Electricity-carbon calculation model parameter self-tuning method, device, equipment, medium and product

Abstract

The invention belongs to the technical field of electrical carbon emission monitoring, and particularly relates to an electrical carbon calculation model parameter self-tuning method, device and equipment, a medium and a product. Comprising the following steps: collecting historical carbon emission and actual power utilization related influence factor data of cities, regions and industries; screening the data according to a hierarchical relationship that a city is used as a top layer and regions and industries are used as sub-layers, and constructing a three-cluster coupling training set; based on the three-cluster coupling training set, constructing an optimal training set containing the carbon emission and the influence factors strongly related to the carbon emission; constructing three electricity-carbon calculation models based on a deep BP neural network from three dimensions of city, region and industry, and training the electricity-carbon calculation models by using the optimal training set; and performing hyper-parameter optimization on the deep BP neural network by using a Bayesian optimization algorithm, and adjusting the hyper-parameter by outputting a result back-band based on a city, region and industry data relationship. According to the invention, the prediction precision and efficiency of the electrical carbon calculation model are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric carbon emission monitoring, and in particular relates to a method, device, equipment, medium and product for self-tuning parameters of an electric carbon calculation model. Background Art

[0002] With the rapid development of smart grids, electricity-carbon calculation models play an increasingly important role in promoting low-carbon energy transformation and power system optimization. Such models need to accurately assess carbon emissions in the power system and provide a scientific basis for policy making and power dispatch. However, traditional electricity-carbon calculation models often rely on fixed parameter settings, resulting in insufficient adaptability and flexibility, and are unable to cope with the complex and changing operating environment and user behavior in the power system.

[0003] In this context, deep BP neural networks have gradually become an important part of electric carbon calculation models due to their powerful data fitting capabilities. However, parameter optimization of deep learning models usually requires a lot of computing resources and time, and the manual adjustment process is easily limited by experience and it is difficult to achieve the optimal effect. In order to solve this problem, Bayesian optimization, as an efficient global optimization method, has gradually been applied to parameter self-tuning of deep learning models. By intelligently exploring the parameter space, Bayesian optimization can find the optimal model parameters with a smaller number of experiments, thereby improving the performance of the model. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for self-tuning parameters of an electric carbon calculation model, comprising:

[0005] Collect data on historical carbon emissions and actual electricity consumption-related influencing factors for cities, regions, and industries;

[0006] The data are screened according to the hierarchical relationship of cities as the top layer and regions and industries as sub-layers, and a three-cluster coupling training set is constructed;

[0007] Constructing an optimal training set including carbon emissions and factors strongly correlated with carbon emissions based on the three-cluster coupled training set;

[0008] Constructing three electric carbon calculation models based on deep BP neural network from three dimensions of city, region and industry, and training the electric carbon calculation models using the optimal training set;

[0009] The Bayesian optimization algorithm is used to optimize the hyperparameters of the deep BP neural network, and the hyperparameters are adjusted by outputting the results based on the relationship between city, region, and industry data.

[0010] Furthermore, the method comprises:

[0011] Evaluate the accuracy of the deep BP neural network electro-carbon calculation model built based on the Bayesian optimization algorithm.

[0012] Furthermore, the construction of the optimal training set containing carbon emissions and factors strongly correlated with carbon emissions includes: first preliminarily screening out weak features in the carbon emission influencing factors based on principal component analysis and correlation analysis, and then further screening using elastic net regression to construct the optimal training set containing carbon emissions and factors strongly correlated with carbon emissions.

[0013] Furthermore, the BP neural network includes an input layer, a hidden layer, and an output layer. The hidden layer is a single layer, and neurons are non-connected within the layer and fully connected outside the layer.

[0014] Furthermore, the electric-carbon calculation model determines the final weights and biases of the electric-carbon calculation model through forward propagation and back propagation.

[0015] Furthermore, the hyperparameters include multiple key parameters such as learning rate, batch size, number of hidden layers, and number of nodes.

[0016] Furthermore, the hyperparameters are adjusted by outputting the results based on the relationship between the city, region and industry data, including: forming a total-score data relationship between the city and the region or / industry, and forming a mutual verification relationship between the region and the industry, that is, the overall carbon emissions of the city are equal to the sum of the regional carbon emissions, and also equal to the sum of the industry carbon emissions.

[0017] Furthermore, the evaluation is based on the accuracy of the deep BP neural network electric carbon calculation model constructed by the Bayesian optimization algorithm, including the use of mean absolute percentage error, mean square error, and determination coefficient to evaluate the model performance from multiple angles.

[0018] The present invention also provides a device for self-tuning parameters of an electric carbon calculation model, which is characterized by comprising:

[0019] Data collection module: used to collect multi-dimensional data of cities, regions and industries;

[0020] Training set construction module: used to screen data according to the hierarchical relationship with cities as the top layer and regions and industries as sub-layers, and construct a three-cluster coupling training set;

[0021] Optimal training set construction module: used to construct the optimal training set including carbon emissions and factors strongly related to carbon emissions;

[0022] Electricity carbon calculation model construction module: used to construct three electricity carbon calculation models based on deep BP neural network from three dimensions: city, region and industry according to the historical carbon emissions and actual electricity consumption data distribution;

[0023] Hyperparameter optimization module: used to optimize the hyperparameters of deep BP neural networks using the Bayesian optimization algorithm, and adjust the hyperparameters by outputting the results based on the relationship between city, region, and industry data.

[0024] Furthermore, the device comprises:

[0025] Model accuracy evaluation module: used to evaluate the accuracy of the deep BP neural network electric carbon calculation model built based on the Bayesian optimization algorithm.

[0026] The present invention also provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the computer program is executed by the processor, the processor executes the steps of the electric carbon calculation model parameter self-tuning method.

[0027] The present invention also provides a computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, the processor executes the steps of the electric carbon calculation model parameter self-tuning method.

[0028] The present invention also provides a computer program product, including a computer program, characterized in that the computer program is executed by a processor to implement the self-tuning method of the electric carbon calculation model parameters.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] 1. By systematically collecting and screening data on carbon emissions and related influencing factors, the quality and relevance of input data are ensured, thereby improving the effectiveness of model training.

[0031] 2. Combining principal component analysis and elastic net regression, an optimal training set containing strongly correlated influencing factors was constructed to ensure the accuracy and stability of the model.

[0032] 3. The deep BP neural network architecture is adopted to effectively improve the fitting ability of the model, enabling it to better reflect the complex electricity-carbon relationship.

[0033] 4. Automatically tuning the model hyperparameters through the Bayesian optimization algorithm significantly reduces the time cost of manual parameter adjustment and improves the efficiency and accuracy of model optimization; through the mutual verification, judgment and return of the Bayesian optimized model parameter self-tuning iteration of the three electric carbon calculation sub-models of city-region-industry, the correctness of the self-tuning of the electric carbon calculation model parameters proposed in the present invention is guaranteed.

[0034] 5. The performance of the model was evaluated by mean absolute percentage error, mean square error and coefficient of determination, ensuring its reliability in practical applications.

[0035] In summary, the present invention not only improves the prediction accuracy and efficiency of the electric carbon calculation model, but also provides strong data support and decision-making basis for realizing the low-carbon development of smart grids, promotes the sustainable development of the power system, and supports the realization of dual carbon goals.

[0036] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for self-tuning parameters of an electric carbon calculation model provided in an embodiment of the present invention is shown;

[0039] Figure 2 It shows the total carbon emissions of a large city from 2005 to 2022;

[0040] Figure 3 A schematic diagram of the optimal training set screening and aggregation process of the electric carbon calculation model based on elastic net regression provided in an embodiment of the present invention is shown;

[0041] Figure 4 A schematic diagram of the optimal training set screening and aggregation process of an electric carbon calculation model based on elastic net regression in a method for self-tuning parameters of an electric carbon calculation model provided in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] See also Figure 1 As shown, a method for self-tuning parameters of an electric carbon calculation model includes:

[0044] S1: Collect data on historical carbon emissions and actual electricity consumption-related influencing factors for cities, regions, and industries.

[0045] See also Figure 2 As shown, data on carbon emissions, electricity consumption, energy, economy, population, etc. of cities, regions and industries are collected through the City Statistical Yearbook and the China Carbon Accounting Database.

[0046] The data types of the city statistical yearbook mainly include:

[0047] Economic data: including GDP area 、Industrial output value Output ind , Service Industry Development Index η serv , investment and consumption indicators η Inv-Con , reflecting the economic development situation.

[0048] Demographics: covers the total population Num p 、Population structure type C p , employment index η emp , Education Level Index etc., to provide social development background.

[0049] Energy consumption: detailed statistics of various energy sources - coal, natural gas, oil usage coal , U gas , U oil and consumption Cos coal 、Cos gas 、Cos oil , reflecting the changing trend of energy use.

[0050] Environment and carbon emissions: including total carbon dioxide emissions Em co2 、Industry emissions Em co2-pro 、Industry emissions Em co2-area 、Renewable energy usage Amn sust etc., to provide a basis for analyzing climate change and environmental protection.

[0051] Transportation: Provides various types of transportation data, including the number of vehicles Num c Traffic c etc., reflecting the impact of transportation development on the environment.

[0052] China's carbon accounting database contains emission and energy inventories for China and its provinces from 2014 to 2019. The data calculates the energy and carbon dioxide emission inventories of 31 provinces in China from 2014 to 2019, including the national emission inventory for 2014-2017 updated based on the latest revised energy activity data of the National Bureau of Statistics. The updated emission inventory continues the previous accounting method and caliber, covering 47 social and economic sectors, 17 types of fossil fuel combustion and cement production-related process emissions.

[0053] S2: Screen the data according to the hierarchical relationship with cities as the top layer and regions and industries as sub-layers, and construct a three-cluster coupling training set.

[0054] The data is filtered according to the hierarchical relationship of cities as the top layer and regions and industries as the sub-layers. Since the regional and industrial data are overlapping, and the sum of the carbon emissions data of each region is equal to the carbon emissions of the city, and the sum of the carbon emissions of each industry is also equal to the carbon emissions of the city, a total-sub-data relationship is formed between cities and regions / industries, and a mutual verification relationship is formed between regions and industries. After filtering the data according to the hierarchical relationship of cities, regions, and industries, the data at these different levels are divided into three different clusters, and there is a certain coupling relationship between these clusters. Based on the data cluster division and coupling relationship, a three-cluster coupling training set is constructed, including the urban data set DT city , Regional Dataset DT area , Industry Dataset DT pro .

[0055] S3: constructing an optimal training set including carbon emissions and factors strongly correlated with carbon emissions based on the three-cluster coupling training set.

[0056] First, based on principal component analysis and correlation analysis, the weak features of carbon emission influencing factors are preliminarily screened, and then elastic net regression is used to screen and construct the optimal training set containing carbon emissions and their strongly correlated influencing factors to improve the accuracy and stability of the model.

[0057] See also Figure 3 As shown in the figure, elastic net adds L2 norm as a penalty term on the basis of LASSO. It has the characteristics of ridge regression and LASSO, and can better handle strong collinearity, group effect and high-dimensional nonlinear data with more variables than samples. The principle is as follows:

[0058] The elastic net data preprocessing process is the same as LASSO, and its coefficient expression is:

[0059]

[0060] make λ=λ1+λ2, the simplified expression is:

[0061]

[0062] The penalty term of the elastic net is a convex combination of the ridge regression penalty term and the LASSO penalty term. Therefore, the elastic net combines the advantages of coefficient shrinkage and variable selection. The singularities at the vertices and on the edges are strictly convex, and the strength of convexity varies with α. When α = 1, the elastic net is the LASSO, and the penalty function is convex but not strictly convex; when 0 < α < 1 (where α = 0.5 for EN), the penalty function is strictly convex; when α = 0, the elastic net is ridge regression.

[0063] To better introduce the dimensionality reduction process of the elastic net, define the relative constraint s as:

[0064]

[0065] where is the elastic net estimate of the variable coefficient, is the least squares estimate of the variable coefficient.

[0066] When s = 0, all variable coefficients are 0, and there are no variables in the model at this time; when 0 < s < 1 + λ2 (λ2 is the coefficient of the L2 norm), the predictor variables gradually enter the model. Among them, the important variables will enter the model earlier and have larger coefficients, while the coefficients of the irrelevant variables will be shrunk to 0. At this time, the optimal s value is obtained according to the model selection criterion (taking the AIC (Akaike Information Criterion) criterion as an example, the model with the minimum AIC value is the optimal model), and the important variables are selected based on the coefficients of each predictor variable to complete data dimensionality reduction; when s = 1 + λ2, all variables are selected into the model, and the elastic net becomes the least squares estimate. Finally, the factors retained by elastic net regression screening are used to construct an optimal training set including carbon emissions and their strongly correlated influencing factors.

[0067] S4: Construct three electricity-carbon calculation models based on deep BP neural networks from three dimensions of city, region, and industry, and use the optimal training set to train the electricity-carbon calculation models.

[0068] Construct three proxy models of "electricity-carbon calculation" based on deep BP neural networks from three dimensions of city, industry, and region, aiming to effectively capture the complex non-linear relationship between electricity consumption and carbon emissions. First, the research starts from the data preparation stage, collecting data related to electricity and carbon emissions, including electricity consumption Cus ele , gross domestic product of the city GDP city , gross domestic product of the region GDP area , demographic data Num p , Cp , η emp , and energy usage coal ,U gas ,U oil}∪{Cos coal ,Cos gas ,Cos oil}}, etc. The BP neural network structure is divided into three parts: input layer, hidden layer and output layer. Neurons are connected within the layer and fully connected outside the layer. During the training process, the signal is first forward propagated to the output layer to obtain preliminary weights and biases; secondly, after the error is calculated in the output layer, it is back-propagated back to the input layer to update the weights and biases of the neurons; finally, after repeated training, the error reaches the target range, the weights and biases in the model can be determined, and the training ends here.

[0069] Signal forward propagation process: Assume that the input signal is X = (x1, x2, ..., x n ) T , the input layer dimension is n; the hidden layer is a single layer with a dimension of p; the output layer dimension is m. The structure of the BP neural network is npm. At this time, the formula for the hidden layer nodes is as follows:

[0070]

[0071] ho j =f(hi j )

[0072] In the formula, hi j and ho j are the input and output of the jth node in the hidden layer; x i is the input of node i in the input layer; w ij is the connection weight between the input layer i node and the hidden layer j node; b j is the bias of the jth node in the hidden layer; f is the activation function of the hidden layer, usually the Sigmoid function, and the formula is:

[0073]

[0074] The input and output of the output layer node j are:

[0075]

[0076] yo k =f(yi k )

[0077] In the formula, yi k and yo k are the input and output of the output layer node k respectively; wjk is the connection weight between hidden layer node j and output layer node k; b k is the bias of node k in the output layer; f is the activation function of the output layer.

[0078] If the error does not meet the actual requirements, the back propagation algorithm will be used to update the weights and biases of the neurons, and the output layer error is:

[0079]

[0080] In the formula, yt k and k are the true value and actual value of the output layer node k respectively.

[0081] The update formula for the weights and bias of the output layer is:

[0082]

[0083] The update formula for the weights and bias of the hidden layer is:

[0084]

[0085] The errors of the output layer and the hidden layer are:

[0086] δ k =(yt k -y k )y k (1-y k )

[0087]

[0088] The adjustment formula for neuron weights and biases is:

[0089]

[0090] Where η is the learning rate, which is between 0 and 1.

[0091] S5: Use the Bayesian optimization algorithm to optimize the hyperparameters of the deep BP neural network, and adjust the hyperparameters by outputting the results based on the relationship between city, region, and industry data.

[0092] In the study of "electric carbon calculation model", the model parameter self-tuning method based on Bayesian optimization provides an effective means to improve model performance. Bayesian optimization is a global optimization method based on Bayesian statistics, which is particularly suitable for optimization problems with high dimensions, high computational costs and no clear formula. In the training process of BP (back propagation) neural network, select appropriate hyperparameters: learning rate ψ, batch size S z 、Number of hidden layers hid、Number of nodes vertex etc., are crucial to the convergence speed and final performance of the model. However, traditional grid search or random search methods are often inefficient and cannot effectively explore the parameter space. Bayesian optimization estimates the distribution of the objective function in the parameter space by constructing a proxy model, usually using a Gaussian process. In each iteration, the optimization algorithm updates the proxy model based on the previous experimental results and calculates an acquisition function Funacq(), such as expected improvement or probability improvement, to guide the next parameter selection. The invented method not only takes into account the predictive performance of the model, but also effectively balances exploration (trials in uncertain areas) and utilization (fine-tuning in areas known to perform well), thereby achieving more intelligent parameter optimization.

[0093] In the implementation process, we first define the hyperparameters to be optimized and their value ranges. Through preliminary experiments, we obtain the initial data point Init data , and then use the Bayesian optimization algorithm to iteratively adjust the parameters. In each iteration, the acquisition function is evaluated based on the current surrogate model, a new parameter combination is selected for training, and the surrogate model is updated. This process continues until the preset number of iterations is reached or the model performance no longer improves significantly. The application of Bayesian optimization in the "Electric Carbon Calculation Model" can effectively improve the training efficiency and prediction accuracy of the model. Compared with traditional methods, Bayesian optimization can find better hyperparameter configurations with fewer experiments, thereby saving computing resources and time. In addition, when faced with high-dimensional parameter space, this method can more flexibly deal with complex optimization problems, ensuring the robustness and adaptability of the model in practical applications.

[0094] The hyperparameter search space is represented by S, and the model parameters to be tuned are the learning rate ψ, the batch size S z 、Number of hidden layers hid 、Number of nodes vertex . Therefore, the objective function can be expressed as:

[0095]

[0096] S, used to determine the best model hyperparameter configuration, can be expressed as s * ∈S, such that:

[0097]

[0098] Here, the observed value of the objective function can be expressed as:

[0099]

[0100] This enables Bayesian optimization to build a probabilistic model that can be used to find the next position in S in the sample search, which can be derived from Bayesian theory. Because the present invention constructs a posterior distribution of the objective function, the subsequent hyperparameter configuration is selected from this distribution. The information of the initial sampling point is used to calculate the shape of the objective function and optimize the hyperparameters of the desired result.

[0101] For further information, see Figure 4 As shown in the figure, based on the self-tuning of the parameters of the electric carbon calculation model through Bayesian optimization, three "electric carbon calculation" sub-models are constructed in three dimensions: city, industry, and region. According to the total-sub-data relationship between cities and regions / industries, and the mutual verification relationship between regions and industries, the hyperparameters of the self-tuning deep BP neural network are fed back through the output results, that is, the sum of the carbon emissions of the industry is equal to the sum of the carbon emissions of the region, which is also equal to the sum of the carbon emissions of the city as a whole. Based on the mutual verification relationship, the "electric carbon calculation" model of the city as a whole, industry, and region can be further optimized.

[0102]

[0103] Judgment: If any of the above formulas is not true, it means that the parameter self-tuning of the three "electricity carbon calculation" sub-models has calculation deviations, and it is necessary to return to the iterative Bayesian optimization model parameter self-tuning method to readjust the parameters. co2 Emissions from the entire industry co2_pro The sum value is within the allowable error range, and the total carbon emissions of the city Em co2 Emissions from multiple regions co2_area If the sum value is within the allowable error range, it means that the Bayesian optimization model parameter self-tuning is completed and the iterative verification is terminated.

[0104] S6: Evaluate the accuracy of the deep BP neural network electro-carbon calculation model built based on the Bayesian optimization algorithm.

[0105] When evaluating the accuracy of the electric carbon calculation model based on the Bayesian optimized deep BP neural network, in addition to the mean absolute percentage error, other evaluation indicators such as mean square error and determination coefficient can also be combined to evaluate the model performance from multiple angles.

[0106] The present invention also provides a device for self-tuning parameters of an electric carbon calculation model, comprising:

[0107] Data collection module: used to collect historical carbon emissions and actual electricity consumption-related influencing factor data of cities, regions, and industries;

[0108] Training set construction module: used to screen the data according to the hierarchical relationship of city as the top layer and region and industry as the sub-layer, and construct a three-cluster coupling training set;

[0109] An optimal training set construction module: used to construct an optimal training set including carbon emissions and factors strongly related to carbon emissions based on the three-cluster coupling training set;

[0110] Electricity-carbon calculation model construction module: used to construct three electricity-carbon calculation models based on deep BP neural network from three dimensions of city, region and industry, and train the electricity-carbon calculation models using the optimal training set;

[0111] Hyperparameter optimization module: used to optimize the hyperparameters of deep BP neural networks using the Bayesian optimization algorithm, and adjust the hyperparameters by outputting the results based on the relationship between city, region, and industry data.

[0112] Model accuracy evaluation module: used to evaluate the accuracy of the deep BP neural network electric carbon calculation model built based on the Bayesian optimization algorithm.

[0113] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the electric carbon calculation model parameter self-tuning method.

[0114] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method for self-tuning parameters of an electric carbon calculation model.

[0115] An embodiment of the present invention also provides a computer program product corresponding to the method for self-tuning parameters of an electric carbon calculation model provided in the aforementioned embodiment. The computer program product includes a computer program, which is executed by a processor to implement the method for self-tuning parameters of an electric carbon calculation model provided in the aforementioned embodiment.

[0116] The above description and the accompanying drawings fully illustrate the embodiments of the present invention so that those skilled in the art can practice them. Other embodiments may include structural and other changes. The embodiments represent only possible variations. Unless explicitly required, individual components and functions are optional, and the order of operations may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. The embodiments of the present invention are not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for self-tuning parameters of an electric carbon calculation model, characterized in that: include: Collect data on historical carbon emissions and actual electricity consumption-related influencing factors for cities, regions, and industries; The data are screened according to the hierarchical relationship of cities as the top layer and regions and industries as sub-layers, and a three-cluster coupling training set is constructed; Constructing an optimal training set including carbon emissions and factors strongly correlated with carbon emissions based on the three-cluster coupling training set; Constructing three electric carbon calculation models based on deep BP neural network from three dimensions of city, region and industry, and training the electric carbon calculation models using the optimal training set; The Bayesian optimization algorithm is used to optimize the hyperparameters of the deep BP neural network, and the hyperparameters are adjusted by outputting the results based on the relationship between city, region, and industry data.

2. The method for self-tuning parameters of the electric carbon calculation model according to claim 1, characterized in that: The method comprises: Evaluate the accuracy of the deep BP neural network electro-carbon calculation model built based on the Bayesian optimization algorithm.

3. The method for self-tuning parameters of the electric carbon calculation model according to claim 1, characterized in that: The construction of the optimal training set including carbon emissions and factors strongly correlated with carbon emissions includes: firstly preliminarily screening out weak features in the carbon emission influencing factors based on principal component analysis and correlation analysis, and then further screening using elastic net regression to construct the optimal training set including carbon emissions and factors strongly correlated with carbon emissions.

4. The method for self-tuning parameters of the electric carbon calculation model according to claim 1, characterized in that: The BP neural network comprises an input layer, a hidden layer and an output layer. The hidden layer is a single layer, and neurons are connected within the layer without connection and fully connected outside the layer.

5. The method for self-tuning parameters of the electric carbon calculation model according to claim 1, characterized in that: The electric-carbon calculation model determines the final weights and biases of the electric-carbon calculation model through forward propagation and back propagation.

6. The method for self-tuning parameters of the electric carbon calculation model according to claim 1, characterized in that: The hyperparameters include multiple key parameters: learning rate, batch size, number of hidden layers, number of nodes, etc.

7. The method for self-tuning parameters of the electric carbon calculation model according to claim 1, characterized in that: The method of adjusting the hyperparameters by outputting the results based on the relationship between the city, region and industry data includes: forming a total-score data relationship between the city and the region or / industry, and forming a mutual verification relationship between the region and the industry, that is, the overall carbon emissions of the city are equal to the sum of the regional carbon emissions, and also equal to the sum of the industry carbon emissions.

8. The method for self-tuning parameters of the electric carbon calculation model according to claim 2, characterized in that: The evaluation is based on the accuracy of the deep BP neural network electric carbon calculation model constructed by the Bayesian optimization algorithm, including the use of mean absolute percentage error, mean square error, and determination coefficient to evaluate the model performance from multiple angles.

9. A self-tuning device for electric carbon calculation model parameters, characterized in that: include: Data collection module: used to collect historical carbon emissions and actual electricity consumption-related influencing factor data of cities, regions, and industries; Training set construction module: used to screen the data according to the hierarchical relationship of city as the top layer and region and industry as the sub-layer, and construct a three-cluster coupling training set; An optimal training set construction module: used to construct an optimal training set including carbon emissions and factors strongly correlated with carbon emissions based on the three-cluster coupling training set; Electricity-carbon calculation model construction module: used to construct three electricity-carbon calculation models based on deep BP neural network from three dimensions of city, region and industry, and train the electricity-carbon calculation models using the optimal training set; Hyperparameter optimization module: used to optimize the hyperparameters of deep BP neural networks using the Bayesian optimization algorithm, and adjust the hyperparameters by outputting the results based on the relationship between city, region, and industry data.

10. The electric carbon calculation model parameter self-tuning device according to claim 9, characterized in that: The device comprises: Model accuracy evaluation module: used to evaluate the accuracy of the deep BP neural network electric carbon calculation model built based on the Bayesian optimization algorithm.

11. The electric carbon calculation model parameter self-tuning device according to claim 9, characterized in that: The construction of the optimal training set including carbon emissions and factors strongly correlated with carbon emissions includes: firstly preliminarily screening out weak features in the carbon emission influencing factors based on principal component analysis and correlation analysis, and then further screening using elastic net regression to construct the optimal training set including carbon emissions and factors strongly correlated with carbon emissions.

12. The electric carbon calculation model parameter self-tuning device according to claim 9, characterized in that: The BP neural network comprises an input layer, a hidden layer and an output layer. The hidden layer is a single layer, and neurons are connected within the layer without connection and fully connected outside the layer.

13. The electric carbon calculation model parameter self-tuning device according to claim 9, characterized in that: The electric-carbon calculation model determines the final weights and biases of the electric-carbon calculation model through forward propagation and back propagation.

14. The electric carbon calculation model parameter self-tuning device according to claim 9, characterized in that: The hyperparameters include multiple key parameters: learning rate, batch size, number of hidden layers, number of nodes, etc.

15. The electric carbon calculation model parameter self-tuning device according to claim 9, characterized in that: The method of adjusting the hyperparameters by outputting the results based on the relationship between the city, region and industry data includes: forming a total-score data relationship between the city and the region or / industry, and forming a mutual verification relationship between the region and the industry, that is, the overall carbon emissions of the city are equal to the sum of the regional carbon emissions, and also equal to the sum of the industry carbon emissions.

16. The electric carbon calculation model parameter self-tuning device according to claim 10, characterized in that: The evaluation is based on the accuracy of the deep BP neural network electric carbon calculation model constructed by the Bayesian optimization algorithm, including the use of mean absolute percentage error, mean square error, and determination coefficient to evaluate the model performance from multiple angles.

17. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the computer program is executed by a processor, the processor executes the steps of the method for self-tuning parameters of the electric carbon calculation model as described in any one of claims 1 to 8.

18. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor executes the steps of the method for self-tuning parameters of the electric carbon calculation model as described in any one of claims 1 to 8.

19. A computer program product comprising a computer program, characterized in that The computer program is executed by a processor to implement the method for self-tuning parameters of an electric carbon calculation model according to any one of claims 1 to 8.

Images