a maximum processing layer; The input layer is used to convert the input industrial distribution point density data and corresponding carbon emission data into one-dimensional data and transmit them to the convolution layer; The convolution layer uses the convolution kernel to perform convolution operation on the input one-dimensional data, and uses the filter to filter the data after the convolution operation to extract the preliminary features of the industrial distribution point density data and the corresponding carbon emission data; The extracted preliminary features are normalized using a normalization layer, and nonlinear learning is performed based on the normalized preliminary features using a ReLU activation function layer to learn the correlation between the spatial layout of the industrial distribution point density data and the corresponding carbon emission data in the training set and carbon emissions, so as to extract high-dimensional nonlinear features of the industrial distribution point density data and the corresponding carbon emission data; The maximum pooling layer uses pooling kernels to perform feature dimensionality reduction processing on the extracted high-dimensional nonlinear features to obtain preliminary correlation features in the industrial distribution point density data and the corresponding carbon emission data.

6. The method for estimating marine industry carbon emissions based on industry density and residual network according to claim 4 is characterized in that: The process of extracting the deep features of industrial distribution point density data and corresponding carbon emission data by using residual network by several sequentially connected residual block modules includes: Each of the residual block modules includes a first convolutional layer, a first normalization layer, a first ReLU activation function layer, a second convolutional layer, a second normalization layer, a jump connection layer and a second ReLU activation function layer, and a plurality of sequentially connected residual block modules form a residual network; Based on the residual network design of deep learning, the first convolution layer is used to perform the first step of convolution on the extracted preliminary correlation features to extract the first deep learning features from the industrial distribution point density data and the corresponding carbon emission data. The expression is: Y1 = conv(X,W1,S); The first normalization layer is used to perform batch normalization on the extracted first-level deep learning features. The expression is: The first ReLU activation function layer introduces nonlinearity based on the first deep learning feature after batch normalization, and learns the nonlinear characteristics between the spatial layout of the industrial distribution point density data and the corresponding carbon emission data in the training set and carbon emissions. The expression is: AND 1relu =max(0,Y 1norm ); The second convolutional layer performs the second convolution on the extracted nonlinear features to extract the second deep learning features from the industrial distribution point density data and the corresponding carbon emission data. The expression is: Y2=conv(X,W2,S); The second normalization layer is used to perform batch normalization on the extracted second-level deep learning features. The expression is: The skip connection layer is used to perform residual connection between the preliminary correlation features of the input residual block module and the second deep learning features after batch normalization processing to obtain the deep features of the industrial distribution point density data and the corresponding carbon emission data. The expression is: Yres=X+Y 2norm ; The second ReLU activation function layer is used to perform nonlinear processing on the extracted deep features to obtain nonlinear and high-dimensional deep features; Among them, X represents the input industry distribution point density data and the corresponding carbon emission data, W represents the convolution kernel, S represents the step size, μ represents the mean, and σ represents the standard deviation.

7. The method for estimating marine industry carbon emissions based on industry density and residual network according to claim 6 is characterized in that: The process of feature dimension adjustment of the extracted deep features by the final layer module includes: The final layer module includes a global maximum pooling layer, a fully connected layer and a linear regression layer; The global maximum pooling layer is used to reduce the dimension of nonlinear and high-dimensional deep features. The expression is: Ygmp=max(Yres); The fully connected layer is used to map the depth features after dimensionality reduction to a 1-dimensional space. The expression is: Y fc =W fc ·Ygmp+b fc ; The linear regression layer is used to perform linear regression processing on the depth features in the 1D space, and the carbon emissions of the marine industry are estimated based on the depth features after linear regression processing, and the estimated value is output. The expression is: Yregression=f(Y fc ): Among them, W fc represents the weight, b fc represents the bias, and f represents the regression function.

8. The method for estimating marine industry carbon emissions based on industry density and residual network according to claim 4 is characterized in that: The Adam optimizer was used to optimize the carbon emission estimation model during the training process, and the initial learning rate was set to 0.

001.

9. A marine industry carbon emission estimation system based on industry density and residual network, characterized in that: The system comprises: The acquisition module is used to obtain the spatial geographic distribution data of the marine industry in the target area; A processing module, used for preprocessing the spatial geographic distribution data; Construction module, used to build carbon emission estimation model; The processing module is also used to divide the pre-processed spatial geographic distribution data into a training set and a test set, use the training set to perform carbon emission estimation training on the carbon emission estimation model, and use the test set to test the trained carbon emission estimation model, and obtain the trained carbon emission estimation model after the test passes; The carbon emission estimation module is used to use the trained carbon emission estimation model to estimate the carbon emissions of the spatial geographic distribution data of the marine industry in the target area to be estimated, and output the estimation results.

10. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images