An electricity-carbon model construction and ecological carbon sink calculation system

Abstract

The present invention relates to the technical field of carbon emission monitoring and management, and specifically discloses an electric-carbon model construction and ecological carbon sink calculation system, comprising a data acquisition module, an electric-carbon model construction module, and an ecological carbon sink calculation module. The scheme introduces multiple external factors affecting carbon emissions, uses a Lasso regression method for feature selection, ensures that the model comprehensively considers external factors, combines an ARIMA model and a support vector regression model, and simultaneously processes the linear and nonlinear characteristics of data to improve the accuracy and robustness of model prediction; integrates lidar data, hyperspectral images, and terrain data to construct a random forest regression model to achieve carbon storage estimation, constructs a multi-scale spectral feature network to extract spectral features of different scales, and at the same time, constructs a multi-scale spatial feature network to extract spatial features, thereby improving the estimation accuracy of ecosystem carbon reserves.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission monitoring and management, and specifically refers to an electricity-carbon model construction and ecological carbon sink measurement system. Background Art

[0002] To achieve the carbon reduction target, an efficient and accurate carbon emission monitoring and prediction system is needed. The traditional electric carbon model only relies on a single prediction method, without taking into account the differences and applicability of different methods, and fails to fully consider the external factors affecting carbon emissions, resulting in low prediction accuracy and lack of robustness of the model; the traditional ecological carbon sink measurement method is too dependent on a certain data source, lacks multi-source data fusion, and cannot fully utilize the advantages of different data. In addition, the traditional hyperspectral image feature extraction ignores the spatial information of the image and cannot fully reveal the inherent structural characteristics of the data, resulting in deviations in carbon storage estimates. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an electric-carbon model construction and ecological carbon sink calculation system. The traditional electric-carbon model only relies on a single prediction method, does not take into account the differences and applicability of different methods, and fails to fully consider the external factors affecting carbon emissions, resulting in low prediction accuracy and lack of robustness of the model. This solution introduces multiple external factors affecting carbon emissions and uses the Lasso regression method (Least Absolute Shrinkage Selection Operator) for feature selection to ensure that the model comprehensively considers external factors, and combines the ARIMA model (Autoregressive Integrated Moving Average Model) to select the external factors that affect carbon emissions. Average) and support vector regression models can simultaneously process the linear and nonlinear characteristics of the data to improve the accuracy and robustness of model predictions; the traditional ecological carbon sink measurement method is too dependent on a certain data source, lacks multi-source data fusion, and cannot fully utilize the advantages of different data. In addition, the traditional hyperspectral image feature extraction ignores the spatial information of the image and cannot fully reveal the inherent structural characteristics of the data, resulting in the problem of deviation in carbon storage estimation. This scheme integrates lidar data, hyperspectral image data, and terrain data to construct a random forest regression model to estimate carbon reserves, construct a multi-scale spectral feature network, extract spectral features of different scales, and at the same time, construct a multi-scale spatial feature network to extract spatial features, thereby improving the estimation accuracy of ecosystem carbon reserves.

[0004] The present invention provides an electric-carbon model construction and ecological carbon sink calculation system, comprising a data acquisition module, an electric-carbon model construction module, and an ecological carbon sink calculation module;

[0005] The data acquisition module acquires power data and external factors affecting carbon emissions, and calculates the carbon emissions of the target province;

[0006] The electric carbon model building module uses historical carbon emission data to fit the ARIMA model, uses the Lasso regression method to perform feature selection on external factors to obtain the selected features, uses the selected features and the carbon emissions of the target province to establish a support vector regression model, and constructs a combined model by determining the weights of the ARIMA model and the support vector regression model;

[0007] The ecological carbon sink estimation module integrates lidar data, hyperspectral images, and terrain data, extracts features from the lidar data, hyperspectral images, and terrain data, uses a recursive feature elimination algorithm to select the optimal feature set from the extracted features, and constructs a random forest regression model to achieve carbon storage estimation.

[0008] Furthermore, the electric carbon model building module includes building an ARIMA model module, a feature selection module, building a support vector regression model module and building a combined model module;

[0009] The steps of establishing the ARIMA model module and fitting the ARIMA model using historical carbon emission data are as follows:

[0010] Step S1: Use the ADF test (Augmented Dickey-Fuller) to assess the stationarity of historical carbon emission data and determine the difference order ;

[0011] Step S2: construct an autocorrelation function graph and a partial autocorrelation function graph, compare the AIC values ​​of different ARIMA models, and determine the optimal ARIMA model parameters;

[0012] Step S3: Based on the optimal ARIMA model parameters, an ARIMA model is established. The formula used is as follows:

[0013] ;

[0014] In the formula, , Indicates time node , The value of , represents the coefficient, , represents the white noise error term, , represents the ARIMA model parameters, represents a constant term;

[0015] Step S4: Check whether the residual sequence obeys the normal distribution, use the ARIMA model to make short-term predictions, and substitute the prediction results into the ARIMA model for re-prediction;

[0016] The feature selection module uses the Lasso regression method to perform feature selection to obtain the selected features. The formula used is as follows:

[0017] ;

[0018] In the formula, represents the parameter vector, represents the transpose of the parameter vector, Indicates the first samples, represents the number of samples, express of norm, represents the regularization parameter, Indicates that we want to find a parameter vector Minimize the following expression;

[0019] The support vector regression model building module uses the carbon emission data and the selected features to build a support vector regression model, including the following steps:

[0020] Step A1: Data preprocessing, normalizing the carbon emission data and the selected features to obtain a data set;

[0021] Step A2: Data partitioning: randomly divide the data set into training set and test set for model training and testing respectively;

[0022] Step A3: Establish a support vector regression model. The formula used is as follows:

[0023] + ;

[0024] ;

[0025] In the formula, represents the Lagrange multiplier, represents the model tolerance, represents the kernel function, represents the regularization parameter, , Indicates the sample index;

[0026] Step A4: Calculate the prediction results using the following formula:

[0027] ;

[0028] In the formula, represents the prediction function, represents the coefficient vector of the model, represents the transpose operation, represents the mapping function, represents the intercept, represents the kernel function;

[0029] Step A5: Model parameter optimization, using grid search and cross validation to optimize the parameters of the support vector regression model;

[0030] Step A6: Model evaluation: Use the test set to evaluate the model fitting effect, using root mean square error, mean absolute error and mean absolute percentage error as evaluation indicators;

[0031] The combined model building module determines the weights of the ARIMA model and the support vector regression model by minimizing the root mean square error, and builds the combined model. The formula used is as follows:

[0032] ;

[0033] ;

[0034] In the formula, represents the root mean square error, represents the total number of observations in the dataset, Indicates The actual carbon emissions of observations, Indicates The predicted carbon emissions of observations, , denote the weights assigned to the ARIMA model and the support vector regression model, respectively. , Respectively represent the ARIMA model and support vector regression model for the The predicted value of the observed value.

[0035] Furthermore, the ecological carbon sink calculation module includes: a data collection module, a feature extraction module, a machine learning model development module, a carbon storage calculation module and an evaluation module;

[0036] The data collection module collects lidar data, hyperspectral images, and terrain data;

[0037] The feature extraction module extracts features from the laser radar data, the hyperspectral image, and the terrain data, including the following steps:

[0038] Step P1: preprocess the lidar data to obtain tree-level features;

[0039] Step P2: Extracting spectral features and spatial features from hyperspectral images using a multi-scale spectral feature network and a multi-scale spatial feature network, including the following steps:

[0040] Step P21: The spectral data of each sample is regarded as a point with multiple wavelengths, and multi-scale segmentation is achieved through wavelength segmentation to obtain subsets of multiple different scales. The specific calculation formula for the spectral data division of each scale is as follows:

[0041] ;

[0042] ;

[0043] In the formula, represents the original spectral data, Indicates the original spectral data The value of the wavelength point, Indicates the number of wavelength points, Indicates that the original spectral data is divided into After part 1 The spectral data contained in each segmented area;

[0044] Step P22: Input each subset into a multi-scale spectral feature network for feature extraction to obtain spectral features;

[0045] Step P23: Divide the hyperspectral image into multiple blocks of the same size, input the hyperspectral image data of each block into the convolutional long short-term memory network, obtain the features of each block, input the features of each block into the next convolutional long short-term memory network, and obtain the spatial features;

[0046] Step P3: Extracting elevation, slope and aspect terrain features from terrain data;

[0047] The developed machine learning model module uses a recursive feature elimination algorithm to select the optimal feature subset from the extracted features, and uses a random forest regression model to predict forest height, volume, and aboveground biomass;

[0048] The carbon storage estimation module uses the predicted forest height as an additional predictor variable of the random forest regression model to estimate the forest volume and aboveground biomass, and calculates the carbon storage based on the predicted aboveground biomass;

[0049] The evaluation module uses root mean square error, relative root mean square error and mean absolute error as evaluation indicators to evaluate the accuracy of random forest regression model prediction.

[0050] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0051] (1) The traditional electric carbon model relies only on a single prediction method, does not take into account the differences and applicability of different methods, and fails to fully consider the external factors affecting carbon emissions, resulting in low prediction accuracy and lack of robustness of the model. This solution introduces multiple external factors affecting carbon emissions and uses the Lasso regression method for feature selection to ensure that the model comprehensively considers external factors. It combines the ARIMA model and the support vector regression model to simultaneously process the linear and nonlinear characteristics of the data to improve the accuracy and robustness of the model prediction.

[0052] (2) The traditional ecological carbon sink estimation method is overly dependent on a certain data source and lacks multi-source data fusion, which makes it unable to fully utilize the advantages of different data. In addition, the traditional hyperspectral image feature extraction ignores the spatial information of the image and cannot fully reveal the inherent structural characteristics of the data, resulting in the problem of deviation in carbon storage estimation. This scheme integrates lidar data, hyperspectral image data, and terrain data to construct a random forest regression model to achieve carbon storage estimation, construct a multi-scale spectral feature network, extract spectral features of different scales, and at the same time, construct a multi-scale spatial feature network to extract spatial features, thereby improving the estimation accuracy of ecosystem carbon storage. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 To use electricity big data to obtain a framework map of carbon emissions in target provinces;

[0054] Figure 2 It is a multi-scale spectral feature network structure diagram;

[0055] Figure 3 It is a multi-scale spatial feature network structure.

[0056] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0058] Embodiment 1: The present invention provides an electricity-carbon model construction and ecological carbon sink calculation system, including a data acquisition module, an electricity-carbon model construction module, and an ecological carbon sink calculation module;

[0059] The data acquisition module acquires power data and external factors affecting carbon emissions, and calculates the carbon emissions of the target province;

[0060] The electric carbon model building module uses historical carbon emission data to fit the ARIMA model, uses the Lasso regression method to perform feature selection on external factors to obtain the selected features, uses the selected features and the carbon emissions of the target province to establish a support vector regression model, and constructs a combined model by determining the weights of the ARIMA model and the support vector regression model;

[0061] The ecological carbon sink estimation module integrates lidar data, hyperspectral images, and terrain data, extracts features from the lidar data, hyperspectral images, and terrain data, uses a recursive feature elimination algorithm to select the optimal feature set from the extracted features, and constructs a random forest regression model to achieve carbon storage estimation.

[0062] Embodiment 2, based on the above embodiment, the electric carbon model building module includes building an ARIMA model module, a feature selection module, building a support vector regression model module and building a combined model module;

[0063] The steps of establishing the ARIMA model module and fitting the ARIMA model using historical carbon emission data are as follows:

[0064] Step S1: Use the ADF test to evaluate the stationarity of historical carbon emission data and determine the difference order ;

[0065] Step S2: construct an autocorrelation function graph and a partial autocorrelation function graph, compare the AIC values ​​of different ARIMA models, and determine the optimal ARIMA model parameters;

[0066] Step S3: Based on the optimal ARIMA model parameters, an ARIMA model is established. The formula used is as follows:

[0067] ;

[0068] In the formula, , Indicates time node , The value of , represents the coefficient, , represents the white noise error term, , represents the ARIMA model parameters, represents a constant term;

[0069] Step S4: Check whether the residual sequence obeys the normal distribution, use the ARIMA model to make short-term predictions, and substitute the prediction results into the ARIMA model for re-prediction;

[0070] The feature selection module uses the Lasso regression method to perform feature selection to obtain the selected features. The formula used is as follows:

[0071] ;

[0072] In the formula, represents the parameter vector, represents the transpose of the parameter vector, Indicates the first samples, represents the number of samples, express of norm, represents the regularization parameter, To find a parameter vector Minimize the following expression;

[0073] The support vector regression model building module uses the carbon emission data and the selected features to build a support vector regression model, including the following steps:

[0074] Step A1: Data preprocessing, normalizing the carbon emission data and the selected features to obtain a data set;

[0075] Step A2: Data partitioning: randomly divide the data set into training set and test set for model training and testing respectively;

[0076] Step A3: Establish a support vector regression model. The formula used is as follows:

[0077] + ;

[0078] ;

[0079] In the formula, represents the Lagrange multiplier, represents the model tolerance, represents the kernel function, represents the regularization parameter, , Indicates the sample index;

[0080] Step A4: Calculate the prediction results using the following formula:

[0081] ;

[0082] In the formula, represents the prediction function, represents the coefficient vector of the model, represents the transpose operation, represents the mapping function, represents the intercept, represents the kernel function;

[0083] Step A5: Model parameter optimization, using grid search and cross validation to optimize the parameters of the support vector regression model;

[0084] Step A6: Model evaluation: Use the test set to evaluate the model fitting effect, using root mean square error, mean absolute error and mean absolute percentage error as evaluation indicators;

[0085] The combined model building module determines the weights of the ARIMA model and the support vector regression model by minimizing the root mean square error, and builds the combined model. The formula used is as follows:

[0086] ;

[0087] ;

[0088] In the formula, represents the root mean square error, represents the total number of observations in the dataset, Indicates The actual carbon emissions of observations, Indicates The predicted carbon emissions of observations, , denote the weights assigned to the ARIMA model and the support vector regression model, respectively. , Respectively represent the ARIMA model and support vector regression model for the The predicted value of the observed value.

[0089] By performing the above operations, the traditional electric carbon model only relies on a single prediction method, does not take into account the differences and applicability of different methods, and fails to fully consider the external factors affecting carbon emissions, resulting in low prediction accuracy and lack of robustness of the model. This solution introduces multiple external factors affecting carbon emissions, uses the Lasso regression method for feature selection, ensures that the model comprehensively considers external factors, combines the ARIMA model and the support vector regression model, and processes the linear and nonlinear characteristics of the data at the same time, thereby improving the accuracy and robustness of the model prediction.

[0090] Embodiment 3, based on the above embodiment, the ecological carbon sink calculation module includes: a data collection module, a feature extraction module, a machine learning model development module, a carbon storage calculation module and an evaluation module;

[0091] The data collection module collects lidar data, hyperspectral images, and terrain data;

[0092] The feature extraction module extracts features from the laser radar data, the hyperspectral image, and the terrain data, including the following steps:

[0093] Step P1: preprocess the lidar data to obtain tree-level features;

[0094] Step P2: Extracting spectral features and spatial features from hyperspectral images using a multi-scale spectral feature network and a multi-scale spatial feature network, including the following steps:

[0095] Step P21: The spectral data of each sample is regarded as a point with multiple wavelengths, and multi-scale segmentation is achieved through wavelength segmentation to obtain subsets of multiple different scales. The specific calculation formula for the spectral data division of each scale is as follows:

[0096] ;

[0097] ;

[0098] In the formula, represents the original spectral data, Indicates the original spectral data The value of the wavelength point, Indicates the number of wavelength points, Indicates that the original spectral data is divided into After part 1 The spectral data contained in each segmented area;

[0099] Step P22: Input each subset into a multi-scale spectral feature network for feature extraction to obtain spectral features;

[0100] Step P23: Divide the hyperspectral image into multiple blocks of the same size, input the hyperspectral image data of each block into the convolutional long short-term memory network, obtain the features of each block, input the features of each block into the next convolutional long short-term memory network, and obtain the spatial features;

[0101] Step P3: Extracting elevation, slope and aspect terrain features from terrain data;

[0102] The developed machine learning model module uses a recursive feature elimination algorithm to select the optimal feature subset from the extracted features, and uses a random forest regression model to predict forest height, volume, and aboveground biomass;

[0103] The carbon storage estimation module uses the predicted forest height as an additional predictor variable of the random forest regression model to estimate the forest volume and aboveground biomass, and calculates the carbon storage based on the predicted aboveground biomass;

[0104] The evaluation module uses root mean square error, relative root mean square error and mean absolute error as evaluation indicators to evaluate the accuracy of random forest regression model prediction.

[0105] By executing the above operations, the traditional ecological carbon sink estimation method is overly dependent on a certain data source, lacks multi-source data fusion, and cannot fully utilize the advantages of different data. In addition, the traditional hyperspectral image feature extraction ignores the spatial information of the image and cannot fully reveal the inherent structural characteristics of the data, resulting in the problem of deviation in carbon storage estimation. This solution integrates lidar data, hyperspectral image data, and terrain data to construct a random forest regression model to achieve carbon storage estimation, construct a multi-scale spectral feature network, extract spectral features of different scales, and at the same time, construct a multi-scale spatial feature network to extract spatial features, thereby improving the estimation accuracy of ecosystem carbon storage.

[0106] Example 4, see Figure 1 Based on the above embodiment, this embodiment calculates the carbon emissions of the target province, and the steps are as follows:

[0107] Step B1: Calculate the CO2 emission factor using the following formula:

[0108] ;

[0109] In the formula, represents the row vector of the carbon emission matrix, represents the carbon emission matrix, represents the inverse matrix of a diagonal matrix;

[0110] Step B2: Estimate the carbon emission matrix using the grid method. The steps are as follows:

[0111] Step B21: Calculate the carbon dioxide emissions from thermal power generation in each region using the following formula:

[0112] ;

[0113] In the formula, Indicates area of carbon dioxide emissions from thermal power generation, Indicates area Thermal power generation fossil fuels The amount of use, Indicates the total number of fuel types, Represents fossil fuels CO2 emission factors;

[0114] The CO2 emission factor for fossil fuels is calculated using the following formula:

[0115] ;

[0116] In the formula, Represents fossil fuels The net calorific value, Represents fossil fuels The carbon content coefficient, Represents fossil fuels Oxidation ratio;

[0117] Step B22: Calculate the total current for each region using the following formula:

[0118] ;

[0119] In the formula, Indicates area The total current, Indicates from the area Transfer to Region of electricity, Indicates area of power generation, Indicates the number of regions, Indicates area of electricity consumption, Indicates from the area Transfer to Region The amount of electricity;

[0120] Step B23: Based on the inflow or outflow of total electricity in each region, construct a direct outflow matrix using the following formula:

[0121] ;

[0122] In the formula, represents a direct outflow matrix, represents the power transmission matrix between regions, Indicates area To area Transmitted electricity in the area The proportion of total current, Representation Matrix The inverse matrix of

[0123] The total current for each region is converted into an equation in matrix form using the following formula:

[0124] ;

[0125] In the formula, represents the power generation in each region;

[0126] Rewrite the above equation as:

[0127] ;

[0128] In the formula, represents the identity matrix, represents the total outflow coefficient matrix, Indicates from the area Transfer to Region of electricity in the region The proportion of total electricity generation in

[0129] Step B24: Construct the power generation consumption matrix, the formula used is as follows:

[0130] ;

[0131] In the formula, represents the power generation consumption matrix;

[0132] Step B25: Calculate the carbon emission matrix using the following formula:

[0133] ;

[0134] In the formula, Matrix representing CO2 emissions from electricity generation;

[0135] Step B3: Calculate the non-electricity related CO2 emissions using the IPCC inventory method using the following formula:

[0136] ;

[0137] In the formula, represents CO2 emissions not related to electricity, Indicates industry No. The consumption of fossil fuels, represents cement production, It represents the carbon dioxide emission per unit of cement production;

[0138] Step B4: Calculate the CO2 emissions associated with electricity using the following formula:

[0139] ;

[0140] In the formula, represents the carbon dioxide emissions associated with electricity, Indicates area CO2 emission factor for electricity consumption, Indicates area Medium Industry Power consumption;

[0141] Step B5: Calculate the historical electricity-to-CO2 conversion factor using the following formula:

[0142] ;

[0143] In the formula, Indicates the target province industry The power-to-CO2 conversion factor, Indicates the target province industry electricity consumption;

[0144] Step B6: constructing an autoregressive AR model of the power-to-carbon dioxide conversion coefficient based on the historical value of the power-to-carbon dioxide conversion coefficient;

[0145] Step B7: Calculate the carbon emissions of the target province using the following formula:

[0146] ;

[0147] In the formula, represents the carbon emissions of the target province, Indicates the number of industries in the target province.

[0148] Embodiment 5: This embodiment is based on the above embodiment. In the data collection module, a Dualix SpectralImaging portable hyperspectral camera GaiaField Pro-V10 is used to acquire hyperspectral images. The spectral range is 400-1000 nanometers and the spectral sampling interval is 3.2 nanometers.

[0149] Example 6, see Figure 2 This embodiment is based on the above embodiment, and the specific process of step P21 is as follows:

[0150] The original spectral data is divided into data sets of multiple scales to achieve a multi-scale effect. This is done by grouping the data in the wavelength dimension. The spectral vector is continuously halved and input into different LSTM networks respectively. Each LSTM network processes its corresponding spectral data segment and extracts the features of that scale. The features of all scales are merged through a fully connected layer to obtain spatial features.

[0151] Embodiment 7, see Figure 3,This embodiment is based on the above embodiment, and the specific process of step P23 is as follows: the hyperspectral image contains not only spectral information but also spatial information. Efficient extraction of spatial information features can further improve the accuracy of subsequent models, and a multi-scale spatial feature network is constructed. Multi-scale spatial information is used to achieve complementarity between features of different scales. The network design principle is: while obtaining spatial sequence information, hierarchical learning of the main branch is ensured, and multi-scale features are extracted by dividing into multiple parts in the spatial dimension;

[0152] First, the hyperspectral image is subjected to an expansion convolution operation to increase the receptive field while ensuring the resolution of the feature map to obtain more feature information. Then, multiple convolution operations are performed, each of which is used as feature information of different scales, thereby obtaining multi-scale hyperspectral feature information. Assuming that the hyperspectral image is data, where is the horizontal axis length of the hyperspectral image, is the vertical axis length of the hyperspectral image, is the wavelength length of the hyperspectral image. The hyperspectral image is divided into multiple Blocks of size blocks, each Size Each block is input into the convolutional long short-term memory network to obtain the characteristics of each block, and then the characteristics of each block are input into the next convolutional long short-term memory network to obtain the hyperspectral spatial feature information. The calculation formula of the convolutional long short-term memory network is expressed as:

[0153] ;

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] in, is the multidimensional input of the current unit, and is the output and state of the previous unit in the convolutional long short-term memory network, , , and Represents the convolutional long short-term memory network in time The input gate, forget gate, cell state and output of the memory cell, represents the output gate, and are the weights and biases of the input gate, and are the weight and bias of the forget gate, and are the weights and biases of the cell state, and are the weights and biases of the output gate, and is the activation function, is the dot product operation, is a convolution operation;

[0159] Figure 3 ConvLSTM stands for convolutional long short-term memory network.

[0160] Embodiment 8. This embodiment is based on the above embodiment. The external factors affecting carbon emissions mainly include four factors: population, economic development, urban construction and industrial energy consumption, specifically including: urbanization rate, industrial land area, green coverage rate, regional GDP, total industrial output value, the proportion of industry and service industries, industrial electricity consumption, industrial wastewater discharge, industrial waste gas emissions, registered population, population density, and per capita disposable income.

[0161] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0162] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0163] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An electricity-carbon model construction and ecological carbon sink calculation system, characterized by: It includes data acquisition module, electric carbon model building module and ecological carbon sink calculation module; The data acquisition module acquires power data and external factors affecting carbon emissions, and calculates the carbon emissions of the target province; The electric carbon model building module uses historical carbon emission data to fit the ARIMA model, uses the Lasso regression method to perform feature selection on external factors to obtain the selected features, uses the selected features and the carbon emissions of the target province to establish a support vector regression model, and constructs a combined model by determining the weights of the ARIMA model and the support vector regression model; The ecological carbon sink estimation module integrates lidar data, hyperspectral images, and terrain data, extracts features from the lidar data, hyperspectral images, and terrain data, selects the optimal feature set from the extracted features using a recursive feature elimination algorithm, and constructs a random forest regression model to achieve carbon storage estimation; The electric carbon model building module includes building an ARIMA model module, a feature selection module, a support vector regression model module and a combined model building module; The steps of establishing the ARIMA model module and fitting the ARIMA model using historical carbon emission data are as follows: Step S1: Use the ADF test to evaluate the stationarity of historical carbon emission data and determine the difference order; Step S2: construct an autocorrelation function graph and a partial autocorrelation function graph, compare the AIC values ​​of different ARIMA models, and determine the optimal ARIMA model parameters; Step S3: Establishing an ARIMA model based on the optimal ARIMA model parameters; Step S4: Check whether the residual sequence obeys the normal distribution, use the ARIMA model to make short-term predictions, and substitute the prediction results into the ARIMA model for re-prediction; The feature selection module uses the Lasso regression method to perform feature selection to obtain selected features; The support vector regression model building module uses the carbon emission data and the selected features to build a support vector regression model, including the following steps: Step A1: normalize the carbon emission data and the selected features to obtain a data set; Step A2: Randomly divide the data set into training set and test set for model training and testing respectively; Step A3: Establish a support vector regression model; Step A4: Calculate the prediction results; Step A5: Optimize the parameters of the support vector regression model using grid search and cross validation; Step A6: Use the test set to evaluate the fitting effect of the support vector regression model, using root mean square error, mean absolute error and mean absolute percentage error as evaluation indicators; The combined model building module determines the weights of the ARIMA model and the support vector regression model by minimizing the root mean square error to build a combined model; The ecological carbon sink calculation module includes: a data collection module, a feature extraction module, a machine learning model development module, a carbon storage calculation module and an evaluation module; The data collection module collects lidar data, hyperspectral images, and terrain data; The feature extraction module extracts features from the laser radar data, the hyperspectral image, and the terrain data, including the following steps: Step P1: preprocess the lidar data to obtain tree-level features; Step P2: Extracting spectral features and spatial features from hyperspectral images using a multi-scale spectral feature network and a multi-scale spatial feature network, including the following steps: Step P21: The spectral data of each sample is regarded as a point with multiple wavelengths, and multi-scale segmentation is achieved through wavelength segmentation to obtain subsets of multiple different scales; Step P22: Input each subset into a multi-scale spectral feature network for feature extraction to obtain spectral features; Step P23: Divide the hyperspectral image into multiple blocks of the same size, input the hyperspectral image data of each block into the convolutional long short-term memory network, obtain the features of each block, input the features of each block into the next convolutional long short-term memory network, and obtain the spatial features; Step P3: Extracting elevation, slope and aspect terrain features from terrain data; The developed machine learning model module uses a recursive feature elimination algorithm to select the optimal feature subset from the extracted features, and uses a random forest regression model to predict forest height, volume, and aboveground biomass; The carbon storage estimation module uses the predicted forest height as an additional predictor variable of the random forest regression model to estimate the forest volume and aboveground biomass, and calculates the carbon storage based on the predicted aboveground biomass; The evaluation module uses root mean square error, relative root mean square error and mean absolute error as evaluation indicators to evaluate the prediction accuracy of the random forest regression model.

Images